基于遥感与集成学习的洞庭湖CO2通量估算及时空分布研究OA

Lake carbon cycling plays a pivotal role in the global carbon cycle,and accurate estimation of the water-air interface CO2 flux(F-CO2)will provide technical support for assessing dynamic changes in carbon sources and sinks.In this study on Dongting Lake,water sample data collected from 104 locations during May and August 2024 was matched with remote sensing data.The performance of four base mod-els(RF,GBR,XGBoost,and SVR)and four ensemble models was then evaluated for CO2 concentration(c-CO2)inversion for Dongting Lake.On this basis,we quantitatively estimated c-CO and analyzed the spatiotemporal distribution of F-CO2 in Dongting Lake.Compared to direct inversion using band com-binations,indirect inversion that includes the measured parameters Secchi depth(ZSD)and water tempera-ture(WT)improved c-CO2 inversion in Dongting Lake.The proposed multi-model ensemble approach in this study(integrating XGBoost,GBR,SVR,RF)was identified as optimal for c-CO2 inversion,achiev-ing an R2 of 0.72,MSE of 43.40 µmol/L,RMSE of 6.59 µmol/L,and MAPE of 12.61%.Compared to the best-performing base model in indirect inversion,this ensemble model improved R² by 26.32%and reduced RMSE by 36.94%,significantly enhancing estimation accuracy under complex hydrological conditions.F-CO2 in Dongting Lake exhibited significant spatiotemporal heterogeneity(P<0.001).The highest F-CO2 occurred in the eastern mainstream of East Dongting Lake[(108.70±19.63)mmol/(m2·d)]in spring and was 1.91 times that of the lowest area(West Dongting Lake:[(57.02±9.37)mmol/(m2·d)].The average values of F-CO2 in spring[(79.46±24.05)mmol/(m2·d)]and summer[(63.20±13.41)mmol/(m2·d)]were significantly higher than those in autumn[(25.92±7.19 mmol/(m2·d)]and winter[(25.71±7.73 mmol/(m2·d)],revealing the effects of hydrological and seasonal variations on lake F-CO2.Addition-ally,Dongting Lake acts as a carbon source across all seasons,but with potential for shifting to a carbon sink in certain areas during winter.