If you want to read the PDF, try requesting it from the authors.

Abstract

Mapping and monitoring land cover and land cover change remains a top priority for land managers. Uniquely, remote sensing offers the capacity to acquire information in a systematic (spatially, temporally, and categorically) and synoptic fashion that is appealing from monitoring and reporting perspectives. The opening of the Landsat archive and new processing and analysis opportunities enable the characterization of large areas and the generation of dynamic, transparent, systematic, repeatable, and spatially exhaustive information products. Best Available Pixel (BAP) approaches enable the production of periodic image composites free of haze, clouds, or shadows over large areas. In this paper we demonstrate an integrated protocol to produce spatially exhaustive annual BAP image composites that are seasonally constrained and free of atmospheric perturbations. These annual BAP composites for the years inclusive of 1998 to 2010 provide for generation of a suite of change metrics for the period 2000 to 2010 using Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) data. The study area is the > 375,000 km2 forested area of Saskatchewan, Canada. We evaluate the robustness of the protocol by comparing in-filled (or proxy) values with a true reference set for surface reflectance. An initial change detection pass is used to aid in the allocation of proxy values (for missing and anomalous values) and to allocate change events to the correct year, with a second pass to characterize key change points and related time series trends (e.g., change year, post-change slopes, among others). In so doing, a multi-temporal data cube (via a series of annual proxy composites) and a set of change metrics are generated. Approximately 35% of the pixels in our study area required proxy values, either as a result of missing data or our noise detection approach. Overall, our results indicate strong agreement between the assigned proxy values and the reference data (R = 0.71– 0.91, RMSE = 0.008–0.025). Agreement was stronger for pixel series with no change events (R = 0.73–0.92, RMSE = 0.007–0.024), relative to pixel series with change events (R = 0.63–0.87, RMSE = 0.010–0.029). The generated change metrics, derived via temporal and spatial analysis of the annual BAP composites, were an important precursor to the generation of valid proxy values, and – importantly – provide valuable information for the further assessment and understanding of land cover and land cover change. Our results indicate that the demonstrated protocol provides a reliable approach to generate proxy image composites containing no data gaps, along with a suite of informative change metrics that provide a comprehensive characterization of land cover changes (including disturbance and recovery) allowing for an improved understanding of landscape dynamics. The protocol is efficient and may be applied over large areas to support regional and national mapping and monitoring activities.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... Spectral indices of vegetation usually follow seasonal patterns that repeat on an annual cycle. The trajectory of a spectral index (or any other annual time series-based predictor) over time can thus be examined from annual time series [34,35]. With the opening of the Landsat archive, algorithms have been developed to detect changes in temporal trajectories of spectral indices and relating them to forest disturbances and structural changes [35][36][37]. ...
... The trajectory of a spectral index (or any other annual time series-based predictor) over time can thus be examined from annual time series [34,35]. With the opening of the Landsat archive, algorithms have been developed to detect changes in temporal trajectories of spectral indices and relating them to forest disturbances and structural changes [35][36][37]. Such approaches implement complex algorithms and perform computationally demanding tasks. ...
... Specifically, best-available pixel (BAP) image composites were first produced from Landsat imagery by selecting observations for each pixel within a specific date range (August 1+/-30 days) based on the scoring functions defined by [49], which rank the presence and distance to clouds and their shadows, the atmospheric quality, and the acquisition sensor. Next, these image composites were further refined by removing noisy observations (e.g., haze and smoke) and infilling data gaps using spectral trend analysis of pixel time series [35]. This process is not part of FOSTER hence we expect annual time series of optical imagery to be available. ...
Article
Full-text available
The uptake of technologies such as airborne laser scanning (ALS) and more recently digital aerial photogrammetry (DAP) enable the characterization of 3-dimensional (3D) forest structure. These forest structural attributes are widely applied in the development of modern enhanced forest inventories. As an alternative to extensive ALS or DAP based forest inventories, regional forest attribute maps can be built from relationships between ALS or DAP and wall-to-wall satellite data products. To date, a number of different approaches exist, with varying code implementations using different programming environments and tailored to specific needs. With the motivation for open, simple and modern software, we present FOSTER (Forest Structure Extrapolation in R), a versatile and computationally efficient framework for modeling and imputation of 3D forest attributes. FOSTER derives spectral trends in remote sensing time series, implements a structurally guided sampling approach to sample these often spatially auto correlated datasets, to then allow a modelling approach (currently k-NN imputation) to extrapolate these 3D forest structure measures. The k-NN imputation approach that FOSTER implements has a number of benefits over conventional regression based approaches including lower bias and reduced over fitting. This paper provides an overview of the general framework followed by a demonstration of the performance and outputs of FOSTER . Two ALS-derived variables, the 95 th percentile of first returns height ( elev_p95 ) and canopy cover above mean height ( cover ), were imputed over a research forest in British Columbia, Canada with relative RMSE of 18.5% and 11.4% and relative bias of -0.6% and 1.4% respectively. The processing sequence developed within FOSTER represents an innovative and versatile framework that should be useful to researchers and managers alike looking to make forest management decisions over entire forest estates.
... However, there are many published works about change detection methodologies. In [11] space-time features were exploited to detect forest disturbances using two consecutive observations and in [12] annual composites were used for detecting changes. In [13]- [15] methodologies to identify changes on the land and their type are proposed. ...
... Step 3 builds up the basis of the proposed methodology. Change detection applications try to find significant changes in data to understand if it was a real change and not a pseudo change (phenology, illumination, external effects) [12], and this is not an exception. FT detection relies on the comparison of consecutive values and FL growth estimation on the inter-annual comparison of values, being fully dependent of the generated time series from this step. ...
... By combining the two identified features (self-learning and robustness to phenolo- gy) we modeled the FL growth using equation (1) and (2), where i is the month being estimated, j the month used for the difference (for the first year it ranges from 1 to 11, after that it is equal to 12), and k the number of months after the FT. The first term corresponds monthly FL growth and the second is the scaling term. ...
Chapter
Mediterranean Europe is strongly affected by wildfires. In Portugal, the Portuguese Institute for Nature Conservation and Forests (ICNF) implemented the national fuel break (FB) network responsible for fire control and suppression. FBs are regions where vegetation is reduced to break up the fuel continuity and create pathways for the firefighting vehicles. The efficiency of this strategy relies on the correct implementation of FBs and on periodic fuel treatments. Multispectral imagery from Sentinel-2 (with high temporal and spatial resolution) facilitates the monitoring of FBs and the implementation of methodologies for their management. In this paper a two stages methodology is proposed for monitoring FBs. The first stage consists in detecting fuel treatments in FBs, to understand if those were correctly executed. This is done through a change detection methodology with resource to an Artificial Neural Network. The second stage monitors the vegetation recovery after a fuel treatment, to aid the scheduling of new treatments, ensuring the efficiency of FBs during the fire season. Both methodologies resort to reflectance bands and spectral indices from Sentinel-2; and timeseries and objects, exploiting the temporal and spatial information. The two stages were tested in different regions across the Portuguese territory, demonstrating their usability for all the national fuel break network. The detection of treatments achieved a relative error lower than 4%, and the vegetation recovery cycle estimated by the second stage match the expectations from ICNF.
... Several other methods exist in literature that combine SITS fitting methods with CD ones in order to produce yearly change maps [8][9][10] . Such methods suffer from the same problem of rather focusing on a single application or assuming cyclic information. ...
... The model coefficients were estimated by the Ordinary Least Squares fitting method and based on only clear sky Landsat images (which reduces the possible amount of information) and only bands 2 and 5 of the sensor, reducing the spectral information and thus the change detection capabilities. Hermosilla et al. 9 , made use of an annual Based Available Pixel (BAP) image composites and break point strategy to detect disturbances over forest based on a Normalized Burn Ratio (NBR) feature. BAPs work at pixel level and thus consider all possible cloud free pixels in the SITS, without penalizing the whole image. ...
... One well-established class of algorithms builds and analyzes annual trends for each pixel series in order to characterize disturbance and recovery over long time periods (Hermosilla et al., 2015a;Kennedy et al., 2010). These approaches rely on systematic observations, with consistency in radiometry and observation period being essential for establishing trends that allow both abrupt and gradual change to be detected (Griffiths et al., 2013;White et al., 2014). ...
... To assess well over these long time frames, such routines typically interpret a signal that is as clean as possible, so that it is intercomparable between years and that both true abrupt changes and slow growth are captured. For example, the Composite-to-Change (C2C) protocol determines the best available pixel annually from Landsat observations to create a representative annual composite for monitoring forest changes, then analyzes trends in those annual composites to interpret change (Hermosilla et al., 2015a(Hermosilla et al., , 2015bHermosilla et al., 2016;White et al., 2017;White et al., 2014). These annual composites are well-calibrated and interpretable for understanding important ongoing inter-annual change characteristics like change rate, magnitude, persistence, direction of change (Gillanders et al., 2008), and forest recovery following disturbance . ...
Article
Full-text available
Knowledge of forest change type and timing is required for forest management, reporting, and science. Time series of historic satellite data (e.g. Landsat) have resulted in an invaluable record of changes in forest conditions. Natural resource management and reporting typically operate at an annual time step, yet the recent addition of data streams from compatible satellites (e.g., Sentinel-2) offer the possibility of generating frequent, management-relevant forest status assessments and maps of change. Analytical approaches that rely on a time series of observations to identify change often struggle to provide reliable estimates of change events in terminal years of the time series until subsequent, additional observations are available. Methods to meaningfully integrate observations from compatible satellite platforms can provide short-term information to augment and refine estimates of change area and type in those terminal years of the time series. In this research we fuse Landsat-8 and Sentinel-2A and -2B data streams to capture, with reduced latency, stand replacing forest change (harvest and wildfire), tagged to a temporal window of occurrence over an ~10,000 km² area of central British Columbia, Canada. We introduce a new algorithm, SLIMS (Shrinking Latency in Multiple Streams), to rapidly and reliably detect change, and then use an established Bayesian approach to meaningfully combine changes detected in the Landsat and Sentinel data streams. Our results indicate that the type and timing of stand-replacing disturbances can be identified in these forests with high accuracy. Overall, 13.9% of the study area was disturbed between the end of 2016 and the end of 2017, with the majority of disturbed area attributable to wildfire and a smaller amount attributed to forest harvesting, mostly in the winter 2016–2017 with some limited summer harvest also occurring. Overall accuracy of the change, assessed using independent validation data, was 95% ± 2.3%. The capacity of these change results to augment a trend-based assessment of change for 2017 was also demonstrated and provides a framework for how short- and long-term change detection approaches provide complementary information that can increase the timeliness and accuracy of change area estimates in the terminal years of a time series. These findings also demonstrate the capacity to regard Landsat and Sentinel-2 sensors as elements of a virtual constellation to obtain forest change information in a timely (i.e., end of growing season) and reliable fashion over large areas.
... Changes in the spectral trajectory of individual pixels to date have most often been undertaken at annual time steps, using annual imagery composites (Hermosilla et al., 2015;Cohen et al., 2018;Griffiths et al., 2019;Corbane et al., 2020). These composites allow for a cloud-free image to be created and ensure that the influence of phenology is minimized using only images acquired within a predefined summer date range. ...
... A number of approaches have been developed to automate the detection of changes in forested landscapes (Kennedy et al., 2010;Hermosilla et al., 2015;White et al., 2017b;Cohen et al., 2018) using monthly, seasonal or annual change detection approaches. One well-established change detection technique designed for weekly or biweekly remote sensing observations is breaks For additive seasonal and trend (Verbesselt et al., 2010), which integrates the iterative decomposition of time series of remote sensing values into trend, seasonal and noise components. ...
Article
Forestry inventory update is a critical component of sustainable forest management, requiring both the spatially explicit identification of forest cover change and integration of sampled or modelled components like growth and regeneration. Contemporary inventory data demands are shifting, with an increased focus on accurate attribute estimation via the integration of advanced remote sensing data such as airborne laser scanning (ALS). Key challenges remain, however, on how to maintain and update these next-generation inventories as they age. Of particular interest is the identification of remotely sensed data that can be applied cost effectively, as well as establishing frameworks to integrate these data to update information on forest condition, predict future growth and yield, and integrate information that can guide forest management or silvicultural decisions such as thinning and harvesting prescriptions. The purpose of this article is to develop a conceptual framework for forestry inventory update, which is also known as the establishment of a ‘living inventory’. The proposed framework contains the critical components of an inventory update including inventory and growth monitoring, change detection and error propagation. In the framework, we build on existing applications of ALS-derived enhanced inventories and integrate them with data from satellite constellations of free and open, analysis-ready moderate spatial resolution imagery. Based on a review of the current literature, our approach fits trajectories to chronosequences of pixel-level spectral index values to detect change. When stand-replacing change is detected, corresponding values of cell-level inventory attributes are reset and re-established based on an assigned growth curve. In the case of non–stand-replacing disturbances, cell estimates are modified based on predictive models developed between the degree of observed spectral change and relative changes in the inventory attributes. We propose that additional fine-scale data can be collected over the disturbed area, from sources such as CubeSats or remotely piloted airborne systems, and attributes updated based on these data sources. Cells not identified as undergoing change are assumed unchanged with cell-level growth curves used to increment inventory attributes. We conclude by discussing the impact of error propagation on the prediction of forest inventory attributes through the proposed near real-time framework, computing needs and integration of other available remote sensing data.
... We used the Landsat 5, 7, and 8 Surface Reflectance Tier 1 products that were harmonized for corresponding bands . We selected growing season months (May-October), masked pixels covered by snow and clouds, and then generated annual composite images using the best pixel approach (Hermosilla et al 2015). Composite images allowed us to mitigate missing data from snow-cover, cloud-cover, and the Landsat 7 scan line error. ...
... Satellite imagery is paramount to developing a fire perimeter database and gaining an accurate understanding of fire regimes characteristics, including annual area burned, (Soja 2004) across eastern Siberian taiga and tundra zones. Developing a fire perimeter database is a computationally intensive task made easier by Earth Engine, which allowed us to map fire across eastern Siberia at a spatial resolution comparable to North American fire databases by generating Landsat composite images that mitigated missing data caused by weather and the scanline error from Landsat 7 (Hermosilla et al 2015). Fire perimeters developed using our approach matched closely with fire perimeters from Alaska 2004 in the MTBS database, except for large fires in close proximity that merged into a single fire perimeter. ...
Article
Full-text available
Circum-boreal and -tundra systems are crucial carbon pools that are experiencing amplified warming and are at risk of increasing wildfire activity. Changes in wildfire activity have broad implications for vegetation dynamics, underlying permafrost soils, and ultimately, carbon cycling. However, understanding wildfire effects on biophysical processes across eastern Siberian taiga and tundra remains challenging because of the lack of an easily accessible annual fire perimeter database and underestimation of area burned by MODIS satellite imagery. To better understand wildfire dynamics over the last 20 years in this region, we mapped area burned, generated a fire perimeter database, and characterized fire regimes across eight ecozones spanning 7.8 million km ² of eastern Siberian taiga and tundra from ∼61–72.5° N and 100° E–176° W using long-term satellite data from Landsat, processed via Google Earth Engine. We generated composite images for the annual growing season (May–September), which allowed mitigation of missing data from snow-cover, cloud-cover, and the Landsat 7 scan line error. We used annual composites to calculate the difference Normalized Burn Ratio (dNBR) for each year. The annual dNBR images were converted to binary burned or unburned imagery that was used to vectorize fire perimeters. We mapped 22 091 fires burning 152 million hectares (Mha) over 20 years. Although 2003 was the largest fire year on record, 2020 was an exceptional fire year for four of the northeastern ecozones resulting in substantial increases in fire activity above the Arctic Circle. Increases in fire extent, severity, and frequency with continued climate warming will impact vegetation and permafrost dynamics with increased likelihood of irreversible permafrost thaw that leads to increased carbon release and/or conversion of forest to shrublands.
... Most of them use large-area surface reflectance products that are generated by year and gap-free. These methods demonstrate a good efficiency in describing both the vegetated landscape spatial discretization and temporal trends (Gómez et al., 2016;Hermosilla et al., 2015Olthof and Fraser, 2014;Zhu and Woodcock, 2014;Roy et al., 2015bib_Hermosilla_et_al_2015). ...
Article
Broad-scale land cover classifications created by using satellite imagery are recognized as important and necessary input to land cover mapping. Evaluation of land cover changes is essential for sustainable watershed management. In this research, a framework for quantitative assessment of long-term land cover changes over large areas is developed. A key feature of the framework is a spatiotemporal fusion of classifications. Herein, a method to resolve the inconsistency of land cover classes at overlapped multi-segment and multi-temporal classifications by considering their probability propagation is proposed. The framework was applied in assessing long-term land cover changes in the Dniester river basin. Two land cover maps, for 2002 and 2018, were obtained with overall accuracy of 81.42% and 81.30%, respectively. The subsequent analysis and quantitative assessment of sixteen-year land cover changes by T. Saaty’s analytic hierarchy procedure resulted in a map of the significance of these long-term land cover changes for the Dniester river basin. The major part of the basin (81.6%) exhibits no changes. Land cover changes maintaining favorable conditions for watershed appear over 10.5% of the area, while harmful changes occur over 7.9% of the territory.
... As a result, the application of atmospheric correction makes change detection results more reliable. Therefore, algorithms with dense time series and especially those over large areas [59,60] typically require the use of surface reflectance data, which has become the lowest level of analysis-ready data standard (e.g., as delivered by Landsat Collection 1 or 2). ...
Article
Full-text available
With uninterrupted space-based data collection since 1972, Landsat plays a key role in systematic monitoring of the Earth’s surface, enabled by an extensive and free, radiometrically consistent, global archive of imagery. Governments and international organizations rely on Landsat time series for monitoring and deriving a systematic understanding of the dynamics of the Earth’s surface at a spatial scale relevant to management, scientific inquiry, and policy development. In this study, we identify trends in Landsat-informed change detection studies by surveying 50 years of published applications, processing, and change detection methods. Specifically, a representative database was created resulting in 490 relevant journal articles derived from the Web of Science and Scopus. From these articles, we provide a review of recent developments, opportunities, and trends in Landsat change detection studies. The impact of the Landsat free and open data policy in 2008 is evident in the literature as a turning point in the number and nature of change detection studies. Based upon the search terms used and articles included, average number of Landsat images used in studies increased from 10 images before 2008 to 100,000 images in 2020. The 2008 opening of the Landsat archive resulted in a marked increase in the number of images used per study, typically providing the basis for the other trends in evidence. These key trends include an increase in automated processing, use of analysis-ready data (especially those with atmospheric correction), and use of cloud computing platforms, all over increasing large areas. The nature of change methods has evolved from representative bi-temporal pairs to time series of images capturing dynamics and trends, capable of revealing both gradual and abrupt changes. The result also revealed a greater use of nonparametric classifiers for Landsat change detection analysis. Landsat-9, to be launched in September 2021, in combination with the continued operation of Landsat-8 and integration with Sentinel-2, enhances opportunities for improved monitoring of change over increasingly larger areas with greater intra- and interannual frequency.
... El programa de monitoreo Landsat se inició aproximadamente hace 40 años (Zhao et al., 2018); por lo tanto, posee el registro temporal más largo y continuo de imágenes que muestran la cobertura del planeta. Además, la resolución espacial y especialmente la resolución espectral de las imágenes de estos satélites, son muy apropiadas para el seguimiento de las actividades humanas y su impacto al ambiente (Hermosilla et al., 2015). ...
Article
Full-text available
p>Mapping of land use and forest cover and assessing their changes is essential in the design of strategies to manage and preserve the natural resources of a country, and remote sensing have been extensively used with this purpose. By comparing four classification algorithms and two types of satellite images, the objective of the research was to identify the type of algorithm and satellite image that allows higher global accuracy in the identification of forest cover in highly fragmented landscapes. The study included a treatment arrangement with three factors and six randomly selected blocks within the Huetar Norte Zone in Costa Rica. More accurate results were obtained for classifications based on Sentinel-2 images compared to Landsat-8 images. The best classification algorithms were Maximum Likelihood, Support Vector Machine or Neural Networks, and they yield better results than Minimum Distance Classification. There was no interaction among image type and classification methods, therefore, Sentinel-2 images can be used with any of the three best algorithms, but the best result was the combination of Sentinel-2 and Support Vector Machine. An additional factor included in the study was the image acquisition date. There were significant differences among months during which the image was acquired and an interaction between the classification algorithm and this factor was detected. The best results correspond to images obtained in April, and the lower to September, month that corresponds with the period of higher rainfall in the region studied. The higher global accuracy identifying forest cover is obtained with Sentinel-2 images from the dry season in combination with Maximum Likelihood, Support Vector Machine, and Neural Network image classification methods.</p
... The objective of this research is therefore to determine the spectral correspondence between USGS and HLS sourced data. Spectral correspondence would enable combined use of both data sources in continued production of land cover (Hermosilla et al., 2018), forest change (Hermosilla et al., 2015a(Hermosilla et al., , 2015b, and forest structure products (Matasci et al, 2018a(Matasci et al, , 2018b based upon annual image composites (e.g. Hermosilla et al., 2016) and enable refinement of science and applications approaches in an on-going fashion. ...
Article
Full-text available
An increase in the temporal revisit of satellite data is often sought to increase the likelihood of obtaining cloud- and shadow-free observations as well as to improve mapping of rapidly- or seasonally-changing features. Currently, as a tandem, Landsat-7 Enhanced Thematic Mapper Plus (ETM+) and −8 Operational Land Imager (OLI) provide an acquisition opportunity on an 8-day revisit interval. Sentinel-2A and -2B MultiSpectral Instrument (MSI), with a wider swath, have a 5-day revisit interval at the equator. Due to robust pre- and post-launch cross-calibration, it has been possible for NASA to produce the Harmonized Landsat Sentinel-2 (HLS) data product from Landsat-8 OLI and Sentinel-2 MSI: L30 and S30, respectively. Knowledge of the agreement of HLS outputs (especially S30) with historic Landsat surface reflectance products will inform the ability to integrate historic time-series information with new and more frequent measures as delivered by HLS. In this research, we control for acquisition date and data source to cross-compare the HLS data (L30, S30) with established Landsat-8 OLI surface-reflectance measures as delivered by the USGS (hereafter BAP, Best Available Pixel). S30 and L30 were found to have high agreement (R = 0.87–0.96) for spectral channels and an r = 0.99 for Normalized Burn Ratio (NBR) with low relative root-mean-square difference values (1.7%–3.3%). Agreement between L30 and BAP was lower, with R values ranging from 0.85 to 0.92 for spectral channels and R = 0.94 for NBR. S30 and BAP had the lowest agreement, with R values ranging from 0.71 to 0.85 for spectral channels and r = 0.90 for NBR. Comparisons indicated a stronger agreement at latitudes above 55° N. Some dependency between spectral agreement and land cover was found, with stronger correspondence for non-vegetated cover types. The level of agreement between S30 and BAP reported herein would enable integration of HLS outputs with historic Landsat data. The resulting increased temporal frequency of data allows for improvements to current cloud screening practices and increases data density and the likelihood of temporal proximity to target date for pixel compositing approaches. Furthermore, additional within-year observations will enable change products with a higher temporal fidelity and allow for the incorporation of phenological trends into land cover classification algorithms.
... Most of them use largearea surface reflectance products that are generated by year and gap-free. These methods demonstrate a good efficiency in describing both the vegetated landscape spatial discretization and temporal trends (Gómez et al. 2016;Hermosilla et al. 2018Hermosilla et al. , 2015Olthof and Fraser 2014;Zhu and Woodcock 2014;Roy et al. 2015). A number of techniques and algorithms for land cover classification have been developed and available in different satellite imagery processing softwares, utilizing multi-spectral and hyperspectral passive remote sensing data. ...
Chapter
Land provides a range of biophysical and socioeconomic goods and services that support the sustainability of ecosystem services, livelihoods, and human well-being. Maintaining ecosystem functions and services, while also supporting human well-being, are the primary goals of sustainable land management. Wide-scale multi-temporal land cover classifications created on the basis of satellite remote sensing data are recognized as necessary and important input to land degradation analysis. Evaluation of long-term land cover changes is of crucial importance for sustainable land management. In this research, we develop a Geo-intelligence framework providing quantitative assessment and mapping of such changes over large areas. To eliminate the inconsistency of land cover classes in overlapping classifications, a spatio-temporal merger of land cover classifications was carried out taking into account their probabilistic distribution. The proposed approach is applied in assessing long-term changes of the land cover in the Southern Ukraine, the region suffering the most from land degradation. Two land cover maps, for two distant years, were obtained with overall relevant accuracy. As a result of quantitative assessment of long-term changes of land cover by slightly modified T. Saaty’s analytic hierarchy procedure, a map of spatial distribution of land cover change of importance to the study area during the period of the study was developed.
... Most of them use largearea surface reflectance products that are generated by year and gap-free. These methods demonstrate a good efficiency in describing both the vegetated landscape spatial discretization and temporal trends (Gómez et al. 2016;Hermosilla et al. 2018Hermosilla et al. , 2015Olthof and Fraser 2014;Zhu and Woodcock 2014;Roy et al. 2015). A number of techniques and algorithms for land cover classification have been developed and available in different satellite imagery processing softwares, utilizing multi-spectral and hyperspectral passive remote sensing data. ...
Chapter
The land use and land cover is changing in different parts of the world, the root cause of which is the increasing urbanization rate. The peri-urban areas are transforming due to this pressure, leading to urban expansion and resulting in major changes in land use along the highway. Such peri-urban areas are largely neglected in policy and practice because they are mostly included in the rural category and come as a region beyond the urban administration. Hence, the present study is focused on the analysis of the impact of land use land cover change on the urbanizing region along a highway using geospatial technology. Seven typical nodes were carefully chosen along the corridor and a growth node was identified. An undeveloped node having an area of 5 km² was chosen as a case study for further analysis and proposal. The entire study area was zoned into three categories (emerging zone, agricultural and tourism zone, and residential zone) based on the prominent land uses for the suggestion of the proposals. The present study can be beneficial to planners, administrators, and policymakers as a stepping stone in promoting the sustainability of peri-urban areas.
... The incorporation of the forest dynamics is proven to significantly improve the accuracy of the model estimations. Most of the research based on LTS imagery composite suggests choosing the best available pixel from the various annual images in each year [27,28]. This process creates an annual composite of LTS over the time period. ...
Article
Full-text available
The forest volumes are essential as they are directly related to the economic and environmental values of the forests. Satellite-based forest volume estimation was first developed in the 1990s, and the accuracy of the estimation has been improved over time. One of the satellite-based forest volume estimation issues is that it tends to overestimate the large volume class and underestimate the small volume class. Free availability of the major satellite imagery and the development of cloud-based computational platforms facilitate an immense amount of satellite imagery in the estimation. In this paper, we set three objectives: (1) to examine whether the long Landsat time series contributes to the improvement of the estimation accuracy, (2) to explore the effectiveness of forest disturbance record and land cover data as ancillary spatial data on the accuracy of the estimation, and (3) to apply the bias correction method to reduce the bias of the estimation. We computed three Tasseled-cap components from the Landsat data for preparation of short (2014–2016) and long (1984–2016) time series. Each data entity was analyzed with harmonic regressions resulting in the coefficients and the fitted values recorded as pixel values in a multilayer raster database. Data included Forest Inventory and Analysis (FIA) unit field inventory measurements provided by the United States Department of Agriculture Forest Service and the National Land Cover Database and disturbance history data added as ancillary information. The totality of the available data was organized into seven distinct Random Forest (RF) models with different variables compared against each other to identify the ones with the most satisfactory performance. A bias correction method was then applied to all the RF models to examine the effectiveness of the method. Among the seven models, the worst one used the coefficients and fitted values of the short Landsat time series only, and the best one used coefficients and fitted values of both short and long Landsat time series. Using the Out-of-bag (OOB) score, the best model was found to be 34.4% better than the worst one. The model that used only the long time series data had almost the same OOB score as the best model. The results indicate that the use of the long Landsat time series improves model performance. Contrary to the previous research employing forest disturbance data as a feature variable had almost no effect on OOB. The bias correction method reduced the relative size of the bias in the estimates of the best model from 3.79% to −1.47%, the bottom 10% bias by 12.5 points, and the top 10% bias by 9.9 points. Depending on the types of forest, important feature variables were differed, reflecting the relationship between the time series remote sensing data we computed for this research and the forests’ phenological characteristics. The availability of Light Detection And Ranging (LiDAR) data and accessibility of the precise locations of the FIA data are likely to improve the model estimates further.
... The area of exposed soils on a single remote sensing scene is limited, and often, the periods in which exposed soils dominate are restricted to short time windows [28] when the soil is in seedbed condition. Compositing techniques of multi-temporal satellite image archives provide an alternative and are widely used in the literature [25][26][27][28][29][30][31][32][33][34]. The compositing approach allows combining all bare soils of all input scenes, which enables a joint estimation of soil parameters for all exposed soils in the observed time period. ...
Article
Full-text available
For food security issues or global climate change, there is a growing need for large-scale knowledge of soil organic carbon (SOC) contents in agricultural soils. To capture and quantify SOC contents at a field scale, Earth Observation (EO) can be a valuable data source for area-wide mapping. The extraction of exposed soils from EO data is challenging due to temporal or permanent vegetation cover, the influence of soil moisture or the condition of the soil surface. Compositing techniques of multitemporal satellite images provide an alternative to retrieve exposed soils and to produce a data source. The repeatable soil composites, containing averaged exposed soil areas over several years, are relatively independent from seasonal soil moisture and surface conditions and provide a new EO-based data source that can be used to estimate SOC contents over large geographical areas with a high spatial resolution. Here, we applied the Soil Composite Mapping Processor (SCMaP) to the Landsat archive between 1984 and 2014 of images covering Bavaria, Germany. Compared to existing SOC modeling approaches based on single scenes, the 30-year SCMaP soil reflectance composite (SRC) with a spatial resolution of 30 m is used. The SRC spectral information is correlated with point soil data using different machine learning algorithms to estimate the SOC contents in cropland topsoils of Bavaria. We developed a pre-processing technique to address the issue of combining point information with EO pixels for the purpose of modeling. We applied different modeling methods often used in EO soil studies to choose the best SOC prediction model. Based on the model accuracies and performances, the Random Forest (RF) showed the best capabilities to predict the SOC contents in Bavaria (R² = 0.67, RMSE = 1.24%, RPD = 1.77, CCC = 0.78). We further validated the model results with an independent dataset. The comparison between the measured and predicted SOC contents showed a mean difference of 0.11% SOC using the best RF model. The SCMaP SRC is a promising approach to predict the spatial SOC distribution over large geographical extents with a high spatial resolution (30 m).
... Indeed, as the phenological status of different crop types evolves differently with time, having information over time allows to better discriminate the crops. Therefore, the use of optical data over large and cloudy areas involves working with composite images (including sometimes spatiotemporal context for residual missing pixels in composite images [18,19]), linearly temporally gap-filled images [20], or pixel-wise weighted least-squares smoothing of the values over time [21,22]. ...
Article
Full-text available
In this upcoming Common Agricultural Policy (CAP) reform, the use of satellite imagery is taking an increasing role for improving the Integrated Administration and Control System (IACS). Considering the operational aspect of the CAP monitoring process, the use of Sentinel-1 SAR (Synthetic Aperture Radar) images is highly relevant, especially in regions with a frequent cloud cover, such as Belgium. Indeed, SAR imagery does not depend on sunlight and is barely affected by the presence of clouds. Moreover, the SAR signal is particularly sensitive to the geometry and the water content of the target. Crop identification is often a pre-requisite to monitor agriculture at parcel level (ploughing, harvest, grassland mowing, intercropping, etc.) The main goal of this study is to assess the performances and constraints of a SAR-based crop classification in an operational large-scale application. The Random Forest object-oriented classification model is built on Sentinel-1 time series from January to August 2020 only. It can identify crops in the Walloon Region (south part of Belgium) with high performance: 93.4% of well-classified area, representing 88.4% of the parcels. Among the 48 crop groups, the six most represented ones get a F1-score higher or equal to 84%. Additionally, this research documents how the classification performance is affected by different parameters: the SAR orbit, the size of the training dataset, the use of different internal buffers on parcel polygons before signal extraction, the set of explanatory variables, and the period of the time series. In an operational context, this allows to choose the right balance between classification accuracy and model complexity. A key result is that using a training dataset containing only 3.2% of the total number of parcels allows to correctly classify 91.7% of the agricultural area. The impact of rain and snow is also discussed. Finally, this research analyses how the classification accuracy depends on some characteristics of the parcels like their shape or size. This allows to assess the relevance of the classification depending on those characteristics, as well as to identify a subset of parcels for which the global accuracy is higher.
... Research using remote sensing to improve estimates of age [114,118,119] and estimate species composition [120,121] is ongoing and is demonstrating the potential to generate wallto-wall raster layers of both attributes. For younger stands, stand age can be derived from time series of satellite imagery when stand records are not available [22,[122][123][124]. The required degree of accuracy of these estimates however depends on how these outputs are used in operational decision making with ongoing research still required on the trade-offs between individual species accuracy and broader applications. ...
Article
Full-text available
Purpose of Review The increasing availability of three-dimensional point clouds, including both airborne laser scanning and digital aerial photogrammetry, allow for the derivation of forest inventory information with a high level of attribute accuracy and spatial detail. When available at two points in time, point cloud datasets offer a rich source of information for detailed analysis of change in forest structure. Recent Findings Existing research across a broad range of forest types has demonstrated that those analyses can be performed using different approaches, levels of detail, or source data. By reviewing the relevant findings, we highlight the potential that bi- and multi-temporal point clouds have for enhanced analysis of forest growth. We divide the existing approaches into two broad categories— – approaches that focus on estimating change based on predictions of two or more forest inventory attributes over time, and approaches for forecasting forest inventory attributes. We describe how point clouds acquired at two or more points in time can be used for both categories of analysis by comparing input airborne datasets, before discussing the methods that were used, and resulting accuracies. Summary To conclude, we outline outstanding research gaps that require further investigation, including the need for an improved understanding of which three-dimensional datasets can be applied using certain methods. We also discuss the likely implications of these datasets on the expected outcomes, improvements in tree-to-tree matching and analysis, integration with growth simulators, and ultimately, the development of growth models driven entirely with point cloud data.
... The Composite-2-Change approach generated the annual best-available pixel (BAP) image composites by inputting optimal observations for each pixel [51]. The spectral trend analysis was applied to the pixel-level time series of these annual composites to further remove noisy observations (such as haze and unscreened clouds) and infill data gaps (due to missing observations and severe cloud cover) with proxy values [52]. The classification image set was derived from all the spectral bands provided by the Shanxi-wide pixel composites using a random forest classifier available in GEE. ...
Article
Full-text available
Detecting the evolution of large-area landscape patterns using long-term remote-sensing images is helpful in supporting research on the relationship between landscape patterns and ecological processes, as well as the development of ecological process simulations and spatiotemporal interaction models. However, detection methods have generally been developed as separate applications, each with a separate type of landscape pattern change; remote-sensing images are acquired at epochal timesteps. Consequently, in practical applications, many omission changes for some types of pattern changes and inaccurate evolution time are presented in the detected map. In this article, state-and-evolution detection models (SEDMs) are promoted to obtain complete information about the evolution of landscape patterns based on yearly land cover data. In the proposed framework, we first define the major categories of landscape pattern changes to comprehensively reveal the characteristics of landscape pattern changes associated with real change cases. Next, a morphological rule-based pattern recognition approach is proposed for quantitative discrimination among these categories. This approach is then applied in annual land cover data to continuously detect landscape pattern evolution processes and evolution time. Finally, the detected evolution time in different evolution processes is applied to measure the timestep between two disparate types. The performances of the SEDMs are presented by Landsat-derived land cover evolution in Shanxi, China. The detected results are indirectly verified by the land cover conversion matrix and connect index, indicating strong robustness and generalization ability of the SEDMs.
... Other instances occurred as a result of the Landsat 7 scan line corrector failure (Wulder et al., 2016), where missing observations were gap-filled with data from the following year. Those artefacts are known from previous research (Hermosilla et al., 2015a) and we here applied a spatial filter to address them: We iteratively merged all patches that shared at least one edge, and had consecutive disturbance years, into one continuous patch. We then assigned the disturbance year of the final patch by majority vote across all pixels. ...
Article
Abiotic forest disturbances are an important driver of ecosystem dynamics. In Europe, storms and fires have been identified as the most important abiotic disturbances in the recent past. Yet, how strongly these agents drive local disturbance regimes compared to other agents (e.g., biotic, human) remains unresolved. Furthermore, whether storms and fires are responsible for the observed increase in forest disturbances in Europe is debated. Here, we provide quantitative evidence for the prevalence of storm and fire disturbances in Europe 1986‐2016. For 27 million disturbance patches mapped from satellite data we determined whether they were caused by storm or fire, using a random forest classifier and a large reference dataset of true disturbance occurrences. We subsequently analyzed patterns of disturbance prevalence (i.e., the share of an agent on the overall area disturbed) in space and time. Storm‐ and fire‐related disturbances each accounted for approximately 7 % of all disturbances recorded in Europe in the period 1986‐2016. Storm‐related disturbances were most prevalent in western and central Europe, where they locally account for >50% of all disturbances, but we also identified storm‐related disturbances in south‐eastern and eastern Europe. Fire‐related disturbances were a major disturbance agent in southern and south‐eastern Europe, but fires also occurred in eastern and northern Europe. The prevalence and absolute area of storm‐related disturbances increased over time, whereas no trend was detected for fire‐related disturbances. Overall, we estimate an average of 127,716 (97,680 – 162,725) ha of storm‐related disturbances per year and an average of 141,436 (107,353 – 181,022) ha of fire‐related disturbances per year. We conclude that abiotic disturbances caused by storm and fire are important drivers of forest dynamics in Europe, but that their influence varies substantially by region. Our analysis further suggests that increasing storm‐related disturbances are an important driver of Europe’s changing forest disturbance regimes.
... The user interface requires no coding expertise and uses the 3I3D algorithm for two purposes: (i) to map forest disturbance from S2 imagery, and (ii) to assist in the estimation of areas of change and related confidence intervals at spatial scales ranging from a few hectares to entire continents. The 3I3D procedure implements the stratified estimator reported by Cochran (1977, p.95) and McRoberts et al. (2002) to reduce the SEs of the disturbance area estimates by increasing the sampling intensity in the mapped change areas that are most subject to greater variability in the form of commission and omission errors, i.e., near the boundaries of predicted forest disturbances Laurin et al., 2021;Francini et al., 2020;Hermosilla et al., 2015a;Hermosilla et al., 2015b). ...
Article
Full-text available
Forest disturbance monitoring is critical for understanding forest-related greenhouse gas emissions and for determining the role of forest management in mitigating climate change. Multiple algorithms for the automated mapping of forest disturbance using remotely sensed imagery have been developed and applied; however, variability in natural and anthropogenic disturbance phenomena, as well as image acquisition conditions, can result in maps that may be incomplete or that contain inaccuracies that prevent their use for directly estimating areas of disturbance. To reduce errors in reporting disturbance areas, stratified estimators can be applied to obtain statistically robust area estimates, while simultaneously circumventing the need to conduct a complete census or in situations where such a census may not be possible. We present a semi-automated procedure for implementation in Google Earth Engine, 3I3D-GEE, for regional to global mapping of forest disturbance (including clear-cut harvesting, fire, and wind damage) and sample-based estimation of related areas using data from the processing capacity of Google Earth Engine. Documentation for the application is also provided in Appendix A. Using Sentinel-2 (S2) imagery, our procedure was applied and tested for 2018 in Italy for which the approximately 11 million ha of forests (mostly Q. pubescens, Q. robur, Q. cerris, Q. petraea, and Fagus sylvatica) serve as an appropriate case study because national statistics on forest disturbance areas are not available. To decrease the overall standard errors of the area estimates, the sampling intensities in areas where greater variability in the form of greater commission and omission errors are expected can be increased. To this end, we augmented the predicted forest disturbance map with a buffer class consisting of a two-pixel buffer (20 m) on each side of the disturbance class boundary. We selected a reference sample of 19,300 points: a simple random sample of 9,300 points from the buffer and simple random samples of 5000 from each of the undisturbed and disturbed classes. The reference sample was photointerpreted using fine resolution orthophotos (30 cm) and S2 imagery. While the estimate of the disturbed area obtained by adding the areas of pixels classified as disturbed was 41,732 ha, the estimate obtained using the unbiased stratified estimator was 27% greater at 57,717±716 ha. Regarding map accuracy, we found several omission errors in the buffer (53.4%) but none (0%) in the undisturbed map class. Similarly, among the 1035 commission errors, the majority (7 4 4) were in the buffer class. The methods presented herein provide a useful tool that can be used to estimate areas of forest disturbance, which many nations must report as part of their commitment to international conventions and treaties. In addition, the information generated can support forest management, enabling the forest sector to monitor stand-replacing forest harvesting over space and time.
... In most cases, the gradual change of forest refers to the process in which the forest changes gradually after moderate disturbance; however, the slow recovery process after the forest encounters a large abrupt change is also a gradual change process . This recovery time is long and can be easily further disturbed from diseases and insect pests during restoration (Hermosilla et al., 2015;Huang et al., 2010;Kennedy et al., 2012;. This study focused on the gradual change in the process of forest growth and did not consider the recovery process. ...
Article
Gradual change is prevalent across the forest landscape and generates long-lasting effects for the landscape surface; thus, tracking long-term gradual change can effectively characterize forest change processes. The objective of this study was to establish a vegetation index for change monitoring so as to determine long-term gradual change processes of forest ecosystems in typical red soil areas. The study area was located in Hengyang City, Hunan Province, China. Landsat images, field survey, and auxiliary data were collected to devise a disturbance sensitive vegetation index (DSVI) as an indicator of forest change. Long-term (1985–2019) forest changes were detected using the LandTrendr algorithm on the Google Earth Engine (GEE), while three evaluation aspects of velocity, frequency, and variance were used to analyze the processes of forest gradual change in red soil regions. Results indicate that the DSVI is a suitable index for forest change detection due to its stronger sensitivity compared to other indexes. Further, it shows excellent change detection ability for different types of gradual changes, such as those caused by drought and significant soil erosion. Furthermore, 97.26% of the forests showed gradual change, and approximately 2/3 of monitored forests showed an increasing growth trend while 1/3 showed a decreasing trend. The dominant (28.33%) forest disturbance frequency indicated instability in red soil regions. Dispersion degree of forest variance was mainly low (48.46%) or medium (28.84%). This research establishes the DSVI as a promising method to track forest gradual change and contributes to a better understanding of gradual change processes of forest landscape over time.
... The need to manage data gaps and the need to detect forest disturbance even on areas often covered by clouds motivated the development of dense-stack change point detection methods, which give satisfying results from that point of view (Zhang and Zhang 2020;Nyland et al. 2018;Coulter et al. 2016). An alternative methodology to circumvent this issue is by incorporating into 3I3D a preprocessing step to calculate Best Available Pixel (BAP) cloud free composites (Hermosilla et al. 2015a(Hermosilla et al. , 2015b(Hermosilla et al. , 2016Thompson et al. 2015) which are constructed on a basis of a set of rules that are used to identify the 'best' pixel observation among several images (White et al. 2014). Thus, a huge amount of data should be downloaded to process large areas or several years. ...
Article
Full-text available
Although estimating forest disturbance area is essential in the context of carbon cycle assessments and for strategic forest planning projects, official statistics are currently not available in several countries. Remotely sensed data are an efficient source of auxiliary information for meeting these needs, and multiple algorithms are commonly used worldwide for this purpose. However, both more accurate maps and precise area estimates are strongly required, especially in Mediterranean ecosystems, and scientific research in this topic area is anything but concluded. In this study, we present the new Three Indices Three Dimensions (3I3D) algorithm for the automated prediction of forest disturbances using statistical analyses of Sentinel-2 data. We tested 3I3D in Tuscany, Italy, for the year 2016, and we compared the results to those obtained using the Global Forest Change Map (GFC), LandTrendr (LT), and the Two Thresholds Method (TTM). The 3I3D map was the most accurate (omissions = 27%, commissions = 30%) followed by TTM (omissions = 35%, commissions = 39%), LT (omissions = 41%, commissions = 43%) and lastly GFC with slightly fewer omissions than LT (39%) but with many more commissions (69%). We also presented a probability sampling framework to estimate the forest harvested area using a model-assisted estimator that can be used at an operational level to produce large-scale statistics. 3I3D and TTM produced the smallest standard errors of the area estimates (8%) followed by LT (13%) and GFC (17%).
... Data from outside of the collection date window were excluded and pixels where no observations were available (e.g., persistently cloudy locations) were labelled as data gaps (White et al., 2014). These BAP composites were further processed to remove unscreened noise (such as from thin clouds, haze or smoke) as well as non-permanent snow occurrences, and fill any remaining data gaps with proxy surface reflectance values using spectral trend information derived the temporal analysis of the time series (Hermosilla et al., 2015). ...
Article
Full-text available
Rapid climate warming has widely been considered as the main driver of recent increases in Arctic tundra productivity. Field observations and remote sensing both show that tundra “greening” has been widespread, but heterogeneity in regional and landscape-scale trends suggest that additional controls are mediating the response of tundra vegetation to warming. In this study, we examined the relationship between changes in vegetation productivity in the western Canadian Arctic and biophysical variables by analyzing trends in the Enhanced Vegetation Index (EVI) obtained from nonparametric regression of annual Landsat surface reflectance composites. We used Random Forests classification and regression tree modelling to predict the trajectory and magnitude of greening from 1984 to 2016 and identify biophysical controls. More than two-thirds of our study area showed statistically significant increases in vegetation productivity, but observed changes were heterogeneous, occurring most rapidly within areas of the Southern Arctic that were: (1) dominated by dwarf and upright shrub cover types, (2) moderately sloping, and (3) located at lower elevation. These findings suggest that the response of tundra vegetation to warming is mediated by regional- and landscape-scale variation in microclimate, topography and soil moisture, and physiological differences among plant functional groups. Our work highlights the potential of the joint analysis of annual remotely sensed vegetation indices and broad-scale biophysical data to understand spatial variation in tundra vegetation change.
... Es en este punto en que la información derivada de los sensores remotos se convierte en una importante herramienta para detectar estas variaciones (Li, Qu y Hao 2010;Colditz y Dech 2007;Hermosilla et al. 2015;Vogelmann et al. 2012). Lo anterior es posible debido a que existe una gran relación entre los índices de vegetación (IV) y la biomasa aérea, el verdor de la vegetación y en general el estado de salud de la cobertura vegetal (Anaya, Chuvieco y Palacios 2008). ...
Article
Full-text available
Climate change and global warming are caused principally by anthropogenic activities. For this reason, understanding the research lines that relate Land Surface Temperature and Vegetation Index time series is of great importance, given the amplitude of different open scientific areas on global warming. The result of this classification is presented to the academic community, which divides the studies into two main representative areas in the study of climate change: (1) Geodata Modeling and Analysis and (2) Remote Sensing. From the last one, two types are derived, some constructed with Linnear Regression Analysis (RL) and others with Nonlinear Regression Analysis (RNL). On the Geodata Modeling and Analysis, the Global Climate Models (GCM) are not the right tool for these analyzes due to their coarse spatial resolution. This implies the development of hybrid models with remote sensing, which are also limited by differences in resolution. On the other hand, remote sensing is the most widely disseminated tool for this type of studies. Finally, a promising window for development in the time series opens with non-linear regression analysis.
... These scoring functions consider acquisition date (August 1st ± 30 days), acquisition sensor (Landsat-7 ETM+ imagery is penalized after the scan line corrector malfunction in 2003), presence of clouds (<70% cloud cover in image) and haze, and distance to clouds and cloud shadows. The resulting annual image composites are then further analyzed to remove unscreened, yet non-representative (e.g., haze, smoke), observations and fill remaining data gaps (pixels with no suitable observations) with proxy values informed by spectral trend analysis (Hermosilla et al., 2015a). Noisy observations are distinguished from actual changes by analyzing the spectral differences between preceding and subsequent years in each of the six Landsat bands considered. ...
Article
Full-text available
Deriving land cover from remotely sensed data is fundamental to many operational mapping and reporting programs as well as providing core information to support science activities. The ability to generate land cover maps has benefited from free and open access to imagery, as well as increased storage and computational power. The accuracy of the land cover maps is directly linked to the calibration (or training) data used, the predictors and ancillary data included in the classification model, and the implementation of the classification, among other factors (e.g., classification algorithm, land cover heterogeneity). Various means for improving calibration data can be implemented, including using independent datasets to further refine training data prior to mapping. Opportunities also arise from a profusion of possible calibration datasets from pre-existing land cover products (static and time series) and forest inventory maps through to observation from airborne and spaceborne lidar observations. In this research, for the 650 Mha forested ecosystems of Canada, we explored approaches to refine calibration data, integrate novel predictors, and optimize classifier implementation. We refined calibration data using measures of forest vertical structure, integrated novel spatial (via distance-to metrics) model predictors, and implemented a regionalized approach for optimizing training data selection and model-building to ensure local relevance of calibration data and capture of regional variability in land cover conditions. We found that additional vetting of training data involved the removal of 44.7% of erroneous samples (e.g. treed vegetation without vertical structure) from the training pool. Nationally, distance to ephemeral waterbodies was a key predictor of land cover, while the importance of distance to permanent water bodies varied on a regional basis. Regionalization of model implementation ensured that classification models used locally relevant descriptors and resulted in improved classification outcomes (overall accuracy: 77.9% ± 1.4%) compared to a generalized, national model (70.3% ± 2.5%). The methodological developments presented herein are portable to other land cover projects, monitoring programs, and remotely sensed data sources. The increasing availability of remotely sensed data for land cover mapping, as well as non-image data for aiding with model development (from calibration data to complementary spatial data layers) provide new opportunities to improve and further automate land cover mapping procedures.
... Generating seamless, cloud-free satellite image mosaics is difficult in mountainous terrain because of the comparatively short snow-free season for image acquisitions and the strong influence of shadows when sun angles are low (Bindschadler et al., 2008;Hansen and Loveland, 2012;Hermosilla et al., 2015;Mateo-García et al., 2018). Our approach for annual image mosaics of the BCA includes Landsat scenes that contain less than 30% cloud cover and those acquired between day of year (DOY) 210 and 250, or Jul. ...
Article
Full-text available
The retreat of mountain glaciers affects mountain hazards and hydrology, and new methods are needed to rapidly map glacier retreat at planetary scales. We automatically map 14,329 glaciers (30,063 km 2) in British Columbia and Alberta, Canada, from 1984 to 2020 using satellite image archives from the Landsat 4, 5, 7 and 8 missions and reveal an acceleration in area loss that commenced in 2011. Glacier fragmentation, disappearance, and proglacial lake development also accelerated, as did the retreat of glaciers to higher elevations. Our annually-resolved method relies on the existence of previously published and manually validated glacier inventories from the mid-1980s and mid-2000's. Our methods performed well for clean ice glaciers, had occasional errors when proglacial lakes were present, and consistently underestimated the area of debris-covered glaciers. Clean ice glacier area loss accelerated sevenfold between the early [1984-2010] and late [2011− 2020] epochs. This acceleration yielded rates of area shrinkage of − 49 ± 7 km 2 a − 1 [early] and − 340 ± 40 km 2 a − 1 [late] with accelerated losses (32-fold increase) for small glaciers on Vancouver Island over the last decade. Glacier fragmentation accelerated from 26 ± 5.6 fragments a − 1 to 88 ± 39 fragments a − 1. About 1141 glaciers fell below our 0.05 km 2 detection limit and so disappeared from our database, representing a loss of 8%. Proglacial lake area growth accelerated from 9.2 ± 1.1 km 2 a − 1 to 49 ± 4.5 km 2 a − 1. We also observed an acceleration in the upwards migration of median glacier elevations for clean ice glaciers from 0.31 ± 0.08 m a − 1 to 4.7 ± 0.7 m a − 1. Our workflow demonstrates the advantages of annual resolution glacier inventories and contributes towards the implementation of planetary mapping of glaciers and glacier attributes at annual resolution.
... The second alternative, where sufficient archive is available, is to generate temporal aggregates around pivot dates, allowing the use of information contained in partially cloudy images, at the expense of a reduced precision on the timing of detected changes (Hansen et al., 2013). Finally for dense temporal time series, pixel-level temporal analysis can be performed (Hermosilla et al., 2015) to optimize the use of the available data and the temporal precision of event detections. In this study, we were interested in the temporal hindsight, and therefore had to deal with very sparse series prior to the launch of Landsat 7 and 8 (and due to the long absence of ground stations in Africa). ...
Article
Woody encroachment and forest progression are widespread in forest-savanna transitional areas in Central Africa. Quantifying these dynamics and understanding their drivers at relevant spatial scales has long been a challenge. Recent progress in open access imagery sources with improved spatial, spectral and temporal resolution combined with cloud computing resources, and the advent of relatively cheap solutions to deploy laser sensors in the field, have transformed this domain of study. We present a study case in the Mpem & Djim National Park (MDNP), a 1,000 km² protected area in the Centre region of Cameroon. Using open source algorithms in Google Earth Engine (GEE), we characterized vegetation dynamics and the fire regime based on Landsat multispectral imagery archive (1975–2020). Current species assemblages were estimated from Sentinel 2 imagery and the open source biodivMapR package, using spectral dissimilarity. Vegetation structure (aboveground biomass; AGB) was characterized using Unmanned Aerial vehicle (UAV) LiDAR scanning data sampled over the study area. Savanna vegetation, which was initially dominant in the MDNP, lost about 50% of its initial cover in <50 years in favor of forest at an average rate of ca. 0.63%.year⁻¹ (6 km².year⁻¹). Species assemblage computed from spectral dissimilarity in forest vegetation followed a successional gradient consistent with forest age. AGB accumulation rate was 3.2 Mg.ha⁻¹.year⁻¹ after 42 years of forest encroachment. In savannas, two modes could be identified along the gradient of spectral species assemblage, corresponding to distinct AGB levels, where woody savannas with low fire frequency store 40% more AGB than open grassy savannas with high fire frequency. A fire occurrence every five year was found to be the fire regime threshold below which woody savannas start to dominate over grassy ones. A fire frequency below that threshold opens the way to young forest transitions. These results have implications for carbon sequestration and biodiversity conservation policies. Maintaining savanna ecosystems in the region would require active management actions to limit woody encroachment and forest progression, in contradiction with global reforestation goals.
... Alternatively, where sufficient archive is available, temporal aggregates can be produced around pivot dates, allowing the use of information contained in partially cloudy images, at the expense of a reduced precision on the timing of detected changes. Finally, for dense temporal series, time series analysis on pixels can be performed (Hermosilla et al., 2015), to optimize the use of the available data and the temporal precision of event detection. In this study, the temporal hindsight was prioritized, and therefore had to deal with very sparse series prior to the launch of Landsat 7. The third approach was thus excluded. ...
Thesis
Full-text available
Understanding the effects of global change (combining anthropic and climatic pressures) on biome distribution needs innovative approaches allowing to address the large spatial scales involved and the scarcity of available ground data. Characterizing vegetation dynamics at landscape to regional scale requires both a high level of spatial detail (resolution), generally obtained through precise field measurements, and a sufficient coverage of the land surface (extent) provided by satellite images. The difficulty usually lies between these two scales as both signal saturation from satellite data and ground sampling limitations contribute to inaccurate extrapolations. Airborne laser scanning (ALS) data has revolutionized the trade-off between spatial detail and landscape coverage as it gives accurate information of the vegetation’s structure over large areas which can be used to calibrate satellite data. Also recent satellite data of improved spectral and spatial resolutions (Sentinel 2) allow for detailed characterizations of compositional gradients in the vegetation, notably in terms of the abundance of broad functional/optical plant types. Another major obstacle comes from the lack of a temporal perspective on dynamics and disturbances. Growing satellite imagery archives over several decades (45 years; Landsat) and available computing facilities such as Google Earth Engine (GEE) provide new possibilities to track long term successional trajectories and detect significant disturbances (i.e. fire) at a fine spatial detail (30m) and relate them to the current structure and composition of the vegetation. With these game changing tools our objective was to track long-term dynamics of forest-savanna ecotone in the Guineo-Congolian transition area of the Central Region of Cameroon with induced changes in the vegetatio structure and composition within two contrasted scenarios of anthropogenic pressures: 1) the Nachtigal area which is targeted for the dam construction and subject to intense agricultural activities and 2) the Mpem et Djim National Park (MDNP) which has no management plan. The maximum likelihood classification of the Spot 6/7 image aided with the information from the canopy height derived from ALS data discriminated the vegetation types within the Nachtigal area with good accuracy (96.5%). Using field plots data in upscaling aboveground biomass (AGB) form field plots estimates to the satellite estimates with model-based approaches lead to a systematic overestimation in AGB density estimates and a root mean squared prediction error (RMSPE) of up to 65 Mg.ha−1 (90%), whereas calibration with ALS data (AGBALS) lead to low bias and a drop of ~30% in RMSPE (down to 43 Mg.ha−1, 58%) with little effect of the satellite sensor used. However, these results also confirm that, whatever the spectral indices used and attention paid to sensor quality and pre-processing, the signal is not sufficient to warrant accurate pixel wise predictions, because of large relative RMSPE, especially above (200–250 Mg.ha−1). The design-based approach, for which average AGB density values were attributed to mapped land cover classes, proved to be a simple and reliable alternative (for landscape to region level estimations), when trained with dense ALS samples. AGB and species diversity measured within 74 field inventory plots (distributed along a savanna to forest successional gradient) were higher for the vegetation located in the MDNP compared to their pairs in the Nachtigal area. The automated unsupervised long-term (45 years) land cover change monitoring from Landsat image archives based on GEE captured a consistent and regular pattern of forest progression into savanna at an average rate of 1% (ca. 6 km².year-1). No fire occurrence was captured for savanna that transited to forest within five years of monitoring. Distinct assemblages of spectral species are apparent in forest vegetation which is consistent with the age of transition. As forest gets older AGBALS recovers at a rate of 4.3 Mg.ha-1.year-1 in young forest stands (< 20 years) compared to 3.2 Mg.ha-1.year-1 recorded for older forest successions (≥ 20 years). In savannas, two modes could be identified along the gradient of spectral species assemblage, corresponding to distinct AGBALS levels, where woody savannas with low fire frequency store 50% more carbon than open grassy savannas with high fire frequency. At least two fire occurrences in five years is found to be the fire regime threshold below which woody savannas start to dominate over grassy ones. Four distinct plant communities were found distributed along a fire frequency gradient. However the presence of fire-sensitive pioneer forest species in all scenarios of fire frequencies (from low to high fire frequencies) would suggest that the limiting effect of fire on woody vegetation is not sufficient to hinder woody encroachment in the area bringing therefore sufficient humidity required for the establishment of pioneer forest saplings within open savannas. These results have implications for carbon sequestration and biodiversity conservation policies. The maintenance of the savanna ecosystem in the region would require active management actions, and contradicts reforestation goals (REDD+, Bonn challenge, etc.).
... There is a developing and important need to understand the capabilities of satellite-based lidar products for informing and updating national-level forest attribute layers. One such example is the National Terrestrial Ecosystem Monitoring System (NTEMS; White et al., 2014), which uses gap-free annual Landsat surface reflectance composites to generate land cover, disturbance (by type), and forest structural attributes (Hermosilla et al., 2015(Hermosilla et al., , 2018. Local and national airborne lidar campaigns (Wulder et al., 2012a) were used to obtain calibration and validation data to produce annual wall-to-wall estimates of forest attributes such as height, biomass, volume, and canopy cover (Matasci et al., 2018b). ...
Article
Forests represent the world's largest terrestrial ecosystem and their monitoring is therefore critical from scientific, ecological, and management perspectives. Present day sustainable forest management practices go beyond forest inventory and increasingly include aspects such as carbon accounting and regeneration assessments. Such monitoring requires often unavailable, spatially exhaustive and up-to-date information on forest attributes over broad areas. Recent developments in the acquisition of broad-scale forest attribute information from remotely sensed data has included the use of multiple technologies that take advantage of globally available data products to derive forest attribute layers. However, less is known about the applicability and performance of such products when used to produce broad-scale, accurate, and up-to-date forest information products. This study aimed to evaluate the agreement between two broad-scale forest canopy height products – Ice, Cloud, and land Elevation Satellite 2 (ICESat-2) and the National Terrestrial Ecosystem Monitoring System (NTEMS) imputed canopy height layers for Canada – across a variety of ecological gradients. Overall, the two datasets showed high correspondence, with a root-mean-square difference of 4.87 m, and 85% of ICESat-2 canopy heights falling within the 95% confidence interval of the NTEMS height estimate. Across ecozones, canopy heights in the Taiga Shield West and Boreal Shield West had stronger agreement (91% of ICESat-2 segments within the 95% confidence interval of NTEMS), while the Taiga Cordillera and Taiga Shield East had lower agreement (< 75% of ICESat-2 segments within the 95% confidence interval of NTEMS). Interestingly, we found that the modeled heights based upon optical satellite data had a less generalized distribution than heights from ICESat-2 as well as achieving a greater representation for the taller (overall and by ecozone) height classes. An increase in absolute difference between data products was also found as a function of increasing slope. Finally, the correspondence between products was evaluated across disturbed areas (35 to 10 years since disturbance) to assess the agreement of the two products in areas of regenerating forest. In general, the analysis found that burned areas, which tend to be more structurally heterogeneous, had lower agreement between products then harvested areas. The high overall correspondence between the data products demonstrate the potential for integration of ICESat-2 to inform (via calibration / validation) or update height products based upon optical satellite data.
... (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Supported by the opening of the archive and expanding computing resources, change detection methods have evolved from two-date comparisons (Coppin et al., 2004) to include time series analyses (Zhu, 2017) that uses annual composite imagery (Hermosilla et al., 2015b;Huang et al., 2010;Kennedy et al., 2010) and all available imagery (Zhu and Woodcock, 2014) as well as ensembles of such algorithms (Healey et al., 2018). ...
Article
Full-text available
Since 1972, the Landsat program has been continually monitoring the Earth, to now provide 50 years of digital, multispectral, medium spatial resolution observations. Over this time, Landsat data were crucial for many scientific and technical advances. Prior to the Landsat program, detailed, synoptic depictions of the Earth's surface were rare, and the ability to acquire and work with large datasets was limited. The early years of the Landsat program delivered a series of technological breakthroughs, pioneering new methods, and demonstrating the ability and capacity of digital satellite imagery, creating a template for other global Earth observation missions and programs. Innovations driven by the Landsat program have paved the way for subsequent science, application , and policy support activities. The economic and scientific value of the knowledge gained through the Landsat program has been long recognized, and despite periods of funding uncertainty, has resulted in the program's 50 years of continuity, as well as substantive and ongoing improvements to payload and mission performance. Free and open access to Landsat data, enacted in 2008, was unprecedented for medium spatial resolution Earth observation data and substantially increased usage and led to a proliferation of science and application opportunities. Here, we highlight key developments over the past 50 years of the Landsat program that have influenced and changed our scientific understanding of the Earth system. Major scientific and pro-grammatic impacts have been realized in the areas of agricultural crop mapping and water use, climate change 2 drivers and impacts, ecosystems and land cover monitoring, and mapping the changing human footprint. The introduction of Landsat collection processing, coupled with the free and open data policy, facilitated a transition in Landsat data usage away from single images and towards time series analyses over large areas and has fostered the widespread use of science-grade data. The launch of Landsat-9 on September 27, 2021, and the advanced planning of its successor mission, Landsat-Next, underscore the sustained institutional support for the program. Such support and commitment to continuity is recognition of both the historic impact the program, and the future potential to build upon Landsat's remarkable 50-year legacy.
... Accurate measurements are best retrieved in cloud-free regions where robust temporal observations can occur throughout the growing season (Frolking et al. 2006). Cloudy days, haze, or smoke from wildfires can reduce quality observations to once a month or less and create spatial gaps in the vineyard monitoring season (Hermosilla et al. 2015;McNairn and Shang 2016). ...
Article
Full-text available
The launch of NASA’s ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) and the European Space Agency’s Sentinel-1A/B synthetic aperture radar (SAR) satellites provides the opportunity to advance a multi-sensor remote sensing approach to crop monitoring. While ECOSTRESS and Sentinel-1A/B have been used separately to assess vegetation conditions, a study that quantifies the synergistic usefulness of both to monitor crops has not been performed. This study assesses the complementary uses of Sentinel-1A SAR and ECOSTRESS land surface temperature (LST) and evapotranspiration (ET) datasets to assess vine growth and conditions in blocks located in Sonoma County, California for 2018. Results indicate Sentinel-1A SAR dual-polarization backscatter measurements (σVV0 and σVH0) have different sensitivities to vine leafiness and moisture content, based on measured vineyard field data and radiometric modeling. SAR and modeled σVV0 backscatter suggest higher sensitivity to surface conditions and trunk and cane moisture, while SAR and modeled σVH0 backscatter indicate higher sensitivity to vine leafiness and canopy moisture. ECOSTRESS LST measurements were sharpened to a 30 m resolution using a data mining sharpener and ET measurements were generated with a retrieval algorithm approach for select dates. Spearman’s rank correlation and linear regressions analyses between SAR backscatter to ECOSTRESS datasets indicate stronger relationships between σVH0 backscatter to LST and ET relative to σVV0 backscatter. The results suggest Sentinel-1A SAR σVH0 backscatter can provide indications of vine leaf volume and moisture state that can be related to LST and ET measurements, providing useful information for vineyard management.
... Scores (i) and (ii) are applied to the whole image, while scores (iii) and (iv) are applied to each pixel. BAP composites calculation was performed using the GEE application developed by Francini et al. (2021) which allows calculating and downloading BAP composites (Hermosilla et al. 2015a, 2015b, White et al. 2017. For more information on the criteria involved in the BAP pixels selection see the BAP-GEE documentation . ...
Article
Full-text available
The study of afforestation is crucial to monitor land transformations and represents a central topic in sustainable development procedures, in terms of climate change, ecosystem services monitoring, and planning policies activities. Although surveying afforestation is important, the assessment of the growing forests is difficult, since land cover has different durations depending on the species. In this context, remote sensing can be a valid instrument to evaluate the afforestation process. Nevertheless, while a vast literature on forest disturbance exists, only a few studies focus on afforestation and almost none directly exploits remote sensing data. This study aims to automatically classify non-forest, afforestation, and forest areas using remote sensing data. To this purpose, we constructed a reference dataset of 61 polygons that suffered a change from non-forest to forest in the period 1988-2020. The reference data were constructed with the Land Use Inventory of Italy and through photointer-pretation of orthophotos (1988-2012, spatial resolution 50 × 50 cm) and very high-resolution images (2012-2020, spatial resolution 30 × 30 cm). Using Land-sat Best Available Pixel composites time-series (1984-2020) we calculated 52 temporal predictors: four temporal metrics (median, standard deviation, Pear-son's correlation coefficient R, and slope) calculated for 13 different bands (the six Landsat spectral bands, three Spectral Vegetation Indices, and four Tasseled Cap Indices). To verify the possibility of distinguishing afforestation from non-forest and forest, given the differences between them can be minimal , we tested four different models aiming at classifying the following categories: (i) non-forest/afforestation, (ii) afforestation/forest, (iii) non-forest/ forest and (iv) non-forest/afforestation/forest. Temporal predictors were used with random forest which was calibrated using random search, validated using k-fold Cross-Validation Overall Accuracy (OAcv), and further using out-of-bag independent data (OAoob). Results illustrate that the distinction of af-forestation/forest reaches the largest OAcv (87%), followed by non-forest/for-est (83%), non-forest/afforestation (75%) and non-forest/afforestation/forest (72%). The different OA values confirm that the difference in photosynthetic activity between forest and afforestation can be analysed through remote sensing to distinguish them. Although remote sensing data are currently not exploited to monitor afforestation areas our results suggest it may be a valid support for country-level monitoring and reporting.
... The opening of the Landsat archive for free access in 2008 (Woodcock et al., 2008) marked a new era for utilizing Earth observation data to document changes to terrestrial ecosystems at large space scales, including developing novel methods in forest disturbance mapping (Hansen and Loveland, 2012). Many algorithms have been developed to estimate the occurrence location, extent, time and duration of disturbances by detecting changes in spectral trajectory from Landsat time series (DeVries et al., 2015;He et al., 2011;Hermosilla et al., 2015;Huang et al., 2010a;Kennedy et al., 2010;White et al., 2017;Zhu et al., 2019b). Forest disturbance products have been produced at national to global scales (Hansen et al., 2013;White et al., 2017;Zhao et al., 2018). ...
Article
Full-text available
Forest disturbances can have broad impact on the climate, local environment, and the regeneration of the forest ecosystem. The nature and magnitude of such impact is largely driven by disturbance intensity. In this study, by integrating field plot measurements collected by the Forest Inventory and Analysis program with time series Landsat observations, we produced the first set of annual forest disturbance intensity map products quantifying the percentage of basal area removal (PBAR) at the 30-m resolution for the conterminous United States (CONUS) from 1986 to 2015. The derived map products revealed that during the 30-year study period, the annual average PBAR values of all disturbed pixels across CONUS ranged from 66% to 70%, and the proportion of those pixels having stand-clearing disturbances ranged from 40% to 58%. High disturbance intensities were concentrated in the Southeastern states from Texas to Virginia and along the Pacific coast and the Cascades in the West. At the national scale, the annual mean PBAR and proportion of stand clearing area (PSCA) values both appeared to follow second order trajectories, with increasing trends at the beginning, decreasing trends towards the end, and turning points around 2003. Overall, there is a net increase of 2% in PBAR and 3% in PSCA from 1986 to 2015. The temporal trends of PBAR and PSCA were also investigated at state and ecoregion levels, with substantial differences found among many states and ecoregions. While states and ecoregions generally follow second order trajectories, the majority had increasing trends throughout much of the study period, reflecting higher disturbance intensities during the later years compared to earlier years. Large increase (>10%) in PBAR was seen in several states (e.g., Virginia, Arkansas, and Minnesota) and ecoregions (e.g., Northern Minnesota Wetlands); however, large decreases (>10%) in PBAR were not observed in any states, and were seen in only one ecoregion, the Blue Mountains in the southeast. The disturbance intensity maps are available from a web portal of the Oak Ridge National Laboratory Distributed Active Archive Center (ORNL-DAAC) at https://doi.org/10.3334/ORNLDAAC/2059.
... Since BAP compiles cloud-free images by selecting the best available observation based on user-defined criteria (Gomez et al., 2016;Griffiths et al., 2013), the BAP composites can retain the source image information from which they came. In addition, since BAP can ensure phenological consistency between multitemporal BAP composites by setting the acquisition day-of-year (DOY) criteria Chen et al., 2021), it is suitable for multiyear change detection and assessment Gomez et al., 2016;Hermosilla et al., 2015;. Accordingly, the BAP method was used to generate cloud-free composites. ...
Article
Full-text available
Monitoring the water clarity of lakes is essential for the sustainable development of human society. However, existing water clarity assessments in China have mostly focused on lakes with areas > 1 km2, and the monitoring periods were mainly in the 21st century. In order to improve the understanding of spatiotemporal variations in lake clarity across China, based on the Google Earth Engine cloud platform, a 30 m long-term LAke Water Secchi depth (SD) dataset (LAWSD30) of China (1985–2020) was first developed using Landsat series imagery and a robust water-color parameter-based SD model. The LAWSD30 dataset exhibited a good performance compared to concurrent in situ SD datasets, with an R2 of 0.86 and a root mean square error of 0.225 m. Then, based on our LAWSD30 dataset, long-term spatiotemporal variations in SD for lakes > 0.01 km2 (N = 40 973) across China were evaluated. The results show that the SD of lakes with areas ≤ 1 km2 exhibited a significant downward trend in the period of 1985–2020, but the decline rate began to slow down and stabilized after 2001. In addition, the SD of lakes with an area > 1 km2 showed a significant downward trend before 2001, and began to increase significantly afterwards. Moreover, in terms of the spatial patterns, the proportion of small lakes (area ≤ 1 km2) showing a decreasing SD trend was the largest in the Mongolian–Xinjiang Plateau Region (MXR) (about 30.0 %), and the smallest in the Eastern Plain Region (EPR) (2.6 %). In contrast, for lakes > 1 km2, this proportion was the highest in MXR (about 23.0 %), and the lowest in the Northeast Mountain Plain Region (NER) (16.1 %). The LAWSD30 dataset and the spatiotemporal patterns of lake water clarity in our research can provide effective guidance for the protection and management of lake environment in China.
... The annual BAP composites are further refined by applying a spectral trend analysis over the Normalized Burn Ratio at pixel level in order to remove unscreened noise, detect changes and fill data gaps with temporally-interpolated values, resulting in annual, gapfree, surface-reflectance image composites (Hermosilla et al., 2015b). During this process, forest disturbances are detected, characterized and attributed to a disturbance agent (i.e., wildfire, harvest, non-stand replacing disturbances) using a Random Forests classification model via the object-based analysis approach (Hermosilla et al., 2015a) with an overall accuracy of 92% AE 2% (Hermosilla et al., 2016). ...
Article
Full-text available
Protected areas (PA) are an effective means of conserving biodiversity and protecting suites of valuable ecosystem services. Currently, many nations and international governments use proportional area protected as a critical metric for assessing progress towards biodiversity conservation. However, this and other common metrics do not assess the effectiveness of PA networks, nor do they assess how representative PA are of the ecosystems they aim to protect. Topography, stand structure, and land cover are all key drivers of biodiversity within forest environments, and are well suited as indicators to assess the representation of PA. Here we examine the protected area network in British Columbia, Canada, through these drivers derived from freely available data and remote sensing products across the provincial biogeoclimatic ecosystem classification system. We examine biases in the protected area network by elevation, forest disturbances, and forest structural attributes, including height, cover, and biomass by comparing a random sample of protected and unprotected pixels. Results indicate that PA are commonly biased towards high elevation and alpine land covers, and that forest structural attributes of the park network are often significantly different in protected vs unprotected areas (426 out of 496 forest structural attributes found to be different; p < 0.01). Analysis of forest structural attributes suggests that establishing additional PA could ensure representation of various forest structure regimes across British Columbia’s ecosystems. We conclude that these approaches using free and open remote sensing data are highly transferable, and can be accomplished using consistent datasets to assess PA representations globally.
... Fassnacht et al. (2016) and Yu et al. (2015) provide comprehensive reviews of RS applications on tree species classifications and forest structural variable estimations. Several studies have used optical and LiDAR data to model breeding birds' diversity and habitat distribution (Foody, 2003;Culbert et al., 2012;Vierling et al., 2013), while most studies on saproxylic beetles have focused on the distribution of endangered species related to environmental and climatic variables, such as growing stock volume, net primary production, land use, and precipitation (Hermosilla et al., 2015a;Chen et al., 2020;Della Rocca and Milanesi, 2020). Just a few studies tested the relation between RS structural metrics and saproxylic abundance, even if good results were reported. ...
Article
Full-text available
Background Rapid climate changes lead to an increase in forest disturbance, which in turn lead to growing concerns for biodiversity. While saproxylic beetles are relevant indicators for studying different aspects of biodiversity, most are smaller than 2 mm and difficult to sample. This, together with a high number of species and trophic roles, make their study remarkably challenging, time-consuming, and expensive. The Landsat mission provides data since 1984 and represents a powerful tool in this scenario. While we believe that remote sensing data cannot replace on-site sampling and analysis, in this study we aim to prove that the Landsat Time Series (TS) may support the identification of insects’ hotspots and consequently guide the selection of areas where to concentrate field analysis. Methods With this aim, we constructed a Landsat-derived NDVI TS (1984–2020) and we summarised the NDVI trend over time by calculating eight Temporal Metrics (TMs) among which four resulted particularly successful in predicting the amount of saproxylic insects: (i) the slope of the regression line obtained by linear interpolating the NDVI values over time; (ii) the Root Mean Square Error (RMSE) between the regression line and the NDVI TS; (iii) the median, and the (iv) minimum values of the NDVI TS. The study area consists of four monitoring sectors in a Mediterranean-managed beech forest located in the Apennines (Molise, Italy), where 60 window flight traps for flying beetles were installed. First, the saproxylic beetle's biodiversities of monitoring sectors were quantified in terms of species richness and alpha-diversity. Second, the capability of TMs in predicting the richness of saproxylic beetles family and trophic categories was assessed in terms of Pearson's product-moment correlation. Results The alpha diversity and species richness analysis indicate dissimilarities across the four monitored sectors (Shannon and Simpson's index ranging between 0.67 to 2.31 and 0.69 to 0.88, respectively), with Landsat TS resulting in effective predictors for estimating saproxylic beetle richness. The strongest correlation was reached between the Monotomidae family and the RMSE temporal metric (R = 0.66). The mean absolute correlation (r) between the NDVI TMs and the saproxylic community was 0.46 for Monotomidae, 0.31 for Cerambycidae, and 0.25 for Curculionidae. Conclusions Our results suggest that Landsat TS has important implications for studying saproxylic beetle distribution and, by helping the selection of monitoring areas, increasing the amount of information acquired while decreasing the effort required for field analysis.
... We uploaded the fire perimeter dataset into Earth Engine to limit the extent of image processing. We created annual growing season composite imagery with the best pixel approach (Hermosilla et al., 2015) using a harmonized Landsat image collection that included imagery from Landsat 5, 7, and 8 (Roy et al., 2016) with clouds and snow masked out. We calculated the dNBR from 1-year prefire and 1-year postfire following the extended assessment by Key and Benson (2006). ...
Article
Wildfire is an essential disturbance agent that creates burn mosaics, or a patchwork of burned and unburned areas across the landscape. Unburned patches, fire refugia, serve as carbon sinks and seed sources for forest regeneration in burned areas. In the Cajander larch (Larix cajanderi Mayr.) forests of north‐eastern Siberia, an unprecedented wildfire season in 2020 and little documentation of landscape patch dynamics have resulted in research gaps about the characteristics of fire refugia in northern latitude forests, which are warming faster than other global forest ecosystems. We aim to characterize the 2010 distribution of fire refugia for these forest ecosystems and evaluate their topographic drivers. North‐eastern Siberia across the North‐east Siberian Taiga and the Cherskii‐Kolyma Mountain Tundra ecozones. 2001–2020. Cajander larch. We used Landsat imagery to define burned and unburned patches, and the Arctic digital elevation model to calculate topographic variables. We characterized the size and density of fire refugia. We sampled individual pixels (n = 80,000) from an image stack that included a binary burned/unburned, elevation, slope, aspect, topographic position index, ruggedness, and tree cover from 2001 to 2020. We evaluated the topographic drivers of fire refugia with boosted regression trees. We found no substantial difference in fire refugia size and density across the region. The fire refugia size averaged 7.2 ha (0.09–150,439 ha). The majority of interior burned patches exceed the potential wind dispersal distance from fire refugia. Topographic position index and terrain steepness were important predictors of fire refugia. Unprecedented wildfires in 2020 did not impact fire refugia formation. Fire refugia are strongly controlled by topographic positions such as uplands and lowlands that influence microsite hydrological conditions. Fire refugia contribute to postfire landscape heterogeneity that preserves ecosystem functions, seed sources, habitat, and carbon sinks.
... The thresholding method mainly determines whether the pixel is a forest or not by setting the threshold value, and the time of exceeding the threshold value is treated as the change time [13]. The segmentation methods divide the long time-series remote sensing data into several segments and determine the location of abrupt changes according to the mean square error and other statistics [14,15]. These methods must determine the number of segments in advance. ...
Article
Full-text available
Examining the characteristics of vegetation change and associated spatial patterns under different protection levels can provide a scientific basis for national park protection and management. Based on the dense time-series Landsat enhanced vegetation index (EVI) data between 1986 and 2020, we utilized the Wild Binary Segmentation (WBS) approach to detect spatial and temporal characteristics of abrupt, gradual, and total changes in Wuyishan National Park. The differences in vegetation change in three protection-level areas (strictly protected [Prots], generally protected [Prot], and non-protected [NP]) were examined, and the contributions to their spatial patterns were evaluated through Geodetector. The results showed the following: (1) The highest percentage of area without abrupt change was in Prots (39.89%), and the lowest percentage was in NP (17.44%). The percentage of abrupt change frequency (larger than three times) increased from 4.40% to 9.10% and 12.49% with the decreases in protection. The significance test showed that the difference in changed frequencies was not significant among these regions, but the interannual variation of abrupt change in Prots was significantly different from other areas. (2) The vegetation coverage of the Wuyishan National Park generally improved. The total EVI change (TEVI) showed that the positive percentage of Prots and Prot was 90.43% and 91.71%, respectively, slightly higher than that of NP (88.44%). However, the mean greenness change of NP was higher than that of Prots and Prot. (3) The park’s EVI spatial pattern in 1986 was the strongest factor determining the EVI spatial pattern in 2020; the explanatory power reduced as the protection level decreased. The explanation power (q value) of abrupt vegetation change was lower and increased as the protection level decreased. The interaction detection showed that EVI1986 and TEVI had the strongest explanatory powers, but the explanatory ability gradually weakened from 0.713 to 0.672 to 0.581 in Prots, Prot, and NP, respectively. This study provided a systematic analysis of vegetation changes and their impacts on spatial patterns.
Article
Fine resolution land cover products are becoming increasingly more available for many regions. These products, however, may not meet the quality requirements of many applications. This study provides an approach for improving land cover mapping by leveraging existing products and clear view Landsat composites. Assessments using independent reference datasets revealed that the CAF-LC30 2020 product derived using this approach over China was more accurate than four existing land cover products. Its overall accuracy with field observations was 2.94% to 10.28% higher than those of the four existing land cover products in northeast China and was 2.10% to 8.18% better across China. It provided a more accurate representation of the land cover types in many regions where the existing land cover products had large classification errors. Forest areas calculated using the CAF-LC30 2020 for the 31 provinces and autonomous regions and municipalities (PARM) in mainland China were better correlated with those reported by the most recent National Forest Inventory (NFI) survey than areas calculated using the other four existing land cover products. Therefore, the CAF-LC30 2020 product should be a better alternative for understanding China’s forests in 2020 than the other four existing land cover products.
Article
Wetlands in drylands are vulnerable to degradation and disappearance due to the combined effects of increasing anthropogenic disturbances and climatic extremes. Such influences may drive non-linear shifts in surface responses that require long-term monitoring approaches for their study. Here, we used a piece-wise regression model to characterize long-term Ecosystem Change Types (ECT) in the surface water and vegetation dynamics of the Inner Niger Delta wetlands in Mali between 2000 and 2019. We also examined the added benefits of using a dense Landsat time series for such segmented trend analysis in comparison with MODIS products that are regularly used for ecosystem trends assessment. Our approach has found statistically significant (p < 0.05) long-term changes in wetland ecosystems, as calculated from Modified Normalized Difference Water Index (MNDWI) and Normalized Difference Vegetation Index (NDVI) image series on both the MODIS and the Landsat scales. The class-specific accuracies of the detected ECTs were evaluated through the validation of temporal trajectories based on the TimeSync logic at selected probability sample locations. Results showed higher user's, producer's, and overall accuracies (OA) when using a dense Landsat time series (OA = 0.89 ± 0.01), outperforming the MOD09A1 time series (OA = 0.37 ± 0.03). Our study provides a robust framework for long-term wetland monitoring that demonstrates the benefits of applying dense Landsat time-series imagery for accurate quantifications of linear and non-linear ecosystem responses in vast highly dynamic floodplain systems. Delivering such an improved assessment, in a spatial resolution that better resolves the characteristics of wetlands ecosystems, has the potential to support the information needs of global conservation and restoration efforts.
Article
Full-text available
Multispectral imagery provides unprecedented information on Earth system processes: however, data gaps due to clouds and shadows are a major limitation. Normalized-Difference Vegetation Index (NDVI) and Normalized-Difference Water Index (NDWI) are two spectral indexes employed for monitoring vegetation phenology, land-cover change and more. Synthetic Aperture Radar (SAR) with its cloud-penetrating abilities can fill data gaps using coincident imagery. In this study, we evaluated C-band Sentinel-1, L-band Uninhabited Aerial Vehicle SAR (UAVSAR) and texture for gap filling using efficient machine learning regression algorithms across three seasons. Multiple models were evaluated including Support Vector Machine, Random Forest, Gradient Boosted Trees and an ensemble of models. The Gap filling ability of SAR was evaluated with Sentinel-2 imagery from the same date, 3 days and 8 days later than both SAR sensors in September. Sentinel-1 and Sentinel-2 imagery from winter and spring seasons were also evaluated. Because SAR imagery contains noise, we compared two robust de-noising methods and evaluated performance against a refined lee speckle filter. Mean Absolute Error (MAE) rates of the cloud gap-filling model were assessed across different dataset combinations and land covers. The results indicated de-noised Sentinel-1 SAR and UAVSAR with GLCM texture provided the highest predictive abilities with random forest R2 = 0.91 (±0.014), MAE = 0.078 (±0.003) (NDWI) and R2 = 0.868 (±0.015), MAE = 0.094 (±0.003) (NDVI) during September. The highest errors were observed across bare ground and forest, while the lowest errors were on herbaceous and woody wetland. Results on January and June imagery without UAVSAR were less strong at R2 = 0.60 (±0.036), MAE = 0.211 (±0.005) (NDVI), R2 = 0.61 (±0.043), MAE = 0.209 (±0.005) (NDWI) for January and R2 = 0.72 (±0.018), MAE = 0.142 (±0.004) (NDVI), R2 = 0.77 (±0.022), MAE = 0.125 (±0.004) (NDWI) for June. Ultimately, the results suggest de-noised C-band SAR with texture metrics can accurately predict NDVI and NDWI for gap-filling clouds during most seasons. These shallow machine learning models are rapidly trained and applied faster than intensive deep learning or time series methods.
Article
Full-text available
Water color is an important parameter in water quality assessment. However, the existing water color investigations have mostly focused on the lakes with areas greater 1 km2. In order to improve the understanding of the color of water bodies in China, a cloud-free composite image of China for the summer of 2015 was generated using time-series of Landsat-8 imagery and the best-available-pixel (BAP) compositing algorithm. Then, the first Forel?Ule index (FUI) water color product with a resolution of 30 m was produced for China using the generated BAP composite and the Google Earth Engine computing platform. Finally, the first national-scale assessment of the FUI of natural lakes with an area > 0.01 km2 (N = 60026) was conducted based on the generated FUI product. The generated FUI product was shown to have a high degree of consistency with in-situ water surface reflectance-derived FUI (R2 = 0.90, P < 0.001). Also, it had a high degree of consistency with the in-situ Secchi depth (SD) (R2 = 0.90, P < 0.001) and Trophic Level Index (TLI) (R2 = 0.62, P < 0.001) datasets. In addition, we found that the most prevalent lake colors in China were yellow (about 49%) and green (about 41%). Besides, the proportion of small lakes (areas < 1 km2) found to be yellow was much larger than for large lakes (area ? 1 km2) (50% against 28%). Our results will provide important information that can be used for preserving and restoring inland water resources.
Article
Full-text available
Time series analysis has gained popularity in forest disturbance monitoring thanks to the availability of satellite and airborne remote sensing images and the development of different time series methods for change detection. Previous research has focused on time series data noise reduction, the magnitude of breakpoints, and accuracy assessment; however, few have looked in detail at how the trend and seasonal model components contribute to disturbance detection in different forest types. Here, we use Landsat time series images spanning 1994–2018 to map forest disturbance in a western Pacific area of Mexico, where both temperate and tropical dry forests have been subject to severe deforestation and forest degradation processes. Since these two forest types have distinct seasonal characteristics, we investigate how trend and seasonal model components, such as the goodness-of-fit (R2), magnitude of change, amplitude, and model length in a stable historical period, affect forest disturbance detection. We applied the Breaks For Additive Season and Trend Monitor (BFAST) algorithm and after accuracy assessment by stratified random sample points, and we obtained 68% and 86% of user accuracy and 75.6% and 86% of producer’s accuracy in disturbance detection, in tropical dry forests and temperate forests, respectively. We extracted the noncorrelated trend and seasonal model components R2, magnitude, amplitude, length of the stable historical period, and percentage of pixels with NA and tested their effects on disturbance detection employing forest-type specific logistic regression. Our results showed that, for all forests combined, the amplitude and stable historical period length contributed to disturbance detection. While for tropical dry forest alone, amplitude was the main predictor, and for the temperate forest alone, the stable historical period length contributed most to the prediction, although it was not statistically significant. These findings provide insights for improving the results of forest disturbance detection in different forest types.
Article
Knowledge about the spatial and temporal distribution of exposed soils is necessary for e.g., soil erosion mitigation. Earth Observation (EO) is a valuable data source for detecting exposed soils on a large scale. In the last couple of years, the multitemporal compositing technique has been used for the generation of so-called exposed soil composites that overcome the limitation of temporarily coverage of the soils with vegetation as it is occurring at agricultural sites. The selection of exposed soil pixels from the stack of multispectral images is mainly done using spectral reflectance indices such as NDVI, NBR2 and others calculated on a per-pixel basis. The definition of the thresholds that are applicable to large areas such as regions, countries or continents is still a challenge and requires a reliable and robust sampling data base. In this study, the Soil Composite Mapping Processor (SCMaP) is used to build exposed soil masks containing all pixels in a given time period showing at least once exposed soil. For this purpose, a modified vegetation index (PV) based on the NDVI is used to separate the soils from other land cover (LC) classes by two PV thresholds. The overall goal of this study is to derive and validate exposed soil masks from multi-year Landsat data stacks for Germany from 1984 to 2019. The first focus is set on the impact of a newly developed sampling approach of LC classes such as urban areas, deciduous forests and agricultural fields that are automatically derived from Corine Land Cover (CLC) data. The spectral-temporal behavior of these LC classes in PVmin/max index composites show larger variability of the PV values compared to a manual sampling for selective LC classes such as urban areas. It reveals that the threshold definition method previously developed by Rogge et al. (2018) is not robust enough and the percentile rule used to define the Tmax threshold had to be adapted from 0.995 to 0.900. On the other hand, the sampling data base has proven to be robust across time and region. The second focus of the paper is to validate all generated exposed soil masks covering Germany for seven time periods from 1984 to 2019. A linear correlation analysis was performed comparing the SCMaP data with surveys from the Federal Statistical Office (Destatis) and the CLC inventories. The comparison with both datasets showed high regression coefficients (R2 = 0.79 to 0.90) with small regional deviations for areas in the Northern part of Germany. Strong correlation was found for time periods based on a higher number of cloud free Landsat images such as from 2000 to 2009. This demonstrates the high potential of SCMaP’s to generate exposed soil masks based on an automated sampling and a robust threshold derivation. To contribute to soil erosion studies that need information about where and when soils are bare, accurate exposed soil masks in suitable time periods can be of great value.
Article
Full-text available
Forest is one of the most important surface coverage types. Monitoring its dynamics is of great significance in global ecological environment monitoring and global carbon circulation research. Forest monitoring based on Landsat time-series stacks is a research hotspot, and continuous change detection is a novel approach to real-time change detection. Here, we present an approach, continuous change detection and classification-spectral trajectory breakpoint recognition, running on Google Earth Engine (GEE) for monitoring forest disturbance and forest long-term trends. We used this approach to monitor forest disturbance and the change in forest cover rate from 1987 to 2020 in Nanning City, China. The high-resolution Google Earth images are collected for the validation of forest disturbance. The classification accuracy of forest, non-forest, and water maps by using the optima classification features was 95.16%. For disturbance detection, the accuracy of our map was 86.4%, significantly higher than 60% of the global forest change product. Our approach can successfully generate high-accuracy classification maps at any time and detect the forest disturbance time on a monthly scale, accurately capturing the thinning cycle of plantations, which earlier studies failed to estimate. All the research work is integrated into GEE to promote the use of the approach on a global scale.
Preprint
Full-text available
Monitoring the water clarity of lakes is essential for the sustainable development of human society. However, existing water clarity assessments in China have mostly focused on lakes with areas > 1 km2, and the monitoring periods were mainly in the 21st century. In order to improve the understanding of spatiotemporal variations in lake clarity across China, based on the Google Earth Engine cloud platform, a 30 m long-term LAke Water Secchi Depth (SD) dataset (LAWSD30) of China (1985–2020) was first developed using Landsat series imagery and a robust water-color-parameter-based SD model. The LAWSD30 dataset exhibited a good performance compared with concurrent in situ SD datasets, with an R2 of 0.86 and a root-mean-square error of 0.225 m. Then, based on our LAWSD30 dataset, long-term spatiotemporal variations in SD for lakes > 0.01 km2 (N = 40,973) across China were evaluated. The results show that the SD of lakes with areas ≤ 1 km2 exhibited a significant downward trend in the period 1985–2020, but the decline rate began to slow down and stabilized after 2001. In addition, the SD of lakes with an area > 1 km2 showed a significant downward trend before 2001, and began to increase significantly afterwards. Moreover, in terms of the spatial patterns, the proportion of small lakes (area ≤ 1 km2) showing a decreasing SD trend was the largest in the Mongolian–Xinjiang Plateau Region (MXR) (about 30.0 %), and the smallest in the Eastern Plain Region (EPR) (2.6 %). In contrast, for lakes > 1 km2, this proportion was the highest in MXR (about 23.0 %), and the lowest in the Northeast Mountain Plain Region (NER) (16.1 %). The LAWSD30 dataset and the spatiotemporal patterns of lake water clarity in our research can provide effective guidance for the protection and management of lake environment in China.
Presentation
Full-text available
Combining infrequent and/or spatially limited lidar based measures of forest structural attributes with frequent and spatially exhaustive optical satellite imagery offers synergies for the mapping of forest attributes over large areas. Airborne lidar data has been demonstrated as a viable source of information relating forest structural attributes (such as height, canopy cover, biomass, and volume) to enable it’s use as reference data for the mapping of forest attributes over large areas in conjunction with optical satellite imagery and model informative spatial data layers. Given the high quality geolocation as well as the accuracy and diversity of forest structural attribution possible with airborne laser scanning systems (ALS), it is an ideal data source for model training and independent accuracy assessment. Purpose collected transects or opportunistically located wall-to-wall ALS data sets have been demonstrated as reference data options. There are a number of cases when ALS data may not be available with the desired traits, including spatial distribution over study region, capture of range of structural conditions present, etc. In such cases, spaceborne lidar datasets may provide new opportunities for systematic gathering of reference data. In this keynote, background on the use of lidar data as reference data will be shared, followed by example implementations representing large areas and long periods of time. Recent experiences in using spaceborne lidar data as reference data will then be offered, followed by sharing of issues and opportunities for further, and improved, use of this novel data capture opportunity.
Article
Full-text available
Aim Forest ecosystems around the globe are facing increasing natural and human disturbances. Increasing disturbances can challenge forest resilience, that is, the capacity of forests to sustain their functions and services in the face of disturbance. Quantifying resilience across large spatial extents remains challenging, as it requires the assessment of the ability of forests to recover from disturbance. Here we analysed the resilience of Europe’s forests by means of satellite-based recovery and disturbance indicators. Location Continental Europe (35 countries). Time period 1986–2018. Major taxa studied Gymnosperm and angiosperm woody plant species. Methods We used a comprehensive set of manually interpreted reference plots and random forest regression to model annual canopy cover from remote sensing data across more than 30 million disturbance patches in Europe over the time period 1986–2018. From annual time series of canopy cover, we estimated the time it takes disturbed areas to recover to pre-disturbance canopy cover levels using space-for-time substitution. We quantified forest resilience as the ratio between canopy disturbance and recovery intervals, with critical resilience defined as forest areas where canopy disturbances occurred faster than canopy recovery. Results On average across Europe, forests recover to pre-disturbance canopy cover within 30 years. The resilience of Europe’s forests to disturbance is high, with recovery being > 10 times faster than disturbance on 69% of the forest area. However, 14% of Europe’s forests had low or critical resilience, with disturbances occurring as fast or faster than forest canopy can recover. Main conclusions We conclude that Europe’s forests are widely resilient to past disturbance regimes, yet changing climate and disturbance regimes could erode resilience.
Article
Full-text available
The success and rate of forest regeneration has consequences for sustainable forest management, climate change mitigation, and biodiversity, among others. Systematically monitoring forest regeneration over large and often remote areas is challenging. Remotely sensed data and associated analytical approaches have demonstrated consistent and transparent options for spatially-explicit characterization of vegetation return following disturbance. Moreover, time series of satellite imagery enable the establishment of spatially meaningful recovery baselines that can provide a benchmark for identifying areas that are either under- or over-performing relative to those baselines. This information allows for the investigation and/or prioritization of areas requiring some form of management intervention, including guiding tree planting initiatives. In this research, we assess recovery following stand replacing disturbances for the 650 Mha forested ecosystems of Canada for the period 1985–2017, wherein ~51 Mha of Canada's forested ecosystems were impacted by wildfire, and ~ 21 Mha were impacted by harvesting. For quantification of forest recovery, we implement the Years to Recovery or Y2R metric using Landsat time series data based on the Normalized Burn Ratio (NBR) to relate the number of years required for a pixel to return to 80% of its pre-disturbance NBR value. By the end of the analyzed period, 76% of areas impacted by wildfire were considered spectrally recovered compared to 93% of harvested areas. On average, we found that harvest areas had more rapid spectral recovery (mean Y2R = 6.1 years) than wildfire (mean Y2R = 10.6 years) and importantly, that Y2R varied by ecozone, disturbance type, pre-disturbance land cover, and latitude. We used airborne laser scanning data to assess whether pixels that were considered spectrally recovered had attained United Nations Food and Agricultural Organization benchmarks of canopy height (>5 m) and cover (>10%) across four geographic regions representing different forest types. Overall, 87% and 97% of recovered pixels sampled in harvests and wildfires, respectively, had achieved at least one of the benchmarks, with benchmarks of height more readily achieved than benchmarks of cover. By analyzing spatial patterns of Y2R, we identified areas that had significant positive or negative spatial clustering in their rate of spectral recovery. Approximately 3.5–4% of areas disturbed by wildfire or harvest had significant positive spatial clustering, indicative of slower spectral recovery rates; these areas were also less likely to have attained benchmarks of height and cover. Conversely, we identified significant negative spatial clustering for 0.94% of areas recovering from harvest and 1.93% of areas recovering from wildfire, indicative of spectral recovery that was more rapid than the ecozonal baseline. Herein, we demonstrated that remote sensing can provide spatial intelligence on the nature of disturbance-recovery dynamics in forested ecosystems over large areas and moreover, can retrospectively quantify and characterize historic forest recovery trends within the past three decades that have implications for forest management, climate change mitigation, and restoration initiatives in the near term.
Article
Knowledge of tree species is required to inform management, planning, and monitoring of forests as well as to characterize habitat and ecosystem function. Remotely sensed data and spatial modeling enable mapping of tree species presence and distribution. Following an assessment of tree species identified in the sample-based National Forest Inventory (NFI), we mapped 37 tree species over the 650-Mha, forest-dominated ecosystems of Canada representing 2019 conditions. Landsat imagery and related spectral indices, geographic and climate data, elevation derivatives, and remote sensing-derived phenology are used as predictor variables trained with calibration samples from Canada's NFI using the Random Forests machine learning algorithm. Based upon prior knowledge of tree species distributions, classification models were implemented on a regional basis, meaning only the tree species that are expected in a given mapping region were modeled using local calibration samples. Modeling resulted in class membership probabilities values for each regionally eligible tree species for all treed pixels as well as an indicator of attribution confidence derived from the distance in feature space between the two leading classes. Accuracy assessment was conducted using independent validation data also drawn from the NFI following the same selection rules and indicated an overall accuracy of 93.1% ± 0.1% (95%-confidence interval). Predictor variables informing on geographic, climatic and topographic conditions had the largest importance on the classification models. Nationally, the most common leading tree species were black spruce (Picea mariana; 203 Mha or 57.3% of the treed area), trembling aspen (Populus tremuloides; 34.7 Mha, 9.8%), and lodgepole pine (Pinus contorta; 21.1 Mha, 5.9%). Regionally, there was ecozone-level dominance of other tree species, including subalpine fir (Abies lasiocarpa; Montane Cordillera), western hemlock (Tsuga heterophylla; Pacific Maritime), and balsam fir (Abies balsamea; Atlantic Maritime). Based upon the per-pixel class membership probabilities, species assemblages akin to those in forest inventories can also be produced. Further, given the calibrated reflectance of Landsat imagery, the methods presented herein can be implemented over a time series of images. The approach uses open data as predictor variables, making the method portable to other areas given availability of tree species training data.
Article
Full-text available
As part of the Global Monitoring for Environment and Security (GMES) Space Component programme, ESA is undertaking the development of an operational optical high-resolution Earth observation mission. Sentinel-2 is a system of two polar-orbiting satellites that will contribute to the continuity and improvement of the SPOT and Landsat series of multispectral missions, and ensure the delivery of high-quality data and applications for operational land monitoring, emergency response and security services.
Article
Full-text available
Free and open access to the more than 40 years of data captured in the Landsat archive, combined with improvements in standardized image products and increasing computer processing and storage capabilities, have enabled the production of large-area, cloud-free, surface reflectance pixel-based image composites. Best-available-pixel (BAP) composites represent a new paradigm in remote sensing that is no longer reliant on scene-based analysis. A time series of these BAP image composites affords novel opportunities to generate information products characterizing land cover, land cover change, and forest structural attributes in a manner that is dynamic, transparent, systematic, repeatable, and spatially exhaustive. Herein, we articulate the information needs associated with forest ecosystem science and monitoring in a Canadian context, and indicate how these new image compositing approaches and subsequent derived products can enable us to address these needs. We highlight some of the issues and opportunities associated with an image compositing approach and demonstrate annual composite products at a national-scale for a single year, with more detailed analyses for two prototype areas using 15 years of Landsat data. Recommendations concerning how to best link compositing decisions to the desired use of the composite (and the information need) are presented, along with future research directions.
Article
Full-text available
We developed a new algorithm called Tmask (multiTemporal mask) for automated masking of cloud, cloud shadow, and snow for multitemporal Landsat images. This algorithm consists of two steps. The first step is based on a single-date algorithm called Fmask (Function of mask) that initially screens most of the clouds, cloud shadows, and snow. The second step benefits from the extra temporal information from the remaining “clear” pixels and further improves the cloud, cloud shadow, and snow mask. Three Top Of Atmosphere (TOA) reflectance bands (Bands 2, 4, and 5 — Landsat-7 band numbering) are used in a Robust Iteratively Reweighted Least Squares (RIRLS) method to estimate a time series model for each pixel. By comparing model estimates with Landsat observations for the three spectral bands, the Tmask algorithm is capable of detecting any remaining clouds, cloud shadows, and snow for an entire stack of Landsat images. Generally, this algorithm will not falsely identify land cover changes as clouds, cloud shadows, or snow, as it is capable of modeling land cover change. The multitemporal images also provide extra information for better discrimination of cloud and snow, which is difficult for single-date algorithm. A snow threshold is derived for Band 5 TOA reflectance for each pixel at each specific time based on a modified Norwegian Linear Reflectance-to-Snow-Cover (NLR) algorithm. By comparing the results of Tmask with a single-date algorithm (Fmask) for multitemporal Landsat images located at Path 12 Row 31, significant improvements are observed for identification of clouds, cloud shadows, and snow. The most significant improvement is observed for cloud shadow detection. Many of the errors in cloud, cloud shadow, and snow detection observed in Fmask are corrected by the Tmask algorithm. The goal is development of a cloud, cloud shadow, and snow detection algorithm that results in a multitemporal stack of images that is free of “noise” factors and thus suitable for detection of land cover change.
Article
Full-text available
Multi-temporal satellite imagery can be composited over a season (or other time period) to produce imagery which is representative of that period, using techniques which will reduce contamination by cloud and other problems. For the purposes of vegetation monitoring, a commonly used technique is the Maximum NDVI Composite, used in conjunction with variety of other constraints. The current paper proposes an alternative based on the medoid (in reflectance space) over the time period (the medoid is a multi-dimensional analogue of the median), which is robust against extreme values, and appears to be better at producing imagery which is representative of the time period. For each pixel, the medoid is always selected from the available dates, so the result is always a single observation for that pixel, thus preserving relationships between bands. The method is applied to Landsat TM/ETM+ imagery to create seasonal reflectance images (four per year), with the aim being a regular time series of reflectance values which captures the variability at seasonal time scales. Analysis of the seasonal reflectance values suggests that resulting temporal image composites are more representative of the time series than the maximum NDVI seasonal composite.
Article
Full-text available
With the advent of free Landsat data stretching back decades, there has been a surge of interest in utilizing remotely sensed data in multitemporal analysis for estimation of biophysical parameters. Such analysis is confounded by cloud cover and other image-specific problems, which result in missing data at various aperiodic times of the year. While there is a wealth of information contained in remotely sensed time series, the analysis of such time series is severely limited due to the missing data. This paper illustrates a technique which can greatly expand the possibilities of such analyses, a Fourier regression algorithm, here on time series of normalized difference vegetation indices (NDVIs) for Landsat pixels with 30-m resolution. It compares the results with those using the spatial and temporal adaptive reflectance fusion model (STAR-FM), a popular approach that depends on having MODIS pixels with resolutions of 250 m or coarser. STAR-FM uses changes in the MODIS pixels as a template for predicting changes in the Landsat pixels. Fourier regression had an R2 of at least 90% over three quarters of all pixels, and it had the highest RPredicted2 values (compared to STAR-FM) on two thirds of the pixels. The typical root-mean-square error for Fourier regression fitting was about 0.05 for NDVI, ranging from 0 to 1. This indicates that Fourier regression may be used to interpolate missing data for multitemporal analysis at the Landsat scale, especially for annual or longer studies.
Article
Full-text available
The Landsat data archive represents more than 40 years of Earth observation, providing a valuable information source for monitoring ecosystem dynamics. In excess of 605 000 images of Canada have been acquired by the Landsat program since 1972. Herein we report several spatial and temporal characteristics of the Landsat observation record for Canada (1972_2012), including image availability by year, growing season, sensor, ecozone, and provincial or territorial jurisdiction. In contrast to the global Landsat archive, which is dominated by Enhanced Thematic Mapper Plus (ETM_) data, the majority of archived Landsat images of Canada were acquired by the Thematic Mapper (TM) sensor (57%). Approximately 55% of archived Landsat images were acquired within ± 30 days of 1 August, and 74% of Worldwide Reference System_2 path_row locations in Canada have more than 200 images acquired between 1 June and 30 September. Issues such as cloud cover and the availability of imagery to support pixel-based image compositing and time series analyses are explored and documented. For a pixel-based image compositing scenario whereby images (TM and ETM_) acquired after 1981 with less than 70% cloud cover and a target date of 1 August 930 days are considered, 60% of the path_row locations have five or fewer years of missing data in the archive. For time series analyses (i.e., ecosystem monitoring scenario) with the same temporal constraint but with less than 10% cloud cover, only 2% of path_row locations are missing five or fewer years of data, with a mean and median of 17 years of missing data. However, if a broader temporal window (1 June to 30 September) is considered for this scenario, 18% of path_row locations have five or fewer years of missing data. Free and open-access to the Landsat data archive, combined with the continuity of new data collections provided by the recently launched Landsat 8 satellite, offer many opportunities for scientific inquiry concerning the status and trends of Canada’s terrestrial ecosystems.
Article
Full-text available
Information on the changing land surface is required at high spatial resolutions as many processes cannot be resolved using coarse resolution data. Deriving such information over large areas for Landsat data, however, still faces numerous challenges. Image compositing offers great potential to circumvent such shortcomings. We here present a compositing algorithm that facilitates creating cloud free, seasonally and radiometrically consistent datasets from the Landsat archive. A parametric weighting scheme allows for flexibly utilizing different pixel characteristics for optimized compositing. We describe in detail the development of three parameter decision functions: acquisition year, day of year and distance to clouds. Our test site covers 42 Landsat footprints in Eastern Europe and we produced three annual composites. We evaluated seasonal and annual consistency and compared our composites to BRDF normalized MODIS reflectance products. Finally, we also evaluated how well the composites work for land cover mapping. Results prove that our algorithm allows for creating seasonally consistent large area composites. Radiometric correspondence to MODIS was high (up to R2 > 0.8), but varied with land cover configuration and selected image acquisition dates. Land cover mapping yielded promising results (overall accuracy 72%). Class delineations were regionally consistent with minimal effort for training data. Class specific accuracies increased considerably (~10%) when spectral metrics were incorporated. Our study highlights the value of compositing in general and for Landsat data in particular, allowing for regional to global LULCC mapping at high spatial resolutions.
Article
Full-text available
Finding an overall linear trend is a common method in scientific studies. It is almost a requirement when one intends to study variability. Nevertheless, when dealing with long climate temporal series, fitting a straight line only seldom has a relevant meaning. This paper proposes and describes a new methodology for finding overall trends, and, simultaneously, for computing a new set of climate parameters: the breakpoints between periods with significantly different trends. The proposed methodology uses a least-squares approach to compute the best continuous set of straight lines that fit a given time series, subject to a number of constraints on the minimum distance between breakpoints and on the minimum trend change at each breakpoint. The method is tested with three climate time series.
Article
Full-text available
A new method called Fmask (Function of mask) for cloud and cloud shadow detection in Landsat imagery is provided. Landsat Top of Atmosphere (TOA) reflectance and Brightness Temperature (BT) are used as inputs. Fmask first uses rules based on cloud physical properties to separate Potential Cloud Pixels (PCPs) and clear-sky pixels. Next, a normalized temperature probability, spectral variability probability, and brightness probability are combined to produce a probability mask for clouds over land and water separately. Then, the PCPs and the cloud probability mask are used together to derive the potential cloud layer. The darkening effect of the cloud shadows in the Near Infrared (NIR) Band is used to generate a potential shadow layer by applying the flood-fill transformation. Subsequently, 3D cloud objects are determined via segmentation of the potential cloud layer and assumption of a constant temperature lapse rate within each cloud object. The view angle of the satellite sensor and the illuminating angle are used to predict possible cloud shadow locations and select the one that has the maximum similarity with the potential cloud shadow mask. If the scene has snow, a snow mask is also produced. For a globally distributed set of reference data, the average Fmask overall cloud accuracy is as high as 96.4%. The goal is development of a cloud and cloud shadow detection algorithm suitable for routine usage with Landsat images.
Chapter
Full-text available
Landscape Assessment primarily addresses the need to identify and quantify fire effects over large areas, at times involving many burns. In contrast to individual case studies, the ability to compare results is emphasized along with the capacity to aggregate information across broad regions and over time. Results show the spatial heterogeneity of burns and how fire interacts with vegetation and topography. The quantity measured and mapped is "burn severity," defined here as a scaled index gauging the magnitude of ecological change caused by fire. In the process, two methodologies are integrated. Burn Remote Sensing (BR) involves remote sensing with Landsat 30-meter data and a derived radiometric value called the Normalized Burn Ratio (NBR). The NBR is temporally differenced between pre- and postfire datasets to determine the extent and degree of change detected from burning (fig. LA-1). Two timeframes of acquisition identify effects soon after fire and during the next growing season for Initial and Extended Assessments, respectively. The latter includes vegetative recovery potential and delayed mortality. The Burn Index (BI) adds a complementary field sampling approach, called the Composite Burn Index (CBI). It entails a relatively large plot, independent severity ratings for individual strata, and a synoptic rating for the whole plot area. Plot sampling may be used to.
Article
Full-text available
The northern high latitudes have warmed by about 0.8°C since the early 1970s, but not all areas have warmed uniformly [Hansen et al., 1999]. There is warming in most of Eurasia, but the warming rate in the United States is smaller than in most of the world, and a slight cooling is observed in the eastern United States over the past 50 years. These changes beg the question, can we detect the biotic response to temperature changes? Here we present results from analyses of a recently developed satellite-sensed normalized difference vegetation index (NDVI) data set for the period July 1981 to December 1999: (1) About 61% of the total vegetated area between 40°N and 70°N in Eurasia shows a persistent increase in growing season NDVI over a broad contiguous swath of land from central Europe through Siberia to the Aldan plateau, where almost 58% (7.3×106km2) is forests and woodlands; North America, in comparison, shows a fragmented pattern of change in smaller areas notable only in the forests of the southeast and grasslands of the upper Midwest, (2) A larger increase in growing season NDVI magnitude (12% versus 8%) and a longer active growing season (18 versus 12 days) brought about by an early spring and delayed autumn are observed in Eurasia relative to North America, (3) NDVI decreases are observed in parts of Alaska, boreal Canada, and northeastern Asia, possibly due to temperature-induced drought as these regions experienced pronounced warming without a concurrent increase in rainfall [Barber et al., 2000]. We argue that these changes in NDVI reflect changes in biological activity. Statistical analyses indicate that there is a statistically meaningful relation between changes in NDVI and land surface temperature for vegetated areas between 40°N and 70°N. That is, the temporal changes and continental differences in NDVI are consistent with ground-based measurements of temperature, an important determinant of biological activity. Together, these results suggest a photosynthetically vigorous Eurasia relative to North America during the past 2 decades, possibly driven by temperature and precipitation patterns. Our results are in broad agreement with a recent comparative analysis of 1980s and 1990s boreal and temperate forest inventory data [United Nations, 2000].
Article
Full-text available
Forest characterization with light detection and ranging (LiDAR) data has recently garnered much scientific and operational attention. The number of forest inventory attributes that may be directly measured with LiDAR is limited; however, when considered within the context of all the measured and derived attributes required to complete a forest inventory, LiDAR can be a valuable tool in the inventory process. In this paper, we present the status of LiDAR remote sensing of forests, including issues related to instrumentation, data collection, data processing, costs, and attribute estimation. The information needs of sustainable forest management provide the context within which we consider future opportunities for LiDAR and automated data processing.
Article
Full-text available
Vegetation Continuous Field (VCF) layers of 30 m percent tree cover, bare ground, other vegetation and probability of water were derived for the conterminous United States (CONUS) using Landsat 7 Enhanced Thematic Mapper Plus (ETM+) data sets from the Web-Enabled Landsat Data (WELD) project. Turnkey approaches to land cover characterization were enabled due to the systematic WELD Landsat processing, including conversion of digital numbers to calibrated top of atmosphere reflectance and brightness temperature, cloud masking, reprojection into a continental map projection and temporal compositing. Annual, seasonal and monthly WELD composites for 2008 were used as spectral inputs to a bagged regression and classification tree procedure using a large training data set derived from very high spatial resolution imagery and available ancillary data. The results illustrate the ability to perform Landsat land cover characterizations at continental scales that are internally consistent while retaining local spatial and thematic detail.
Article
Full-text available
We present a method for a supervised classification of Normalized Difference Vegetation Index (NDVI) time series that identifies vegetation type and vegetation coverage, absolute in %coverage or relative to a reference NDVI cycle. The shape of the NDVI cycle, which is diagnostic for certain vegetation types, is our primary classifier. A Discrete Fourier Filter is applied to time series data in order to minimize the influence of high-frequency noise on class assignments. Similarity between filtered NDVI cycles is evaluated using a linear regression technique. The correlation coefficients calculated between the Fourier filtered reference cycle and likewise filtered target cycles describe the similarity of their phenology, and the corresponding regression coefficients are an expression of coverage relative to the reference. The regression coefficients are correlated with field measured vegetation coverage. The Fourier Filtered Cycle Similarity method (FFCS) compensates phenological shifts, which are typical in areas with a strong climate gradient, and prevents the break-up of classes of identical vegetation types on the basis of vegetation coverage. Some other advantages compared to traditional unsupervised classifications are: synoptic visualization of vegetation type and coverage variation, independence from scene statistics, and consistent classification of biophysical characteristics only, without rock/soil reflectance dominating class assignment as it often does in unsupervised classifications of sparsely vegetated areas. Using the FFCS classification we differentiated a total of five rangeland vegetation types for the area of Syria including their intra-class coverage variation. Classified classes are dominated by one of two shrub types, one of two annual grass types or a bare soil/sparsely vegetated type.
Article
Full-text available
Managing forests for sustainable use requires that both the biological diversity of the forests and a viable forest industry be maintained. A current approach towards maintaining biological diversity is to pattern forest management practices after those of natural disturbance events. This paradigm hypothesizes that ecological processes will be maintained best where active management approximates natural disturbance events. The forest management model now used in most sub-boreal and boreal forests calls for regularly dispersed clearcuts no greater than 60–100 ha in size. However, the spatial characteristics of the landscape produced by this model are distinctly different from the historic pattern generated by wildfire, which was heretofore the dominant stand-replacing process in these forests. Wildfire creates a more complex landscape spatial pattern with greater range in patch size and more irregular disturbance boundaries. Individual wildfires are often over 500 ha but leave patches of unburned forest within them. The combination of these attributes is not present in recent clearcuts. Allowing a proportion of larger (i.e.>500ha) harvest units may provide distinct economic advantages that could outweight the opportunity costs of leaving some patches of forest behind. For the forest type examined, further evaluation of modelling forest harvest patterns more closely after the patterns created by wildfire is required as it may achieve a good balance and strike a suitable compromise between certain ecological and economic objectives of sustainable development.