Classification of inland lake water quality levels based on Sentinel-2 images using convolutional neural networks and spatiotemporal variation and driving factors of algal bloom

Water quality monitoring in inland lakes is crucial to ensuring the health and stability of aquatic ecosystems. For regional water environment agencies and researchers, remote sensing offers a cost-effective alternative to traditional in-situ water sampling methods. In this study, we designed a convolutional neural network (CNN) based on AlexNet to represent the relationship between Sentinel-2 images and in situ water quality levels of Lake Dianchi from November 2020 to April 2023. The model incorporated an algal bloom extraction algorithm and utilized correlation analysis, redundancy analysis (RDA), and random forest (RF) method to establish connections between two environmental factors: water quality and meteorology, to the area of algal bloom (AAB). The findings revealed an improvement in Lake Dianchi's water quality, with Levels A (good water quality) and B (mildly polluted water quality) averaging 1.24% and 84.28%, respectively. Starting in October 2022, water quality stabilized at Level B, averaging at 98.17%. Seasonal variations demonstrated the best water quality in spring and the worst in summer (Level C, severely polluted water quality, accounting for 5.19% and 21.68%, respectively). Algal bloom presence was minimally observed, with an average AAB value of 1.75%, peaking in autumn (4.05%) and hitting a low in winter (0.38%). A significant correlation was identified between water quality levels and AAB, with a notable spatial trend of decreasing Level C water quality and AAB from north to south, featuring lower AAB in the Southern Waihai compared to the Central Waihai. Statistical analysis pinpointed total phosphorus (TP) as the dominant factor influencing AAB, while meteorological factors such as wind speed (WS), relative humidity (RH), and precipitation (PP) playing secondary roles. Despite fluctuations in TP concentration, a recent stabilization at 0.05 mg/L suggests a positive trajectory for future algal bloom management in Lake Dianchi. •Convolutional neural networks (CNN) are used in lake water quality levels classification.•Since October 2022, the water quality of Lake Dianchi has maintained at Level B (mildly polluted).•Total phosphorus (TP) was the main influencing factor of area of algal bloom.•The algal bloom in Lake Dianchi has a good prospect of control in the future.