Remote sensing of total suspended matter concentration in lakes across China using Landsat images and Google Earth Engine

Abstract

Total suspended matter (TSM) has a crucial impact on light propagation in the water column, and often co-varies with nutrients, heavy metal and micropollutant fluxes. The objective of this study was to explore the feasibility of mapping TSM in lakes (area > 1 ha) across China using Landsat surface reflectance product embedded in Google Earth Engine. We conducted 44 sampling campaigns at 423 natural and manmade lakes across China, and determined TSM from 2036 water samples collected during 2011–2020. Landsat surface reflectance was matched with water sampling events within ±7 days of satellite overpasses, yielding 1908 matched pairs. We divided the TSM dataset into nine subsets, with eight subsets dedicated to building of simple regression and Random Forest (RF) models, and the remaining subset used to validate model performance. Regression analysis indicated strong associations between TSM concentration and both the Red band (Band 4; R2 = 0.76, RMSE = 21.4 mg/L) and the suspended matter index [(Nir + Red)/2); R2 = 0.71, RMSE = 17.1 mg/L]. The RF model outperformed the Red band model as indicated by higher coefficient of determination for model calibration (R2 = 0.95) and validation (R2 = 0.81). We also assembled samples (N = 1703) from lakes in different continents, and both the RF (RMSE = 16.04 mg/L) and Red band (RMSE = 18.9 mg/L) models exhibited acceptable performance. The models were further evaluated using a time series of TSM records from six large lakes in the USA and Japan (in situ TSM collected during 1988–2018). Finally, we used 460 scenes of Landsat/OLI images mainly acquired in 2019 to map TSM in lakes across China. Results showed considerable regional variability (TSM range: 0.12–860 mg/L), with lakes in Northeast China (44.6 mg/L), East China (38.8 mg/L) and Xinjiang-Inner Mongolia (39.7 mg/L) exhibiting much higher TSM concentration than lakes in the Yungui Plateau (17.62 mg/L) and the Tibetan Plateau (5.31 mg/L). Results confirmed the stability and spatial transferability of both models. Given the relationships among TSM and various indicators (e.g. nutrients, water clarity, chlorophyll-a) of the trophic state of aquatic ecosystems, these models would greatly facilitate TSM monitoring in lakes, and provide water resources managers with additional tools to assess the impact of water protection measures.