added a research item
Illumination variations in non-atmospherically corrected high-resolution satellite (HRS) images acquired at different dates/times/locations pose a major challenge for large-area environmental mapping and monitoring. This problem is exacerbated in cases where a classification model is trained only on one image (and often limited training data) but applied to other scenes without collecting additional samples from these new images. In this research, by focusing on caribou lichen mapping, we evaluated the potential of using conditional Generative Adversarial Networks (cGANs) for the normalization of WorldView-2 (WV2) images of one area to a source WV2 image of another area on which a lichen detector model was trained. In this regard, we considered an extreme case where the classifier was not fine-tuned on the normalized images. We tested two main scenarios to normalize four target WV2 images to a source 50 cm pansharpened WV2 image: (1) normalizing based only on the WV2 panchromatic band, and (2) normalizing based on the WV2 panchromatic band and Sentinel-2 surface reflectance (SR) imagery. Our experiments showed that normalizing even based only on the WV2 panchromatic band led to a significant lichen-detection accuracy improvement compared to the use of original pansharpened target images. However, we found that conditioning the cGAN on both the WV2 panchromatic band and auxiliary information (in this case, Sentinel-2 SR imagery) further improved normalization and the subsequent classification results due to adding a more invariant source of information. Our experiments showed that, using only the panchromatic band, F1-score values ranged from 54% to 88%, while using the fused panchromatic and SR, F1-score values ranged from 75% to 91%.
Relating ground photographs to UAV orthomosaics is a key linkage required for accurate multi-scaled lichen mapping. Conventional methods of multi-scaled lichen mapping, such as random forest models and convolutional neural networks, heavily rely on pixel DN values for classification.However, the limited spectral range of ground photos requires additional characteristics to differentiate lichen from spectrally similar objects, such as bright logs. By applying a neural network to tiles of a UAV orthomosaics, additional characteristics, such as surface texture and spatial patterns, can be used for inferences. Our methodology used a neural network (UAV LiCNN) trained on ground photo mosaics to predict lichen in UAV orthomosaic tiles. The UAV LiCNN achieved mean user and producer accuracies of 85.84% and 92.93%, respectively, in the high lichen class across eight different orthomosaics. We compared the known lichen percentages found in 77 vegetation microplots with the predicted lichen percentage calculated from the UAV LiCNN, resulting in a R2 relationship of 0.6910. This research shows that AI models trained on ground photographs effectively classify lichen in UAV orthomosaics. Limiting factors include the misclassification of spectrally similar objects to lichen in the RGB bands and dark shadows cast by vegetation.
Arctic temperatures have increased at almost twice the global average rate since the industrial revolution. Some studies also reported a further amplified rate of climate warming at high elevations; namely, the elevation dependency of climate change. This elevation-dependent climate change could have important implications for the fate of glaciers and ecosystems at high elevations under climate change. However, the lack of long-term climate data at high elevations, especially in the Arctic, has hindered the investigation of this question. Because of the linkage between climate warming and plant phenology changes and remote sensing’s ability to detect the latter, remote sensing provides an alternative way for investigating the elevation dependency of climate change over Arctic mountains. This study investigated the elevation-dependent changes to plant phenology using AVHRR (Advanced Very High Resolution Radiometer) time series from 1985 to 2013 over five study areas in Canada’s Arctic. We found that the start of the growing season (SOS) became earlier faster with an increasing elevation over mountainous study areas (i.e., Sirmilik, the Torngat Mountains, and Ivvavik National Parks). Similarly, the changes rates in the end of growing season (EOS) and the growing season length (GSL) were also higher at high elevations. One exception was SOS in the Ivvavik National Park: “no warming trend” with the May-June temperature at a nearby climate station decreased slightly during 1985–2013, and so no elevation-dependent amplification.
Mining activities in Canada’s pristine Arctic (e.g., driving on unpacked roads, blasts, rock grinding, diesel combustion, and garbage incineration) could add local sources of airborne fine particulate matter with a diameter of < 2.5 μm (PM2.5) to their surrounding area. The increase in PM2.5 above the background level around a mine represents a potential disturbance to caribou. To quantify the spatial distribution of the elevated PM2.5, we investigated three different sampling schemes to measure PM2.5 concentration using a portable monitor. We found that the best sampling scheme was to use the regional background PM2.5 as the reference and analyze the anomaly of PM2.5 measured at sites around the mine complex from the background level. The regional background PM2.5 values were measured at the Daring Lake Tundra Research Station during 2018 and 2019. Our results indicated that the background PM2.5 was not a low and constant value but varied with rain events, wind direction, and the impacts of forest fire smoke. After excluding periods affected by forest fires smokes, we found the background PM2.5 was close to 0 μg m⁻³ for the first few hours after rain, and then increased logistically with the time after rain (tar) to the maximum of 5 (or 10) μg m⁻³ when the wind came from the north (or south) of the NW-SE axis. The NW-SE axis in western Canada divides the tundra north with few anthropogenic PM2.5 sources from the forested south with many PM2.5 sources from forest fire smokes and human activities. Analyses of PM2.5 anomaly from the background (i.e., PM2.5 measured at a site around the mining complex—the background level at the corresponding tar and wind direction) revealed that the zone of elevated PM2.5 around the mine (Zepm) expanded with tar. In the first few hours after rain, PM2.5 was close to 0 everywhere except within meters of a source (e.g., a truck exhaust) in the downwind direction. During tar = 6 to 96 h, Zepm expanded to 6.3 km in the downwind direction when the wind came from south of the NW-SE axis. A similar result was found in the downwind direction when the wind came from north of the NW-SE axis, with Zepm = 4.4 km. In the upwind direction, the value of Zepm was much smaller, being 0.7 km (or 1.0 km) when the wind came from the north (or south) of the NW-SE axis. For the period of tar between 96 and 192 hours, Zepm further expanded to 21.2 km when the wind from the south of the NW-SE axis. The results from this study indicated that this reference paradigm that uses the regional background PM2.5 as the reference in combination with a portable PM2.5 monitor worked well for quantifying the tempo-spatial patterns of PM2.5 at locations in remote and mostly pristine Arctic. However, their effectiveness for other regions needs further investigation.
Lichen is an important food source for caribou in Canada. Lichen mapping using remote sensing (RS) images could be a challenging task, however, as lichens generally appear in unevenly distributed, small patches, and could resemble surficial features. Moreover, collecting lichen labeled data (reference data) is expensive, which restricts the application of many robust supervised classification models that generally demand a large quantity of labeled data. The goal of this study was to investigate the potential of using a very-high-spatial resolution (1-cm) lichen map of a small sample site (e.g., generated based on a single UAV scene and using field data) to train a subsequent classifier to map caribou lichen over a much larger area (~0.04 km2 vs. ~195 km2) and a lower spatial resolution image (in this case, a 50-cm WorldView-2 image). The limited labeled data from the sample site were also partially noisy due to spatial and temporal mismatching issues. For this, we deployed a recently proposed Teacher-Student semi-supervised learning (SSL) approach (based on U-Net and U-Net++ networks) involving unlabeled data to assist with improving the model performance. Our experiments showed that it was possible to scale-up the UAV-derived lichen map to the WorldView-2 scale with reasonable accuracy (overall accuracy of 85.28% and F1-socre of 84.38%) without collecting any samples directly in the WorldView-2 scene. We also found that our noisy labels were partially beneficial to the SSL robustness because they improved the false positive rate compared to the use of a cleaner training set directly collected within the same area in the WorldView-2 image. As a result, this research opens new insights into how current very high-resolution, small-scale caribou lichen maps can be used for generating more accurate large-scale caribou lichen maps from high-resolution satellite imagery.
Since mid-1980’s, the population of the Bathurst barren ground caribou (Rangifer tarandus) in Canada’s Arctic has declined by 93%. In order to develop and implement an effective recovery plan, it is important to know how various factors have cumulatively impacted the population decline. To contribute to the knowledge, we investigated the following two questions: how have changes in climate-induced habitat conditions impacted the peak calving date of the Bathurst caribou, and what was the implication of the impact on the population? Our results indicate that the peak calving date was impacted by changes in habitat conditions (e.g., the start date of vegetation growing season SOS) in a complex manner. Large inter-annual variations in SOS on the calving ground and summer range of the Bathurst herd were observed during 1985 and 2012, with the largest difference being 29 days. A 1-day delay of SOS in year i − 1 on the calving ground (SOScg(i − 1)) from its normal date could result in a 0.5-day delay in the peak calving date in year i, likely caused by the delay in the conception date in the previous fall. However, advances in SOScg(i − 1) did not alter the peak calving date in year i. Furthermore, a 1-day delay (or advance) in the current year’s SOS on the summer range (SOSsr(i)) might cause a 0.23-day delay (or advance) in the peak calving date in the current year, likely through changing the caribou’s gestation duration. Together SOScg(i − 1) and SOSsr(i) explained 69.1% of the variation in the peak calving date of the Bathurst caribou herd during 1985–2012, indicating the cumulative impacts on the peak calving date by the changing habitat conditions over a period of 2 years and thus the validation of the cumulative habitat impact hypothesis. Finally, our results also show that a 1-day delay in the peak calving date corresponded approximately 2–3% reduction in the birth rate of the Bathurst caribou, and thus might have been partially responsible for the population decline.
This study explores how dust from the Ekati Diamond Mine potentially affects the availability and quality of forage on the seasonal range of the Bathurst caribou herd. Understanding the effects of dust as a source of disturbance is important because the Bathurst caribou population has declined by 93% since the middle 1980s and there are reports that caribou in general may avoid mining projects. There are several challenges for quantifying dust impacts: 1) Natural variations (e.g., topography, natural disturbance, and soil pH) may also impact forage availability and quality for caribou. To minimize their masking effect, we stratified survey sites into seven land cover classes and selected the most populous class (i.e., the dwarf shrub) for assessing the impact. 2) Within class variation (e.g., the proportion of area covered by rocks where vascular plants and lichen do not grow) can further skew the analysis. We eliminated this problem by examining only the area not covered by rocks. 3) Coarse and fine suspended particulates have different spatial coverages, chemical compositions, and pH values. Consequently, their impacts on caribou forage can be different. To distinguish their impacts, we sampled two areas: transects from the Misery Haul Road that has been in active use vs. those from a rarely used spur road outside the Misery Camp. We sampled percent vegetation cover, soil pH, and dust on leaves along these transects during the summers of 2015 and 2016. Our results indicated that the amount of dust on leaves in a zone of ~1000 m from the Misery Haul Road was 3 - 9 times than that of background sites. The zone of reduced lichen percent cover was also about 1000 m. In contrast, these road dust-induced changes in caribou forage were not observed for the dust-free transect from the spur road.
Leaf area index (LAI) is an important structural vegetation parameter that is commonly derived from remotely sensed data. It has been used as a reliable indicator for vegetation's cover, status, health and productivity. In the past two decades, various Canada-wide LAI maps have been generated by the Canada Centre for Remote Sensing (CCRS). These products have been produced using a variety of very coarse satellite data such as those from SPOT VGT and NOAA AVHRR satellite data. However, in these LAI products, the mapping of the Canadian northern vegetation has not been performed with field LAI measurements due in large part to scarce in situ measurements over northern biomes. The coarse resolution maps have been extensively used in Canada, but finer resolution LAI maps are needed over the northern Canadian ecozones, in particular for studying caribou habitats and feeding grounds.In this study, a new LAI algorithm was developed with particular emphasis over northern Canada using a much finer resolution of remotely sensed data and in situ measurements collected over a wide range of northern arctic vegetation. A statistical relationship was developed between the in situ LAI measurements collected over vegetation plots in northern Canada and their corresponding pixel spectral information from Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) data. Furthermore, all Landsat TM and ETM+ data have been pre-normalized to NOAA AVHRR and SPOT VGT data from the growing season of 2005 to reduce any seasonal or temporal variations. Various spectral vegetation indices developed from the Landsat TM and ETM + data were analysed in this study. The reduced simple ratio index (RSR) was found to be the most robust and an accurate estimator of LAI for northern arctic vegetation. An exponential relationship developed using the Theil–Sen regression technique showed an R of 0.51 between field LAI measurement and the RSR. The developed statistical relationship was applied to a pre-existing Landsat TM 250 m resolution mosaic for northern Canada to produce the final LAI map for northern Canada ecological zones. Furthermore, the 250 m resolution LAI estimates, per ecological zone, were almost generally lower than those of the CCRS Canada-wide VGT LAI maps for the same ecozones. Validation of the map with LAI field data from the 2008 season, not used in the derivation of the algorithm, shows strong agreement between the in situ LAI measurement values and the map-estimated LAI values.
Information on biomass distribution is needed to estimate GHG emissions and removals from land use changes in Canada's north for UNFCCC reporting. This paper reports aboveground biomass measurements along the Dempster Highway transect in 2004, and around Yellowknife and the Lupin Gold Mine in 2005. The measured aboveground biomass ranges are 10–100 t ha−1 for woodlands, 1–100 t ha−1 for shrub sites, and 0.5–10 t ha−1 for grass/herbs sites. The root mean squared error (RMSE) of measurements is 21%, and the median absolute percentage error (MedAPE) is 14%. The combination of JERS backscatter and Landsat TM4/TM5 gives the best biomass equation for the Dempster Highway transect, with r 2 = 0.72 when using a one‐step approach (i.e. using all points) and 0.78 when using a two‐step approach (i.e. stratifying data into three classes: grass, shrub, and woodlands). The two‐step approach reduces the MedAPE from 53% to 33%. The validation against Yellowknife & Lupin data indicates that the equations have good transferability. The improvement of two‐step approach over the one‐step approach, however, is not significant for the validation dataset, suggesting that the one‐step approach is as good as the two‐step approach when applied over areas outside where the equations are developed. The relationships and error analysis of this study, as well as the final estimate of GHG emission/removal over Canada's north have been incorporated into Canada's 2006 UNFCCC report.