Study On Spatio-temporal Characteristics And Prediction Of NEP Index In Vegetation Ecosystem Of Yangtze River Economic Belt

Vegetation has powerful ecological service functions.Not only does it serve as a critical element in terrestrial ecosystems,but it also plays a vital function in regulating the dynamic equilibrium of global carbon elements and preserving biodiversity.The development and changes of vegetation phenology not only affect the size of plant biomass,but also impact the concentration of CO₂in the atmosphere worldwide,which have important implications for global climate change and carbon cycling.Vegetation ecosystems possess a noteworthy capacity for carbon sequestration,with nearly all of the terrestrial ecosystem’s carbon reserve deriving from these systems.As such,they play a crucial role in the examination of global climate change and the cycling of carbon on a global scale.As one of the highest developed economic entities in China,the Yangtze River Economic Belt has gradually expanded in size in recent years.Along with this expansion,the problem of carbon emissions has also increased.However,the Yangtze River Economic Belt has abundant vegetation resources and is an important ecological security barrier for the sustainable development of the country.Examining the carbon source/sink status of the vegetation ecosystem in the Yangtze River Economic Belt can offer scientific insight into safeguarding,utilizing,and revitalizing this vital ecological system.This study took the vegetation ecosystem in the Yangtze River Economic Belt as the research area,and used MODIS data and meteorological data to estimate the NEP index to reflect the carbon source/sink situation of the vegetation ecosystem and to explore the spatio-temporal characteristics of NEP.Based on relevant literature under the condition of not considering the impact of human activities,11 environmental factors were selected.This study used a random forest feature selection model and Spearman correlation coefficient method to screen the environmental factors,and determine longitude,latitude,normalized vegetation index,terrain undulation,temperature,and precipitation as the input variables of the NEP prediction model for the vegetation ecosystem in the Yangtze River Economic Belt.To overcome the limitations of traditional prediction models in forecasting the NEP index,this research constructed three prediction models:a random forest prediction model enhanced by a genetic algorithm,a lightweight gradient boosting machine prediction model optimized by a particle swarm optimization algorithm,and an extreme gradient boosting prediction model optimized by a simulated annealing algorithm.This study compared and analyzed the prediction results of five models,including linear regression prediction models and Lasso regression prediction models,based on four indicators:mean squared error,mean absolute error,root mean squared error and R².The results showed that the machine learning prediction models are superior to the traditional prediction models,and the extreme gradient boosting prediction model optimized by simulated annealing algorithm has the smallest root mean squared error and the closest prediction results to the true values.