Abstract:Based on the GF-1 satellite images effectively observed in 2018 and 2019 and the chlorophyll-a concentration data in-situ observed on the lake surface, a random forest machine learning algorithm is used to quantitatively evaluate the importance measures and contribution rate of the band reflectance and select effective feature band combinations. Then a remote sensing inversion model of chlorophyll-a concentration in Taihu Lake based on in-situ automatic monitoring data is established in this paper. The results show that the green light band (0.52-0.59 μm) and the red light band (0.63-0.69 μm) are the key bands, which can be combined with other bands to estimate chlorophyll-a concentration. It is better to construct the estimation model of chlorophyll-a concentration in Taihu Lake by seasons, and the determination coefficients R2 of the spring, summer, autumn, and winter models are 0.84, 0.85, 0.96, and 0.82, respectively. The concentration of chlorophyll-a in Taihu Lake is highest in summer, followed by autumn and spring, and lowest in winter. The spatial changes of chlorophyll-a concentration in spring, autumn and summer are more obvious, while that in winter is not obvious. The areas with high chlorophyll-a concentration are mainly concentrated in the western coastal area, Zhushan Lake, Meiliang Lake and some lake core areas. Studies have shown that the random forest model can objectively determine the effective bands for chlorophyll-a concentration inversion, and achieve high-precision estimation of chlorophyll-a concentration in large inland water bodies.