An enhanced deep learning approach to assessing inland water quality and the affecting factors using Landsat 8 and Sentinel-2

The estimation of water quality parameters (WQPs) using remote sensing is difficult due to the complex correlation between WQPs and water optical properties, the interactions of WQPs, and the impacts of climate. We proposed enhanced multimodal deep learning (EMDL) models for Chlorophyll-a (Chla), total phosphorous (TP), total nitrogen (TN), Secchi disk depth (SDD), dissolved organic carbon (DOC), and dissolved oxygen (DO) estimation in Lake Simcoe, Canada. The EMDL models were developed and validated using the remote sensing reflectance derived from the harmonized Landsat and Sentinel-2 images, synchronized in-situ water quality measurements, water surface temperature, and climate data (N = 950). Using the EMDL models, the spatiotemporal water quality patterns of Lake Simcoe from 2013 to 2019 were reconstructed. Besides, we quantitatively analyzed the impacts of 12 potential natural and anthropogenic factors on the water quality of Lake Simcoe. The results showed that the EMDL models had the potential to detect the spatiotemporal dynamics of water quality with the Slope being close to 1 (0.84−0.95), normalized mean absolute error ≤ 20.17%, and Bias ≤ 14.68%. Human activities such as urban development and agricultural activities mainly affected the water quality of Lake Simcoe. This study provides a practical approach to supporting the environmental management of regional inland watersheds.