Tan, Z, Simis, SGH, Yang, C, Shen, M, Li, J and Duan, H 2025 Revealing two decades of chlorophyll-a dynamics in arid oligotrophic lakes of Xinjiang, China using a deep recurrent approach. Water Research, 285. 124058. 10.1016/j.watres.2025.124058
Full text not available from this repository.Abstract/Summary
Oligotrophic lakes in arid and semi-arid regions are vital to local livelihoods and ecosystems, yet research has primarily focused on water levels and volumes in these areas, with limited attention to water quality. As a sensitive indicator of ecological change, chlorophyll-a (Chl-a), which is detectable via remote sensing, can provide valuable insights into subtle variations in water quality. To address this research gap, we investigated Chl-a concentrations in lakes larger than 100 km2 across the expansive Xinjiang region of China using two decades (2002–2023) of MODIS imagery. A robust deep learning model was developed to accurately estimate spatiotemporal Chl-a dynamics, overcoming the limitations of conventional algorithms for oligotrophic lakes. Unlike existing machine learning models that treat remote sensing spectra as static multi-dimensional points, our approach models wavebands as dynamic sequences with a Recurrent Neural Network (RNN) framework. A comparative evaluation with four conventional Chl-a inversion models, four machine learning and five deep learning models demonstrated the superiority of the RNN model with an RMSE of 0.75 mg m−3 and R2 of 0.72. Results over the ice-free period revealed that Chl-a concentrations in major lakes of Xinjiang generally remained below 6 mg m−3, with annual averages below 3 mg m−3, and manifested a slight declining trend in recent years. Further analysis of environmental variables revealed that Chl-a variability in lakes influenced by anthropogenic activities was primarily driven by solar radiation, vegetation cover, and runoff. In contrast, alpine lakes with minimal human disturbance exhibited dynamics largely shaped by lake expansion. This study presents a robust framework for Chl-a estimation, potentially extendable to other water parameters, and advances the understanding of long-term ecological dynamics in arid environments of Xinjiang.
Item Type: | Publication - Article |
---|---|
Additional Keywords: | Chlorophyll-aOligotrophic lakesXinjiangRecurrent neural networkMODISTime series |
Divisions: | Plymouth Marine Laboratory > National Capability categories > National Centre for Earth Observation |
Depositing User: | S Hawkins |
Date made live: | 26 Jun 2025 14:10 |
Last Modified: | 26 Jun 2025 14:10 |
URI: | https://plymsea.ac.uk/id/eprint/10435 |
Actions (login required)
![]() |
View Item |