Weight Analysis Of Evapotranspiration Response Factors Based On Machine Learning

Evapotranspiration is the total water vapor flux transported to the atmosphere by vegetation and the whole ground.This dissertation focuses on the correlation of evapotranspiration response factors,using machine learning model to replace traditional statistical methods.design experiments to explore the response of evaporation factor importance ranking.The latent heat flux is obtained by evapotranspiration multiplied by the latent heat constant of water gasification,so this experiment needs to conduct variable screening,outlier elimination,normalization and standardization operations on the data measured from the flux tower in Yanchi county,Ningxia from 2012 to 2017(for some algorithms).and then respectively set up RF model,XGBoost model,SVR model,ANN model,selects the determination coefficient,MSE and MAE as evaluation coefficients to optimize the model,Adjust the parameters to optimize the model.And then select the effect of the optimal model to analyze its response factor importance,correlation ranking of response factors by correlation coefficient method.Finally,according to data for every year and every two years,we build a model to analyze the impact of the amount of data on the results.The experimental results show that the random forest model and XGBoost model are the optimal models,and their decision coefficients are between 0.84 and 0.85,MAE is 7.3726 and 7.4799,MSE is 279.2584 and 261.1653,respectively.In the process of assessing the importance of response factors,the rank of photosynthetic effective radiation and net radiation of the upper layer always ranked the top two.The rank of response factor score of 7 indexes,such as CO2 flux and water vapor pressure deficit,varied to a certain extent under different circumstances,but all of them ranked within 11(22 indexes in total).Other indexes ranked low and varied.In the process of data volume analysis,the results of data analysis every two years are in the same trend with the overall results,while the results obtained from data every year are different from the overall results.In the process of analyzing the correlation between the data,the rank of carbon dioxide flux increased with the enhancement of light,and the influence of air temperature also increased significantly with the deletion of night data.The result is the following:(1)In general,the light factor was greater than the temperature factor was greater than the light factor.(2)At least two years of data should be selected in the process of importance analysis of response factors,otherwise,it will affect the experimental results.(3)Carbon dioxide flux and air temperature have a certain correlation with light conditions.