1. A decade-long chlorophyll-a data record in lakes across China from VIIRS observations.
- Author
-
Cao, Zhigang, Wang, Menghua, Ma, Ronghua, Zhang, Yunlin, Duan, Hongtao, Jiang, Lide, Xue, Kun, Xiong, Junfeng, and Hu, Minqi
- Subjects
- *
BODIES of water , *DEEP learning , *MACHINE learning , *LAKES , *INFRARED imaging , *REMOTE-sensing images , *FOOD chains - Abstract
Chlorophyll-a (Chl-a) is one of the optically active constituents in waters, and its concentration is frequently utilized as a proxy for lake trophic levels. However, generating a large-scale, long-term, and consistent data record of Chl-a in lakes from satellite images has been a challenging undertaking due to the limitations of conventional algorithms in monitoring inland waters spanning various optical properties. Here, we develop a practical deep neural network (DNN) model to generate a long-term Chl-a series (2012−2021) in 217 large lakes (> 50 km2) across China from the Visible Infrared Imaging Radiometer Suite (VIIRS) imagery. The assessment showed that the NOAA operational VIIRS remote sensing reflectance (R rs (λ)) products were reliable over 28 of China's examined lakes (N = 340, bias = −12%, mean absolute percentage error [MAPE] = 38%), particularly at bands ranging from the green to near-infrared domain. The DNN model performed satisfactorily on Chl-a retrievals (bias = 5%, MAPE = 32%) in 79 lakes over three orders of magnitude (0.1–300 μg L−1) spanning clear/deep to turbid/shallow waters, with significant improvements compared with the existing algorithms and other machine learning algorithms. The algorithm was applied to VIIRS images to produce a data record of spatial and temporal variations in Chl-a for China's large lakes over the past decade. The VIIRS-derived data record showed that China's lakes have an average Chl-a of 9.5 μg L−1 and are to 45.5% eutrophic. The results revealed a spatial trend of lower Chl-a in the western deep lakes than that in the eastern shallow lakes. In addition, we observed a significant increase in Chl-a in the lakes of the China East Plain but a decreasing trend of Chl-a in the Tibetan Plateau. This study highlights the feasibility of a machine learning approach based on synchronous matchups to derive Chl-a data in various lakes from satellite images. Our results provide a comprehensive understanding of overall changes to the optical conditions of China's lakes and enable scientists to elucidate the roles of climate and human activities in regulating lake productivity. • The performance of VIIRS-derived R rs was evaluated in clear and turbid lakes. • A robust deep-learning algorithm was developed for Chl-a retrievals in lakes. • The algorithm showed capability of integrating VIIRS and OLCI for Chl-a retrievals. • Chl-a variations in 217 large Chinese lakes were mapped from VIIRS images. • Chl-a increased in East China Plain but decreased in the Tibetan Plateau since 2012. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF