Search

Lakes and reservoirs provide important ecosystem services that support human welfare and socio-economic activities. However, in many world regions, the ecological integrity of lakes and reservoirs is threatened by human perturbations and climate change. The Secchi disk depth (SDD) is a widely-used proxy representing the trophic status of lakes and reservoirs and can be retrieved from remote sensing data. Despite their potential for large-scale (regional, global) and long-term lake and reservoir water clarity assessment, the transferability of remote sensing-based models has been a major limitation. In this study, we assembled in situ SDD in lakes and reservoirs across North America (NA) from five different sources. We identified a subset of 3235 samples collected within ±7 days of a Landsat satellite overpass. Relationships between various spectral index models calculated from Landsat top-of-atmosphere reflectance and in situ SDD were analyzed. A model based on Landsat blue/green plus blue/red ratios (denoted as RGRB) was selected to retrieve the SDD of all NA lakes and reservoirs. The RGRB model performed well during calibration (R2 = 0.81) and validation (R2 = 0.78, MAPE = 30.85 %). This model also exhibited stable and reliable performances regardless of the Landsat sensors (TM, ETM+, and OLI), despite spectral configuration differences among these sensors. RGRB was implemented to generate SDD maps for all lakes and reservoirs (water surface area ≥1 ha) across NA in 2019. More than 2.9 million lakes and reservoirs were mapped with Landsat OLI images, resulting in an average SDD of 3.84 ± 1.77 m. A strong positive relationship between average SDD and log-transformed water surface area (R2 = 0.80, p < 0.001) indicated that large lakes and reservoirs tend to be more transparent than small ones. Latitudinal variations were found in the water clarity gradient, with maximum SDD recorded at the 35°N–60°N latitude and lower SDD at the 10°N–30°N latitude. This model can be implemented using the Google Earth Engine platform to derive SDD for NA lakes and reservoirs at annual or even seasonal time steps to assess water eutrophication variation in both time and space at the continental scale. [ABSTRACT FROM AUTHOR]

The global mining sector generates billions of tons of tailings stored in thousands of tailing ponds. The occasional spills of tailings resulting from dam failures or pipe damage can have devastating consequences, threatening nearby human populations and ecosystems, particularly those located along river corridors. Satellite remote sensing technology is a vital supplementary method to traditional field methods for monitoring and evaluating the water pollution caused by spilled tailings. The researchers have developed workflows to evaluate the effect of tailing spills on water quality using low and medium-spatial satellite imagery from satellite sensors. Due to insufficient spatial resolution, these workflows were hard to apply to monitor the water pollution caused by spilled tailing in small rivers. Using valuable on-site data from a river water pollution incident caused by spilled tailing, a workflow utilizing high-resolution satellite imagery (GF1) was developed. This workflow incorporates a machine learning algorithm (improved DeepLabV3+) to extract water masks first and a novel spectral index method to determine TSM concentrations. The improved DeepLabV3+ algorithm can obtain an accurate water mask no matter the water pixels, whether influenced by the tailing spills from GF1 imagery with IoU of 82.66%, Precision of 93.21%, Recall of 87.96%, and F1-score of 90.51%. A new spectral index combination algorithm that provides reliable TSM products for an extensive TSM magnitude range was presented to assess the level of water contamination. The strong correlation (R2 = 0.97) between in situ TSM and Mo concentrations suggests that the retrieved TSM products are suitable for assessing the water pollution caused by the spilled tailing. This workflow provides a method for monitoring and evaluating water pollution resulting from spilled tailings in small rivers. It utilizes high-resolution satellite data to observe and analyze the pollution levels. • A practical workflow to track water pollution caused by tailing spills using HSS. • An improved DeepLabV3+ algorithm to extract river water pixels accurately. • A new spectral index to retrieve TSM for evaluating river water pollution. • Retrieved TSM products were a good indicator of heavy metal water pollution. • Better temporal resolution is needed for HSS to monitor tailing water pollution. [ABSTRACT FROM AUTHOR]

Wetlands provide numerous ecosystem services essential for biodiversity conservation, climate change mitigation, and human well-being. Nevertheless, there are still no complete and uniform wetland maps providing information on the location, distribution, size, and changing status of wetlands in Africa. In this study, we used 130,000 Landsat satellite images and 270,000 sampling points evenly distributed across Africa to classify wetlands with a random forest classifier based on the Google Earth Engine (GEE) cloud platform. This resulted in the first systematic and comprehensive map of existing wetlands in Africa that quantified the area and distribution of these wetlands at a spatial resolution of 30 m, with an overall wetland producer accuracy of 90.22 % and user accuracy of 95.74 %. To address the seasonal variation in African wetlands, we used a dynamic threshold segmentation method based on the Otsu method, which can extract the extent of wetlands under different periods more accurately based on Landsat time series and African phenology data. Our results show that the total African existing wetlands extent from 2019 to 2021 (here in 2020), was approximately 1,341,500 km2, excluding shallow marine waters, comprises 3.98 % of Africa's total land area. Of these wetlands, inland wetlands account for 81 %, and coastal wetlands account for 19 %. The wetlands in Africa were symmetrically distributed around the equator and decreased with increasing latitude, accounting for 97 % of the total wetland area being within the 20°N and 20°S latitudes. African marshes were highly dynamic, expanding by 92.22 % from the dry to wet season. Our study demonstrates the substantial potential of large data samples, multiple wetland classification features (spectral-temporal, topographic, and hydrological features), and random forest classification in GEE for large-scale wetland mapping. The resulting wetland maps provide new baseline data for large-scale monitoring of seasonal dynamics and long-term trends in African wetlands. [ABSTRACT FROM AUTHOR]