Key points are not available for this paper at this time.
Lakes in monsoon-dominated regions are highly vulnerable to climate change and eutrophication. Chlorophyll-a, a measure of phytoplankton biomass, is a critical indicator for detecting changes in trophic state. Readily available satellite images combined with machine learning techniques can enable long-term monitoring of chlorophyll-a in lakes. We evaluated 24 combinations of models and satellite images for Phewa Lake, Nepal (eight algorithms across three satellite combinations). An ensemble learning model combining a Support Vector Regression (SVR) and Random Forest (RF) based on Sentinel-2 imagery achieved the best relative performance amongst the tested models, although overall predictive accuracy was moderate. Although microwave imagery from Sentinel-1 can penetrate clouds, and therefore provide continuous monitoring during periods of persistent cloud cover, Sentinel-2 achieved higher accuracy (MAE = 0.2 mg/m 3 ), due to the availability of high spectral resolution images and red-edge sensitivity. Analysis of Sentinel-2 images of Phewa Lake from 2018 to 2024 revealed relative seasonal patterns of chlorophyll-a consistent with limnological processes, with relatively higher concentrations during post-monsoon than other seasons. Model-generated maps showed relatively homogeneous spatial distributions of chlorophyll-a in post-monsoon, winter, and pre-monsoon, but highly heterogeneous and dynamic spatial patterns during monsoon, a season of high inflows and mixing. Remote sensing combined with machine learning offers a low-cost and scalable approach for freshwater monitoring that is particularly valuable in monsoonal and low-income countries. In Nepal, which has more than 5,000 lakes, such approaches have strong potential for national-scale monitoring and management. An effort to implement and validate machine learning models in other lakes can be beneficial for sustainable monitoring.
Joshi et al. (Mon,) studied this question.