No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
A novel linear spectral unmixing-based method for tree decline monitoring by fusing UAV-RGB and optical space-borne data
International Journal of Remote Sensing
Abstract
Remote sensing-assisted monitoring of forest health entails methods that can provide up-to-date and accurate information on decline and mortality of individual trees, while maintaining time and cost efficiency. However, the trade-off of applying consumer-grade UAV-RGB data as the most affordable and accessible data source at the catchment level is constrained by its poor spectral information content. We developed a method based on the fusion of UAV-RGB data with space-borne Sentinel-2 Multispectral Instrument (MSI) at the level of tree crowns, with the specific target of supporting studies on semi-arid tree decline. We applied linear spectral unmixing (Spectral Unmixing-Based data Fusion method, LSUBF) by considering a limited number of endmem-bers and calculating the abundances (fractional covers) from the UAV data, and evaluated the results by high-resolution MSI space-borne data including SPOT-6 (1.5 m spatial resolution) and PlanetScope (3 m spatial resolution). This method suggested an increase in the coefficient of determination of the applied generalized additive model for decline severity estimation at tree crown level from 0.61 to 0.69, while it was improved from 0.70 to 0.91 when fitting a non-parametric random forest model. The results of sensitivity analysis demonstrated that the additional spectral information obtained from the proposed method results in higher accuracy in estimating decline severity. We suggest this method as a cost-effective alternative to monitor periodical tree decline, in particular across semi-arid ecosystems.
Supplementary resources (3)
Data
February 2024
Data
February 2024
Data
February 2024
ResearchGate has not been able to resolve any citations for this publication.
Tree decline is a highly complex process and is inherently a function of manifold climatic, physiologic, and anthropogenic factors. Monitoring decline processes and their underlying dynamics primarily entails identifying their location and intensity across different ecosystems, for which airborne and satellite remote sensing approaches offer cost-effective and spatially explicit alternatives to field methods. Consumer-grade unmanned aerial vehicles (UAVs) can barely be used as standalone means for large-area monitoring due to their constrains in spatial and spectral domains. However, they could effectively be integrated alongside satellite data to unlock their information for subsequent upscaling on landscape level. We designed a novel two-step workflow to describe the severity of tree decline by linking UAV-RGB information to space-borne multispectral and digital elevation model (DEM) data over 15 forest sites dominated by Persian oak across the latitudinal gradient of Zagros Forests in western Iran. We display how to 1) leverage UAV as reference data across multiple structurally different Persian oak-dominated sites in semi-arid Zagros mountains of Iran; 2) link UAV, Copernicus DEM, and Sentinel-2 data to retrieve decline information within a model-driven context; and 3) analyze the sensitivity of models by means of a global variance-based sensitivity analysis. Results suggested a high association between UAV and field data on the intensity of decline, which enabled using sampled UAV data as reference to estimate the decline severity using space-borne data by means of semi-parametric generalized additive model (GAM) and non-parametric random forest (RF) approaches. Conclusively, this study provided a baseline for multi-scale analysis of tree decline using budget and partially free data sources, which can be of high scientific and practical assets for monitoring in remote, sparse, mountainous, and continuously degrading forest areas.
In this study, all the papers about oak decline phenomena in the forests of Iran have been reviewed. One hundred and twelve papers from 51 journals were reviewed, and the information presented in each paper was analyzed and finally summarized. Eighty-seven percent of the papers were about Brant’s oak (Quercus brantii Lindl.). Most of the studies focused on the pests and diseases causing oak decline. The relationship between decline and quantitative characteristics of trees has produced conflicting results. Regarding the relationship between decline and stand characteristics, certain results were observed in only two cases: the clump pattern of declined trees and the direct relationship between stand density and oak decline. Findings were contradictory in most cases regarding the relationship between physiographic characteristics and tree decline. One certain result about the relation between decline and soil properties was that tree decline in shallow soils was more pronounced than in deep soils. Concerning climatic research, a reduction in precipitation and an increase in air temperature, wind speed, solar radiation, evapotranspiration, and dust have been shown to increase the probability of decline or intensification of the phenomenon. Furthermore, various studies emphasized the contribution of local people to addressing oak decline and strengthening their economies. According to the literature reviews, using crescent-shaped barriers for regional preservation and the sanitary cutting of branches are more likely to mitigate the process of decline in trees. Overall, the published papers suggest two important features of the oak decline phenomenon: multidimensionality, complexity, and sporadic research. Additionally, it seems that over time, the areas affected by this phenomenon have increased, and the lack of essential research about this phenomenon is evident.
The unmixing-based spatiotemporal fusion model is one of the effective ways to solve limitations in temporal and spatial resolution tradeoffs in a single satellite sensor. By using fusion data from different satellite platforms, high resolution in both temporal and spatial domains can be produced. However, due to the ill-posed characteristic of the unmixing function, the model performance may vary due to the different model setups. The key factors affecting the model stability most and how to set up the unmixing strategy for data downscaling remain unknown. In this study, we use the multisource land surface temperature as the case and focus on the three major factors to analyze the stability of the unmixing-based fusion model: (1) the definition of the homogeneous change regions (HCRs), (2) the unmixing levels, and (3) the number of HCRs. The spatiotemporal data fusion model U-STFM was used as the baseline model. The results show: (1) The clustering-based algorithm is more suitable for detecting HCRs for unmixing. Compared with the multi-resolution segmentation algorithm and k-means algorithm, the ISODATA clustering algorithm can more accurately describe LST’s temporal and spatial changes on HCRs. (2) For the U-STFM model, applying the unmixing processing at the change ratio level can significantly reduce the additive and multiplicative noise of the prediction. (3) There is a tradeoff effect between the number of HCRs and the solvability of the linear unmixing function. The larger the number of HCRs (less than the available MODIS pixels), the more stable the model is. (4) For the fusion of the daily 30 m scale LST product, compared with STARFM and ESTARFM, the modified U-STFM (iso_USTFM) achieved higher prediction accuracy and a lower error (R2: 0.87 and RMSE:1.09 k). With the findings of this study, daily fine-scale LST products can be predicted based on the unmixing-based spatial–temporal model with lower uncertainty and stable prediction.
The unmanned aerial vehicles (UAVs) are widely used for agricultural monitoring due to reduce the cost and time of crop monitoring via the acquisition of images with high spatial-temporal resolution. The normalized difference vegetation index (NDVI) is the most widely studied and used for mapping crop growth. A relatively expensive multispectral sensor is required to produce an NDVI map. The visible vegetation indices (VIs) derived from UAV images showed potential capabilities for predicting crop growth. The purpose of this paper is to evaluate the RGB indices to monitor the growth of the rice crop. The images were obtained from the study area by DJI PM4 multispectral UAV. The multispectral images were used to calculate NDVI as a reference vegetation index and different RGB indices were implemented and compared with the reference index. The results showed that RGB indices can be used as the vegetation index in the case of unavailable multispectral images.
Monitoring tree decline in arid and semi-arid zones requires methods that can provide up-to-date and accurate information on the health status of the trees at single-tree and sample plot levels. Unmanned Aerial Vehicles (UAVs) are considered as cost-effective and efficient tools to study tree structure and health at small scale, on which detecting and delineating tree crowns is the first step to extracting varied subsequent information. However, one of the major challenges in broadleaved tree cover is still detecting and delineating tree crowns in images. The frequent dominance of coppice structure in degraded semi-arid vegetation exacerbates this problem. Here, we present a new method based on edge detection for delineating tree crowns based on the features of oak trees in semi-arid coppice structures. The decline severity in individual stands can be analyzed by extracting relevant information such as texture from the crown area. Although the method presented in this study is not fully automated, it returned high performances including an F-score = 0.91. Associating the texture indices calculated in the canopy area with the phenotypic decline index suggested higher correlations of the GLCM texture indices with tree decline at the tree level and hence a high potential to be used for subsequent remote-sensing-assisted tree decline studies.
Since remote sensing images offer unique access to the distribution of land cover on earth, many countries are investing in this technique to monitor urban sprawl. For this purpose, the most widely used methodology is spectral unmixing which, after identifying the spectra of the mixed-pixel constituents, determines their fractional abundances in the pixel. However, the literature highlights shortcomings in spatial validation due to the lack of detailed ground truth knowledge and proposes five key requirements for accurate reference fractional abundance maps: they should cover most of the area, their spatial resolution should be higher than that of the results, they should be validated using other ground truth data, the full range of abundances should be validated, and errors in co-localization and spatial resampling should be minimized. However, most proposed reference maps met two or three requirements and none met all five. In situ and remote data acquired in Venice were exploited to meet all five requirements. Moreover, to obtain more information about the validation procedure, not only reference spectra, synthetic image, and fractional abundance models (FAMs) that met all the requirements, but also other data, that no previous work exploited, were employed: reference fractional abundance maps that met four out of five requirements, and fractional abundance maps retrieved from the synthetic image. Briefly summarizing the main results obtained from MIVIS data, the average of spectral accuracies in root mean square error was equal to 0.025; using FAMs, the average of spatial accuracies in mean absolute error (MAEk-Totals) was equal to 1.32 and more than 78% of these values were related to sensor characteristics; using reference fractional abundance maps, the average MAEk-Totals value increased to 1.97 because errors in co-localization and spatial-resampling affected about 29% of these values. In conclusion, meeting all requirements and the exploitation of different reference data increase the spatial accuracy, upgrade the validation procedure, and improve the knowledge of accuracy.
Reasonable cultivation is an important part of the protection work of endangered species. The timely and nondestructive monitoring of chlorophyll can provide a basis for the accurate management and intelligent development of cultivation. The image analysis method has been applied in the nutrient estimation of many economic crops, but information on endangered tree species is seldom reported. Moreover, shade control, as the common seedling management measure, has a significant impact on chlorophyll, but shade levels are rarely discussed in chlorophyll estimation and are used as variables to improve model accuracy. In this study, 2-year-old seedlings of tropical and endangered Hopea hainanensis were taken as the research object, and the SPAD value was used to represent the relative chlorophyll content. Based on the performance comparison of RGB and multispectral (MS) images using different algorithms, a low-cost SPAD estimation method combined with a machine learning algorithm that is adaptable to different shade conditions was proposed. The SPAD values changed significantly at different shade levels (p < 0.01), and 50% shade in the orthographic direction was conducive to chlorophyll accumulation in seedling leaves. The coefficient of determination (R²), root mean square error (RMSE), and average absolute percent error (MAPE) were used as indicators, and the models with dummy variables or random effects of shade greatly improved the goodness of fit, allowing better adaption to monitoring under different shade conditions. Most of the RGB and MS vegetation indices (VIs) were significantly correlated with the SPAD values, but some VIs exhibited multicollinearity (variance inflation factor (VIF) > 10). Among RGB VIs, RGRI had the strongest correlation, but multiple VIs filtered by the Lasso algorithm had a stronger ability to interpret the SPAD data, and there was no multicollinearity (VIF < 10). A comparison of the use of multiple VIs to estimate SPAD indicated that Random forest (RF) had the highest fitting ability, followed by Support vector regression (SVR), linear mixed effect model (LMM), and ordinary least squares regression (OLR). In addition, the performance of MS VIs was superior to that of RGB VIs. The R² of the optimal model reached 0.9389 for the modeling samples and 0.8013 for the test samples. These findings reinforce the effectiveness of using VIs to estimate the SPAD value of H. hainanensis under different shade conditions based on machine learning and provide a reference for the selection of image data sources.
The mountain pine (Pinus mugo ssp. Mugo Turra) is an important component of the alpine treeline ecotone and fulfills numerous ecosystem functions. To understand and quantify the impacts of increasing logging activities and climatic changes in the European Alps, accurate information on the occurrence and distribution of mountain pine stands is needed. While Earth observation provides up-to-date information on land cover, space-borne mapping of mountain pines is challenging as different coniferous species are spectrally similar, and small-structured patches may remain undetected due to the sensor’s spatial resolution. This study uses multi-temporal optical imagery from PlanetScope (3 m) and Sentinel-2 (10 m) and combines them with additional features (e.g., textural statistics (homogeneity, contrast, entropy, spatial mean and spatial variance) from gray level co-occurrence matrix (GLCM), topographic features (elevation, slope and aspect) and canopy height information) to overcome the present challenges in mapping mountain pine stands. Specifically, we assessed the influence of spatial resolution and feature space composition including the GLCM window size for textural features. The study site is covering the Sarntal Alps, Italy, a region known for large stands of mountain pine. Our results show that mountain pines can be accurately mapped (PlanetScope (90.96%) and Sentinel-2 (90.65%)) by combining all features. In general, Sentinel-2 can achieve comparable results to PlanetScope independent of the feature set composition, despite the lower spatial resolution. In particular, the inclusion of textural features improved the accuracy by +8% (PlanetScope) and +3% (Sentinel-2), whereas accuracy improvements of topographic features and canopy height were low. The derived map of mountain pines in the Sarntal Alps supports local forest management to monitor and assess recent and ongoing anthropogenic and climatic changes at the treeline. Furthermore, our study highlights the importance of freely available Sentinel-2 data and image-derived textural features to accurately map mountain pines in Alpine environments.
In recent years, technological advances have led to the increasing use of unmanned aerial vehicles (UAVs) for forestry applications. One emerging field for drone application is forest health monitoring (FHM). Common approaches for FHM involve small-scale resource-extensive fieldwork combined with traditional remote sensing platforms. However, the highly dynamic nature of forests requires timely and repetitive data acquisition, often at very high spatial resolution, where conventional remote sensing techniques reach the limits of feasibility. UAVs have shown that they can meet the demands of flexible operation and high spatial resolution. This is also reflected in a rapidly growing number of publications using drones to study forest health. Only a few reviews exist which do not cover the whole research history of UAV-based FHM. Since a comprehensive review is becoming critical to identify research gaps, trends, and drawbacks, we offer a systematic analysis of 99 papers covering the last ten years of research related to UAV-based monitoring of forests threatened by biotic and abiotic stressors. Advances in drone technology are being rapidly adopted and put into practice, further improving the economical use of UAVs. Despite the many advantages of UAVs, such as their flexibility, relatively low costs, and the possibility to fly below cloud cover, we also identified some shortcomings: (1) multitemporal and long-term monitoring of forests is clearly underrepresented; (2) the rare use of hyperspectral and LiDAR sensors must drastically increase; (3) complementary data from other RS sources are not sufficiently being exploited; (4) a lack of standardized workflows poses a problem to ensure data uniformity; (5) complex machine learning algorithms and workflows obscure interpretability and hinders widespread adoption; (6) the data pipeline from acquisition to final analysis often relies on commercial software at the expense of open-source tools.
As an important application in remote sensing, landcover classification remains one of the most challenging tasks in very-high-resolution (VHR) image analysis. As the rapidly increasing number of Deep Learning (DL) based landcover methods and training strategies are claimed to be the state-of-the-art, the already fragmented technical landscape of landcover mapping methods has been further complicated. Although there exists a plethora of literature review work attempting to guide researchers in making an informed choice of landcover mapping methods, the articles either focus on the review of applications in a specific area or revolve around general deep learning models, which lack a systematic view of the ever advancing landcover mapping methods. In addition, issues related to training samples and model transferability have become more critical than ever in an era dominated by data-driven approaches, but these issues were addressed to a lesser extent in previous review articles regarding remote sensing classification. Therefore, in this paper, we present a systematic overview of existing methods by starting from learning methods and varying basic analysis units for landcover mapping tasks, to challenges and solutions on three aspects of scalability and transferability with a remote sensing classification focus including (1) sparsity and imbalance of data; (2) domain gaps across different geographical regions; and (3) multi-source and multi-view fusion. We discuss in detail each of these categorical methods and draw concluding remarks in these developments and recommend potential directions for the continued endeavor.
Detecting and mapping individual trees accurately and automatically from remote sensing images is of great significance for precision forest management. Many algorithms, including classical methods and deep learning techniques, have been developed and applied for tree crown detection from remote sensing images. However, few studies have evaluated the accuracy of different in-dividual tree detection (ITD) algorithms and their data and processing requirements. This study explored the accuracy of ITD using local maxima (LM) algorithm, marker-controlled watershed segmentation (MCWS), and Mask Region-based Convolutional Neural Networks (Mask R-CNN) in a young plantation forest with different test images. Manually delineated tree crowns from UAV imagery were used for accuracy assessment of the three methods, followed by an evaluation of the data processing and application requirements for three methods to detect individual trees. Overall, Mask R-CNN can best use the information in multi-band input images for detecting individual trees. The results showed that the Mask R-CNN model with the multi-band combination produced higher accuracy than the model with a single-band image, and the RGB band combination achieved the highest accuracy for ITD (F1 score = 94.68%). Moreover, the Mask R-CNN models with multi-band images are capable of providing higher accuracies for ITD than the LM and MCWS algorithms. The LM algorithm and MCWS algorithm also achieved promising accuracies for ITD when the canopy height model (CHM) was used as the test image (F1 score = 87.86% for LM algorithm, F1 score = 85.92% for MCWS algorithm). The LM and MCWS algorithms are easy to use and lower computer computational requirements, but they are unable to identify tree species and are limited by algorithm parameters, which need to be adjusted for each classification. It is highlighted that the application of deep learning with its end-to-end-learning approach is very efficient and capable of deriving the information from multi-layer images, but an additional training set is needed for model training, robust computer resources are required, and a large number of accurate training samples are necessary. This study provides valuable information for forestry practitioners to select an optimal approach for detecting individual trees.
In this work we develop a novel fully unsupervised autoencoder-based scheme for nonlinear hyperspectral pixel unmixing. A unique approach is derived where high noise and unresponsive pixels are accounted for, by a unique averaging approach based on spatially-aware filters built using radial basis function (RBF) kernels. A novel technique is implemented via calculating rank-equivalent kernel covariance matrices in order to estimate the unknown number of endmembers contributing to the data. Utilization of spatial information is done via RBF-based weighted averaging, which is then followed by endmember estimation via K-means clustering. The RBF distances from the cluster centers are determined to measure the position of the mixed pixels in relation to the centers, which is utilized as a preliminary estimation of the abundances. The proposed framework is robust in the presence of unresponsive pixels, while highly versatile working with different nonlinear unmixing models. Extensive numerical tests establish the superiority of the novel approach with respect to state-of-the-art methods.
Semi-arid tree covers, in both high and coppice growth forms, play an essential role in protecting water and soil resources and provides multiple ecosystem services across fragile ecosystems. Thus, they require continuous inventories. Quantification of forest structure in these tree covers provides important measures for their management and biodiversity conservation. We present a framework, based on consumer-grade UAV photogrammetry, to separately estimate primary variables of tree height (H) and crown area (A) across diverse coppice and high stands dominated by Quercus brantii Lindl. along the latitudinal gradient of Zagros mountains of western Iran. Then, mul-tivariate linear regressions were parametrized with H and A to estimate the diameter at breast height (DBH) of high trees because of its importance to accelerate the existing practical DBH inventories across Zagros Forests. The estimated variables were finally applied to a model tree above-ground biomass (AGB) for both vegetative growth forms by local allometric equations and Random Forest models. In each step, the estimated variables were evaluated against the field reference values , indicating practically high accuracies reaching root mean square error (RMSE) of 0.68 m and 4.74 cm for H and DBH, as well as relative RMSE < 10% for AGB estimates. The results generally suggest an effective framework for single tree-based attribute estimation over mountainous, semi-arid coppice, and high stands.
Many countries share an effort to understand the impact of growing urban areas on the environment. Spatial, spectral, and temporal resolutions of remote sensing images offer unique access to this information. Nevertheless, their use is limited because urban surface materials exhibit a great diversity of types and are not well spatially and spectrally distinguishable. This work aims to quantify the effect of these spatial and spectral characteristics of urban surface materials on their retrieval from images. To avoid other sources of error, synthetic images of the historical center of Venice were analyzed. A hyperspectral library, which characterizes the main materials of Venice city and knowledge of the city, allowed to create a starting image at a spatial resolution of 30 cm and spectral resolution of 3 nm and with a spectral range of 365–2500 nm, which was spatially and spectrally resampled to match the characteristics of most remote sensing sensors. Linear spectral mixture analysis was applied to every resampled image to evaluate and compare their capabilities to distinguish urban surface materials. In short, the capability depends mainly on spatial resolution, secondarily on spectral range and mixed pixel percentage, and lastly on spectral resolution; impervious surfaces are more distinguishable than pervious surfaces. This analysis of capability behavior is very important to select more suitable remote sensing images and/or to decide the complementarity use of different data.
Global biodiversity monitoring systems through remote sensing can support consistent assessment, monitoring, modelling and reporting on biodiversity which are key activities intended for sustainable management. This work presents an overview of biodiversity monitoring components, i.e. biodiversity levels, essential biodiversity variables, biodiversity indicators, scale, biodiversity inventory, biodiversity models, habitat, ecosystem services, vegetation health and biogeochemical heterogeneity and discusses what remote sensing through Earth Observations has contributed to the study of biodiversity. The technological advancements in remote sensing have enabled information-rich data on biodiversity. Remote sensing data are making a strong contribution in providing unique information relevant to various biodiversity research and conservation applications. The extensive use of Earth observation data are not yet realized in biodiversity assessment, monitoring and conservation. The development of direct remote sensing approaches and the techniques for quantifying biodiversity at the community to species level is likely to be a great challenge for comprehensive earth observation-based monitoring strategy.
Mapping crop types and land cover in smallholder farming systems in sub-Saharan Africa remains a challenge due to data costs, high cloud cover, and poor temporal resolution of satellite data. With improvement in satellite technology and image processing techniques, there is a potential for integrating data from sensors with different spectral characteristics and temporal resolutions to effectively map crop types and land cover. In our Malawi study area, it is common that there are no cloud-free images available for the entire crop growth season. The goal of this experiment is to produce detailed crop type and land cover maps in agricultural landscapes using the Sentinel-1 (S-1) radar data, Sentinel-2 (S-2) optical data, S-2 and PlanetScope data fusion, and S-1 C2 matrix and S-1 H/α polarimetric decomposition. We evaluated the ability to combine these data to map crop types and land cover in two smallholder farming locations. The random forest algorithm, trained with crop and land cover type data collected in the field, complemented with samples digitized from Google Earth Pro and DigitalGlobe, was used for the classification experiments. The results show that the S-2 and PlanetScope fused image + S-1 covariance (C2) matrix + H/α polarimetric decomposition (an entropy-based decomposition method) fusion outperformed all other image combinations, producing higher overall accuracies (OAs) (>85%) and Kappa coefficients (>0.80). These OAs represent a 13.53% and 11.7% improvement on the Sentinel-2-only (OAs < 80%) experiment for Thimalala and Edundu, respectively. The experiment also provided accurate insights into the distribution of crop and land cover types in the area. The findings suggest that in cloud-dense and resource-poor locations, fusing high temporal resolution radar data with available optical data presents an opportunity for operational mapping of crop types and land cover to support food security and environmental management decision-making.
Oak decline syndromes such as Acute Oak Decline (AOD) and Chronic Oak Decline (COD) are becoming increasingly prevalent and with this comes the need for more quantitative, sensitive and standardised visual oak health monitoring.
Phenotyping protocols were developed to specifically measure oak decline severity and were based on a comprehensive set of simple to measure phenotypic descriptors. A total of 36 phenotypic measurements describing oak decline status included aspects of tree size, crown condition, the presence of biotic agents and a number of derived composite descriptors.
Phenotypic measurements were collected from a total of 174 Quercus robur, surveyed from 9 sites across England and included healthy, AOD, COD and AOD trees in remission. Using these data, the Phenotypic Decline Index (PDI) and the Decline Acuteness Index (DAI) were developed to quantitatively describe and differentiate the acute and chronic oak decline severity spectrum. These decline indexes were derived from unsupervised random forest machine learning models, trained using the collected phenotypic data.
The suitability of the decline indexes for describing decline severity and type were assessed by comparing decline index scores to manual decline status classifications along with an assessment of descriptor importance and contribution to the decline index models. Crown condition and trees size descriptors such as ‘composite crown volume’ contributed positively to the PDI. Trees with smaller crowns in poor condition had greater PDI values. Tree stature and the presence of stem bleeding contributed highly to the DAI, allowing differentiation between trees with AOD and COD syndromes. AOD trees had relatively larger stature and the presence of stem bleeding where as COD trees had small stature and stem bleeding was absent.
The oak decline indexes are simple but sensitive measures of tree decline severity and allow easy comparisons of oak trees both spatially and temporally. These have the potential to provide useful tools for forest monitoring and management as well application to remote sensing and omics research.
The growing demand for high-quality imaging data and the current technological limitations of imaging sensors require the development of techniques that combine data from different platforms in order to obtain comprehensive products for detailed studies of the environment. To meet the needs of modern remote sensing, the authors present an innovative methodology of combining multispectral aerial and satellite imagery. The methodology is based on the simulation of a new spectral band with a high spatial resolution which, when used in the pansharpening process, yields an enhanced image with a higher spectral quality compared to the original panchromatic band. This is important because spectral quality determines the further processing of the image, including segmentation and classification. The article presents a methodology of simulating new high-spatial-resolution images taking into account the spectral characteristics of the photographed types of land cover. The article focuses on natural objects such as forests, meadows, or bare soils. Aerial panchromatic and multispectral images acquired with a digital mapping camera (DMC) II 230 and satellite multispectral images acquired with the S2A sensor of the Sentinel-2 satellite were used in the study. Cloudless data with a minimal time shift were obtained. Spectral quality analysis of the generated enhanced images was performed using a method known as “consistency” or “Wald’s protocol first property”. The resulting spectral quality values clearly indicate less spectral distortion of the images enhanced by the new methodology compared to using a traditional approach to the pansharpening process.
Time-series remote sensing data are important in monitoring land surface dynamics. Due to technical limitations, satellite sensors have a trade-off between temporal, spatial and spectral resolutions when acquiring remote sensing images. In order to obtain remote sensing images with high spatial resolution and high temporal frequency, spatiotemporal fusion methods have been developed. In this paper, we propose a Linear Spectral Unmixing-based Spatiotemporal Data Fusion Model (LSUSDFM) for spatial and temporal data fusion. In this model, the endmember abundance of the low-resolution image pixel is calculated based on that of the high-resolution image by the spectral mixture analysis. The endmember spectrum signals of low-resolution images are then calculated continuously within an optimized moving window. Subsequently, the fused image is reconstructed according to the endmember spectrum and its corresponding abundance map. A simulated dataset and real satellite images are used to test the fusion model, and the fusion results are compared with a current spectral unmixing based downscaling fusion model (SUDFM). Our experimental work shows that, compared to the SUDFM, the proposed LSUSDFM can achieve better quality and accuracy of fused images, especially in effectively eliminating the “plaque” phenomenon in the results by the SUDFM. The LSUSDFM has great potential in generating images with both high spatial resolution and high temporal frequency, as well as increasing the number of spectral bands of the high spatial resolution data.
Non-destructive crop monitoring over large areas with high efficiency is of great significance in precision agriculture and plant phenotyping, as well as decision making with regards to grain policy and food security. The goal of this research was to assess the potential of combining canopy spectral information with canopy structure features for crop monitoring using satellite/unmanned aerial vehicle (UAV) data fusion and machine learning. Worldview-2/3 satellite data were tasked synchronized with high-resolution RGB image collection using an inexpensive unmanned aerial vehicle (UAV) at a heterogeneous soybean (Glycine max (L.) Merr.) field. Canopy spectral information (i.e., vegetation indices) was extracted from Worldview-2/3 data, and canopy structure information (i.e., canopy height and canopy cover) was derived from UAV RGB imagery. Canopy spectral and structure information and their combination were used to predict soybean leaf area index (LAI), aboveground biomass (AGB), and leaf nitrogen concentration (N) using partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), and extreme learning regression (ELR) with a newly proposed activation function. The results revealed that: (1) UAV imagery-derived high-resolution and detailed canopy structure features, canopy height, and canopy coverage were significant indicators for crop growth monitoring, (2) integration of satellite imagery-based rich canopy spectral information with UAV-derived canopy structural features using machine learning improved soybean AGB, LAI, and leaf N estimation on using satellite or UAV data alone, (3) adding canopy structure information to spectral features reduced background soil effect and asymptotic saturation issue to some extent and led to better model performance, (4) the ELR model with the newly proposed activated function slightly outperformed PLSR, RFR, and SVR in the prediction of AGB and LAI, while RFR provided the best result for N estimation. This study introduced opportunities and limitations of satellite/UAV data fusion using machine learning in the context of crop monitoring.
Weeds are one of the major restrictions to sustaining crop productivity. Weeds often outcompete crops for nutrients, soil moisture, solar radiation, space and provide platforms for breeding of pests and diseases. The ever-growing global food insecurity triggers the need for spatially explicit innovative geospatial technologies that can deliver timely detection of weeds within agro-ecological systems. This will help pinpoint maize fields to be prioritized for weed control. Satellite remote sensing offers incomparable opportunities for precision agriculture, ecological applications and vegetation characterisation, with vast socioeconomic benefits. This work compares and evaluates the strength of Sentinel-2 (S2) satellite with the constellation of Dove nanosatellites i.e. PlanetScope (PS) data in detecting and mapping Striga (Striga hermonthica) weed within intercropped maize fields in Rongo sub-county in western Kenya. We applied the S2 and PS derived spectral data and vegetation indices in mapping the Striga occurrence. Data analysis was implemented, using the Guided Regularised Random Forest (GRRF) classifier. Comparatively, Sentinel-2 demonstrated slightly lower Striga detection capacity than PlanetScope, with an overall accuracy of 88% and 92%, respectively. The results further showed that the VNIR (Blue, Green Red and NIR) and the Atmospheric resistance Vegetation Index (ARVI) were the most fundamental variables in detecting and mapping Striga presence in maize fields. Findings from this work demonstrate that Sentinel-2 data has the capability to provide spatial explicit near real-time field level Striga detection – a previously daunting task with broadband multispectral sensors.
Landsat images have been widely used in support of responsible development of natural resources, disaster risk management (e.g., forest fire, flooding etc.), agricultural production monitoring, as well as environmental change studies due to its medium spatial resolution and rich spectral information. However, its availability and usability are largely constrained by its low revisit frequency. On the other hand, MODIS (Moderate Resolution Imaging Spectroradiometer) images for land studies have much more frequent coverage but with a lower spatial resolution of 250–500 m. To take advantages of the two sensors and expand their availability and usability, during the last decade, a number of image fusion methods have been developed for generating Landsat-like images from MODIS observations to supplement clear-sky Landsat imagery. However, available methods are typically effective or applicable for certain applications. For a better result, a new Prediction Smooth Reflectance Fusion Model (PSRFM) for blending Landsat and MODIS images is proposed. PSRFM consists of a dynamic prediction model and a smoothing filter. The dynamic prediction model generates synthetic Landsat images from a pair of Landsat and MODIS images and another MODIS image, either forward or backward in time. The smoothing filter combines the forward and backward predictions by weighted average based on elapsed time or on the estimated prediction uncertainty. Optionally, the smooth filtering can be applied with constraints based on Normalized Difference Snow Index (NDSI) or Normalized Difference Vegetation Index (NDVI). In comparison to some published reflectance fusion methods, PSRFM shows the following desirable characteristics: (1) it can deal with one pair or two pairs of Landsat and MODIS images; (2) it can incorporate input image uncertainty during prediction and estimate prediction uncertainty; (3) it can track gradual vegetation phenological changes and deal with abrupt land-cover type changes; and (4) for predictions using two pairs of input images, the results can be further improved through the constrained smoothing filter based on NDSI or NDVI for certain applications. We tested PSRFM to generate a Landsat-like image time series by using Landsat 8 OLI and MODIS (MOD09GA) images and compared it to two reflectance fusion algorithms: STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model) and ESTARFM (Enhanced version of STARFM). The results show that the proposed PSRFM is effective and outperforms STARFM and ESTARFM both visually and quantitatively.
Land cover maps are indispensable for decision making, monitoring, and management in agricultural areas, but they are often only available after harvesting. To obtain a timely crop map of a small-scale arable landscape in the Swiss Plateau, we acquired uncalibrated, very high-resolution data, with a spatial resolution of 0.05 m and four spectral bands, using a consumer-grade camera on an unmanned aerial vehicle (UAV) in June 2015. We resampled the data to different spatial and spectral resolutions, and evaluated the method using textural features (first order statistics and mathematical morphology), a random forest classifier for best performance, as well as number and size of the structuring elements. Our main findings suggest the overall best performing data consisting of a spatial resolution of 0.5 m, three spectral bands (RGB—red, green, and blue), and five different sizes of the structuring elements. The overall accuracy (OA) for the full set of crop classes based on a pixel-based classification is 66.7%. In case of a merged set of crops, the OA increases by ~7% (74.0%). For an object-based classification based on individual field parcels, the OA increases by ~20% (OA of 86.3% for the full set of crop classes, and 94.6% for the merged set, respectively). We conclude the use of UAV to be most relevant at 0.5 m spatial resolution in heterogeneous arable landscapes when used for crop classification.
The study investigates the potential of UAV-based remote sensing technique for monitoring of Norway spruce health condition in the affected forest areas. The objectives are: (1) to test the applicability of UAV visible an near-infrared (VNIR) and geometrical data based on Z values of point dense cloud (PDC) raster to separate forest species and dead trees in the study area; (2) to explore the relationship between UAV VNIR data and individual spruce health indicators from field sampling; and (3) to explore the possibility of the qualitative classification of spruce health indicators. Analysis based on NDVI and PDC raster was successfully applied for separation of spruce and silver fir, and for identification of dead tree category. Separation between common beech and fir was distinguished by the object-oriented image analysis. NDVI was able to identify the presence of key indicators of spruce health, such as mechanical damage on stems and stem resin exudation linked to honey fungus infestation, while stem damage by peeling was identified at the significance margin. The results contributed to improving separation of coniferous (spruce and fir) tree species based on VNIR and PDC raster UAV data, and newly demonstrated the potential of NDVI for qualitative classification of spruce trees. The proposed methodology can be applicable for monitoring of spruce health condition in the local forest sites.
Increased forest vulnerability is being reflected as more widespread and severe drought-induced decline episodes. In particular, the Mediterranean area is revealing a high susceptibility to phenomena of loss in tree vitality across species. Within tree species, oaks (Quercus spp.) are experiencing extensive decline in many countries. However, in the wake of the so-called “oak decline phenomenon”, the attention on these species has generally been limited. In this paper, we review the current available literature on oakdecline cases reported within the Mediterranean Basin, with particular remark for those occurred in Italy and Spain. More specifically our main aims were to: (i) provide an update on the patterns and mechanisms of decline by focusing on tree-ring and wood-anatomical variables; (ii) provide some hints for improving the resistance and resilience of oak stands experiencing decline. Our review reveals that drought is reported as the main driver triggering oak decline within the Mediterranean Basin, although other causes (i.e., increasing temperature, pathogens attack or excessive stand density) could exacerbate decline. In most reported cases, drought induced a substantial reduction of growth and changes in some wood anatomical properties. Indeed, growth decline prior death is also indicated as an early-warning signal of impending death. In ring-porous oak species, declining trees were often characterized by a very low production of latewood and a decrease in lumen area of the widest earlywood vessels, suggesting a potential reduction of hydraulic conductivity. Moreover, hydraulic dysfunction is reported as the main cause of decline. Finally, regarding management actions that should be considered for improving the resilience of declining stands and preserve the species-specific stand composition, it could be useful to shorten the rotation period of coppice stands or promoting their gradual conversion towards high forests, and favoring more drought-resistant species should also be considered. In addition, regeneration prior to regeneration cuts should be improved by anticipating seed dispersal or by planting oak seedlings obtained from local germoplasm.
Vegetation Indices (VIs) obtained from remote sensing based canopies are quite simple and effective algorithms for quantitative and qualitative evaluations of vegetation cover, vigor, and growth dynamics, among other applications. These indices have been widely implemented within RS applications using different airborne and satellite platforms with recent advances using Unmanned Aerial Vehicles (UAV). Up to date, there is no unified mathematical expression that defines all VIs due to the complexity of different light spectra combinations, instrumentation, platforms, and resolutions used. Therefore, customized algorithms have been developed and tested against a variety of applications according to specific mathematical expressions that combine visible light radiation, mainly green spectra region, from vegetation, and nonvisible spectra to obtain proxy quantifications of the vegetation surface. In the real-world applications, optimization VIs are usually tailored to the specific application requirements coupled with appropriate validation tools and methodologies in the ground. The present study introduces the spectral characteristics of vegetation and summarizes the development of VIs and the advantages and disadvantages from different indices developed. This paper reviews more than 100 VIs, discussing their specific applicability and representativeness according to the vegetation of interest, environment, and implementation precision. Predictably, research, and development of VIs, which are based on hyperspectral and UAV platforms, would have a wide applicability in different areas.
The purpose of this study is to validate the accuracy of abundance map reference data (AMRD) for three airborne imaging spectrometer (IS) scenes. AMRD refers to reference data maps ("ground truth") that are specifically designed to quantitatively assess the performance of spectral unmixing algorithms. While classification algorithms typically label whole pixels as belonging to certain ground cover classes, spectral unmixing allows pixels to be composed of fractions or abundances of each class. The AMRD validated in this paper were generated using our previously-proposed remotely-sensed reference data (RSRD) technique, which spatially aggregates the results of standard classification or unmixing algorithms from fine spatial-scale IS data to produce AMRD for co-located coarse-scale IS data. Validation of the three scenes was accomplished by estimating AMRD in 51 randomly-selected 10 m×10 m plots, using seven independent methods and observers. These independent estimates included field surveys by two observers, imagery analysis by two observers and RSRD by three algorithms. Results indicated statistically-significant differences between all versions of AMRD. Even AMRD from our two field surveys were significantly different for two of the four ground cover classes. These results suggest that all forms of reference data require validation prior to use in assessing the performance of classification and/or unmixing algorithms. Given the significant differences between the independent versions of AMRD, we propose that the mean of all (MOA) versions of reference data for each plot and class is most likely to represent true abundances. Our independent versions of AMRD were compared to MOA to characterize error and uncertainty. Best case results were achieved by a version of imagery analysis, which had mean coverage area differences of 2.0%, with a standard deviation of 5.6%. One of the RSRD algorithms was nearly as accurate, achieving mean differences of 3.0%, with a standard deviation of 6.3%. Further analysis of statistical equivalence yielded an overall zone of equivalence between [-7.0%, 7.2%] for this version of RSRD. The relative accuracy of RSRD methods is promising, given their potential to efficiently generate scene-wide AMRD. These results provide the first known validated abundance level reference data for airborne IS data.
Although many advances have been made in past decades, land cover classification of fine-resolution
remotely sensed (RS) data integrating multiple temporal, angular, and spectral features remains limited,
and the contribution of different RS features to land cover classification accuracy remains uncertain. We
proposed to improve land cover classification accuracy by integrating multi-source RS features through
data fusion. We further investigated the effect of different RS features on classification performance.
The results of fusing Landsat-8 Operational Land Imager (OLI) data with Moderate Resolution Imaging
Spectroradiometer (MODIS), China Environment 1A series (HJ-1A), and Advanced Spaceborne Thermal
Emission and Reflection (ASTER) digital elevation model (DEM) data, showed that the fused data integrating
temporal, spectral, angular, and topographic features achieved better land cover classification accuracy
than the original RS data. Compared with the topographic feature, the temporal and angular
features extracted from the fused data played more important roles in classification performance, especially
those temporal features containing abundant vegetation growth information, which markedly
increased the overall classification accuracy. In addition, the multispectral and hyperspectral fusion successfully
discriminated detailed forest types. Our study provides a straightforward strategy for hierarchical
land cover classification by making full use of available RS data. All of these methods and findings
could be useful for land cover classification at both regional and global scales.
Crop advisors and farmers increasingly use drone data as part of their decision making. However, the vast majority of UAS-based vegetation mapping services support only the calculation of a relative NDVI derived from compressed JPEG pixel values and do not include the possibility to include more complex aspects like soil correction. In our ICPA12 contribution, we demonstrated the effects and consequences of the above shortcomings. Here, we present the stepwise development of a solution to ensure reliable input for crop advisors as a basis for site-specific crop management based on drone data. UAS flights are executed with a Trimble UX5 (HP) over a Belgian farm comprising four different crop types during a 3 month interval. Vegetation index maps derived from compressed JPEG imagery as well as preprocessed raw sensor data from a modified conventional CIR camera are evaluated against those from a true multispectral camera, and we examine the ability to calibrate the maps. Resulting maps are compared to NDVI values from the active close-range Trimble GreenSeeker crop sensor. Based on the results, we discuss under which conditions the three different data types can be used to complement traditional measurements in addressing within-season crop variability.
This paper presents two novel hyperspectral mixture models and associated unmixing algorithms. The two models assume a linear mixing model corrupted by an additive term whose expression can be adapted to account for multiple scattering nonlinearities (NL), or mismodelling effects (ME). The NL model generalizes bilinear models by taking into account higher order interaction terms. The ME model accounts for different effects such as endmember variability or the presence of outliers. The abundance and residual parameters of these models are estimated by considering a convex formulation suitable for fast estimation algorithms. This formulation accounts for constraints such as the sum-to-one and non-negativity of the abundances, the non-negativity of the nonlinearity coefficients, the spectral smoothness of the ME terms and the spatial sparseness of the residuals. The resulting convex problem is solved using the alternating direction method of multipliers (ADMM) whose convergence is ensured theoretically. The proposed mixture models and their unmixing algorithms are validated on both synthetic and real images showing competitive results regarding the quality of the inference and the computational complexity when compared to the state-of-the-art algorithms.
The evaluation at detailed spatial scale of soil status after severe fires may provide useful information on the recovery of burned forest ecosystems. Here, we aimed to assess the potential of combining multispectral imagery at different spectral and spatial resolutions to estimate soil indicators of burn severity. The study was conducted in a burned area located at the northwest of the Iberian Peninsula (Spain). One month after fire, we measured soil burn severity in the field using an adapted protocol of the Composite Burn Index (CBI). Then, we performed soil sampling to analyze three soil properties potentially indicatives of fire-induced changes: mean weight diameter (MWD), soil moisture content (SMC) and soil organic carbon (SOC). Additionally, we collected post-fire imagery from the Sentinel-2A MSI satellite sensor (10–20 m of spatial resolution), as well as from a Parrot Sequoia camera on board an unmanned aerial vehicle (UAV; 0.50 m of spatial resolution). A Gram-Schmidt (GS) image sharpening technique was used to increase the spatial resolution of Sentinel-2 bands and to fuse these data with UAV information. The performance of soil parameters as indicators of soil burn severity was determined trough a machine learning decision tree, and the relationship between soil indicators and reflectance values (UAV, Sentinel-2 and fused UAV-Sentinel-2 images) was analyzed by means of support vector machine (SVM) regression models. All the considered soil parameters decreased their value with burn severity, but soil moisture content, and, to a lesser extent, soil organic carbon discriminated at best among soil burn severity classes (accuracy = 91.18 %; Kappa = 0.82). The performance of reflectance values derived from the fused UAV-Sentinel-2 image to monitor the effects of wildfire on soil characteristics was outstanding, particularly for the case of soil organic carbon content (R² = 0.52; RPD = 1.47). This study highlights the advantages of combining satellite and UAV images to produce spatially and spectrally enhanced images, which may be relevant for estimating main impacts on soil properties in burned forest areas where emergency actions need to be applied.
Soil seed bank is an important driver of vegetation dynamics, particularly in fire-prone Mediterranean ecosystems. In this study, we disentangle the effects of fire-related cues on the dynamics of the soil seed bank in semiarid oak forests of western Iran. Soil samples were collected in two types of forests: below closed canopy or below the open canopy of more intensively used forests. In a greenhouse experiment, the effects of five treatments were tested: application of ash (Ash), smoke (Smoke), 50°C temperature (Heat50), 100°C temperature (Heat100) and control. Soil samples were regularly monitored and the number of germinated seeds (seedlings) by species were recorded in each treatment. Diversity indices were also computed. We identified 18 species from 16 genera and 9 families. 56% of the seedlings emerged from the closed canopy forest soils and 44% from the open canopy forests. Seedling density and diversity values were the lowest in the Heat100 treatment. No annual grasses and tree species were recorded in this treatment. In contrast, the influence of the other fire-related cues was mediated by the forest type covertype of forest cover: the number of emerged seedlings was the highest in Smoke and Heat50 in the open canopy and closed canopy, respectively. This mediation effect was also reflected in the composition of the control as tree species were only present in the closed canopy and annual forbs were less abundant there compared to the open canopy type. Considering all fire treatments together, the seedling density was significantly higher in the closed canopy type than in the open canopy (309.6 ± 28.03 vs. 240.8 ± 20.78 seeds/m2). Our results showed that intense fires leading to high soil temperatures are largely detrimental to soil seed bank germination and species diversity, whereas less intense fires can be favourable depending on species life form and the type of forest. Our findings highlight that post-fire regeneration from the soil seed bank is more promising in closed forests than in forests with an open canopy in the studied ecosystem.
Global urbanization changes land cover patterns and affects the living environment of humans. However, urbanization and its evolution process, i.e., conversions among diverse land covers, are hard to measure, as existing land cover maps usually have low temporal resolutions; conversely, long-term and temporally dense land cover maps, such as vegetation-impervious-soil decomposition maps base on MODIS, ignore the important land cover of cropland in urban evolution process (UEP). To resolve the issue, this study suggests a novel model named time-extended non-crop vegetation-impervious-cropland (Time V-I-C) to represent and quantify different stages of UEP; then, a normalized multi-objective T-ConvLSTM (NMT) method is proposed to unmix cropland, non-crop vegetation, and impervious based on the intra-annual remotely-sensed time series, and obtain their fractions in each pixel for generating UEP maps. Consequently, UEP maps from 2001 to 2018 are generated for two Chinese urban agglomerations, i.e., Beijing-Tianjin-Hebei and Yangtze River Delta urban agglomerations. The mapping results have high accuracies with a small standard error of regression (SER) of 13.1%, small root mean square error (RMSE) of 12.6%, and small mean absolute error (MAE) of 8.4%, and the maps reveal the different UEP in the two urban agglomerations. Therefore, this study provides a new idea for expressing UEP and contributes to a wide range of urbanization studies and sustainable city development.
Spatiotemporal reflectance fusion has received considerable attention in recent decades. However, various challenges remain despite varying levels of success, especially regarding the recovery of spatial details with complex land cover changes. Taking the blending of Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) images as an example, this article presents a locally weighted unmixing-based spatiotemporal image fusion model (LWU-STFM) that focuses on recovering complex land cover changes. The core idea is to redefine the land use class of each pixel featuring land cover change at the prediction date. The spatial unmixing process is enhanced using a proposed geographically spectrum-weighted regression (GSWR), and then, we optimize similar neighboring pixels for the final weighted-based prediction. Experiments are conducted using semisimulated and actual time-series Landsat–MODIS datasets to demonstrate the performance of the proposed LWU-STFM compared with the classic spatial and temporal adaptive reflectance fusion model (STARFM), flexible spatiotemporal data fusion (FSDAF), two enhanced FSDAF models (SFSDAF and FSDAF 2.0), and a virtual image pair-based spatiotemporal fusion model for spatial weighting (VIPSTF-SW). The results reveal that the proposed LWU-STFM outperforms the other five models with the best quantitative accuracy. In terms of the relative dimensionless global error (ERGAS) index, the errors of Landsat-like images generated using LWU-STFM are 2.8%–63.4% lower than those of other models. From visual comparisons, LWU-STFM predictions illustrate encouraging improvements in recovering spatial details of pixels with complex land cover changes in heterogeneous landscapes and, thus, advancing applications of spatiotemporal image fusion for continuous and fine-scale land surface monitoring.
Vegetation cover maps across ecologically-fragile and particularly arid and semi-arid forest ecosystems are prerequisites for their monitoring and management. Direct and field-based measurements of vegetation cover pose serious challenges due to high costs and inaccessibility in harsh terrains, whereas multispectral remote sensing offers objective, spatially-explicit and rapid alternatives. One of the most straightforward tools is the use of broadband vegetation indices (VIs), which are mathematical derivations from multispectral bands that are correlated with various vegetation traits. There are a number of broadband VIs that reach their optimum performance by calibrating their regulatory parameters. We improved the performance of selected VIs for both greenness estimation and land-cover classification across semi-arid woodlands by optimizing their regulatory parameters. We showed this across two separate areas in highly-fragile and sparse vegetation of Zagros mountains of Iran. Regulatory parameters were optimized by multi-objective particle swarm optimization (MOPSO) for Enhanced VI (EVI) and two innovative, more complex broadband indices that use red, blue, and near-infrared multispectral bands. Then, they were applied to estimate greenness and classify vegetation, and were validated by subsets of very high-resolution optical imagery. The results suggest high accuracy of these indicators for estimating and classifying vegetation compared with the commonly-used broadband VIs. Amongst the improved VIs, the one with a more complex combination of spectral bands comparatively returned the best performance, that was 1.34× and 1.33× higher in greenness estimation and 1.58× higher in classification compared with the benchmark NDVI. They also described a higher variance across systematic transects in both regions. In conclusion, both greenness estimation and classification of semi-arid, sparse woodlands were more accurate by optimizing their regulatory parameters.
Urban tree species classification is a challenging task due to spectral and spatial diversity within an urban environment. Unmanned aerial vehicle (UAV) platforms and small-sensor technology are rapidly evolving, presenting the opportunity for a comprehensive multi-sensor remote sensing approach for urban tree classification. The objectives of this paper were to develop a multi-sensor data fusion technique for urban tree species classification with limited training samples. To that end, UAV-based multispectral, hyperspectral, LiDAR, and thermal infrared imagery was collected over an urban study area to test the classification of 96 individual trees from seven species using a data fusion approach. Two supervised machine learning classifiers, Random Forest (RF) and Support Vector Machine (SVM), were investigated for their capacity to incorporate highly dimensional and diverse datasets from multiple sensors. When using hyperspectral-derived spectral features with RF, the fusion of all features extracted from all sensor types (spectral, LiDAR, thermal) achieved the highest overall classification accuracy (OA) of 83.3% and kappa of 0.80. Despite multispectral reflectance bands alone producing significantly lower OA of 55.2% compared to 70.2% with minimum noise fraction (MNF) transformed hyperspectral reflectance bands, the full dataset combination (spectral, LiDAR, thermal) with multispectral-derived spectral features achieved an OA of 81.3% and kappa of 0.77 using RF. Comparison of the features extracted from individual sensors for each species highlight the ability for each sensor to identify distinguishable characteristics between species to aid classification. The results demonstrate the potential for a high-resolution multi-sensor data fusion approach for classifying individual trees by species in a complex urban environment under limited sampling requirements.
Non-photosynthetic vegetation (NPV) is an essential component in various vegetation-soil ecosystems. Both phenology and disturbance lead to a transition from photosynthetic vegetation to NPV and vice versa. Due to the similar spectral reflectance of NPV and bare soil (BS) in the visible-near infrared region (400–1000 nm), NPV and BS separation is relying on the shortwave infrared (SWIR) bands in most cases. The lignin and cellulose absorption feature is around 2100 nm, which is the most distinctive feature of NPV. However, the water absorption feature is much stronger in the SWIR, increasing the difficulty for NPV-BS separation when wet. Recently, Sentinel-2/3 satellites add more bands in the near infrared (NIR), which provide an extra opportunity for index building and application. Based on the difference captured by derivative spectra, a spectral index, NPV-Soil Separation Index (NSSI), is proposed to realize the separation using two NIR bands within 750–900 nm range in this study. Using spectra of photosynthetic vegetation (PV), NPV, and BS acquired from world-recognized spectral libraries, NSSI is built and validated as effective for lab-collected data. With the triangle method, one of the linear spectral unmixing methods, the fractional cover of PV, NPV, and BS can be estimated. Over a woodland study area, the fractional cover retrieved by cellulose absorption index (CAI) and NDVI combination of ZY1-02D AHSI hyperspectral image is 26.41%, 37.56%, 36.03% for PV, NPV, and BS in order. With the proposed NSSI-NDVI combination, the corresponding estimated fractional cover is 23.31%, 38.44%, 38.25% using Sentinel-2 MSI and 24.58%, 36.74%, and 38.68% using Sentinel-3 OLCI image. The comparable validation result confirms that the proposed NSSI is effective for NPV-BS separation. Moreover, the triangle method of NSSI-NDVI combination is applied on both grassland and cropland images to examine its feasibility on varied types of typical vegetation-soil ecosystems, and the well-built triangular space supports its feasibility. Relying on NIR bands, NSSI can avoid strong water absorption in the SWIR. Also, the feasibility of NSSI being used on multiple multispectral satellite sensors, especially the Sentinel series, makes continuous mapping for NPV over a large spatial scale possible.
Several vegetation indices have been developed, with the normalized difference vegetation index (NDVI) been the most studied and commonly used. To generate an NDVI map, a relatively high-cost multispectral sensor is required; but currently, most UAVs are equipped with low-cost RGB cameras. For that reason, other indices that utilize RGB data have been developed to generate maps similar to NDVI and minimize the data acquisition cost, such as the triangular greenness index (TGI) and the visible atmospheric resistant index (VARI). However, several studies found that these indices cannot be recommended as reliable general-purpose crop health indicators. This study utilizes a genetic algorithm to develop a new visible index (visible NDVI; vNDVI) that estimates NDVI values of vegetation from uncalibrated RGB cameras mounted on UAVs (or other remote sensing platforms). Three experiments were conducted to create and validate the proposed index. First, the NDVI values generated from a multispectral camera were compared with the NDVI values generated by a hyperspectral camera. In the second experiment, the vNDVI formula was created using a genetic algorithm. The third experiment validates the proposed vNDVI, generated from two uncalibrated RGB cameras, in three different crops (citrus, grapes, and sugarcane). The proposed vNDVI proved to be highly accurate on estimating NDVI values by just using RGB cameras, with an overall mean percentage error of 6.89% and a mean average error of 0.052 in all three crops, providing a low-cost alternative for remote sensing and plant phenotyping.
The Zagros Mountains forests extend across 11 provinces in Iran and constitute approximately 40.0% of the country’s woodlands. These forests have important soil conservation and water regulation functions. Over the last decade, these forests have been declining in oak populations in many places, triggered by factors such as drought, pathogens like the fungus Biscogniauxia mediterranea, and pests such as borer beetles. Mapping the regions that show such a decline is the first step to addressing and managing the risks posed by this environmental calamity. In this research, we focus on the forests surrounding Malekshahi city in the Ilam province of Iran. Using Landsat data from the years 2000 to 2016, we determined the spatial distribution of oak decline in the region. After applying a forest/non-forest classification, appropriate spectral indices including Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) were selected. Together with ground truth data, two regression methods (linear regression and support vector regression (SVR)) were used to model the decline score of each pixel based on the slope of variation of selected spectral indices during the observed 17 years. The oak forests were then classified into four categories: healthy forests, low-severity-declined forests, mid-severity declined forests, and high-severity declined forests, based on the respective estimated decline scores. SVR mapped different severities of oak decline with an overall accuracy of 51%, which appears to be due to the dependency of the method on the time of decline during the 17-year timeframe. However, in a binary classification mode – meaning classifying decline score to be either ‘Healthy’ or ‘decline’ – both regression methods were able to detect declined pixels with a producer’s accuracy of 100%.
The foundation of all quantitative remote sensing is based on the ability of sensors to convert energy into digital signals. This process requires an understanding of the radiometric and spectral behaviour of the instrument. This information is rarely available for the consumer-grade digital cameras commonly used in unmanned aerial vehicle (UAV) systems. This study measured the spectral and radiometric characteristics of consumer-grade digital cameras in laboratory-based experiments. The spectral characteristics showed broad overlapping spectral bands in the visible and near-infrared wavelengths with spectral band widths between 70 and 100 nm. The radiometric response functions were non-linear in nature. The nonlinearity resulted in the inability of reliable spectral reflectance calculations. When compared to the spectral reflectance values computed from an ASD FieldSpec-3, differences were as high as 42% from the true value with average deviations between 23% and 32%. When characterized, the spectral reflectance values compared more favourably to ASD-computed spectral reflectance.
An unmanned aerial system (UAS)-based imaging technology has gained great interests in modern photogrammetry and remote sensing. However, due to the limitations of UAS imaging devices, image enhancement (IE) has become a necessary process for improving the visual appearance of UAS images. Although a great amount of effort has been focused on improving image quality from different aspects, the major obstacles are from computational efficiency and complexity, such as manually adjusting the associated algorithmic parameters that account for various image luminance. To overcome these drawbacks, we propose a new adaptive yet highly efficient luminance enhancement method, namely, adaptive trigonometric transformation function (ATTF), for enhancing the visual quality of digital color images captured by a UAS. The ATTF is derived from a tangent-based transformation function whose characteristics adaptively change with respect to the variation of the image luminance. By combining ATTF with a Laplacian operator and a color restoration process, a well-balanced color image is obtained. The effectiveness of the proposed technique is evaluated on various UAS-based images and compared with other IE techniques.
Canopy height (Hcanopy) and aboveground biomass (AGB) of crops are two basic agro-ecological indicators that can provide important indications on the growth, light use efficiency, and carbon stocks in agro-ecosystems. In this study, hundreds of stereo images with very high resolution were collected to estimate Hcanopy and AGB of maize using a low-cost unmanned aerial vehicle (UAV) system. Millions of point clouds that are related to the structure from motion (SfM) were produced from the UAV stereo images through a photogrammetric workflow. Metrics that are commonly used in airborne laser scanning (ALS) were calculated from the SfM point clouds and were tested in the estimation of maize parameters for the first time. In addition, the commonly used spectral vegetation indices calculated from the UAV orthorectified image were also tested. Estimation models were established based on the UAV variables and field measurements with cross validation, during which the performance of the UAV variables was quantified. Finally, the following results were achieved: (1) the spatial patterns of maize Hcanopy and AGB were predicted by a multiple stepwise linear (SWL) regression model (R2 = 0.88, rRMSE = 6.40%) and a random forest regression (RF) model (R2 = 0.78, rRMSE = 16.66%), respectively. (2) The UAV-estimated maize parameters were proved to be comparable to the field measurements with a mean error (ME) of 0.11 m for Hcanopy, and 0.05 kg/m2 for AGB. (3) The SfM point metrics, especially the mean point height (Hmean) greatly contributed to the estimation model of maize Hcanopy and AGB, which can be promising indicators in the detection of maize biophysical parameters. To conclude, the variations in spectral and structural attributes for maize canopy should be simultaneously considered when only simple RGB images are available for estimating maize AGB. This study provides some suggestions on how to make full use of the low-cost and high-resolution UAV stereo images in precision agro-ecological applications and management.