Simulation Of Forest Evapotranspiration Based On Attention-LSTM Model

Forest evapotranspiration is one of the important components of terrestrial evapotranspiration,and it is also the key link of hydrological cycle and energy cycle in forest ecosystem.Accurate simulation of forest evapotranspiration is of great significance for the management and effective utilization of forest water resources and the sustainable development of forestry.This paper takes the Luya Mountain Nature Reserve in Shanxi Province as the study area.The purpose of this paper is to explore the suitable artificial intelligence models for forest evapotranspiration simulation,and to investigate the impact of different environmental factors on the simulation.The main conclusions are as follows:(1)In the study area,air temperature(Ta),water vapor pressure(Pva),wind speed(WS),net radiation(Rn)and soil heat flux(SHF)were positively correlated with evapotranspiration(ET),and Rn had the highest correlation with ET.Relative humidity(RH)is negatively correlated with ET.The correlations between various environmental factors and evapotranspiration from high to bottom were Rn,Ta,SHF,RH,WS and Pva.(2)In the simulation of forest evapotranspiration,Root Mean Squared Error(RMSE)and Mean Absolute Deviation(MAE)of Long Short-Term Memory(LSTM)are 0.137 and 0.073respectively,and coefficient of determination(R~2)is 0.735.Compared with the Support Vector Machines(SVM)model and the Random Forest(RF)model,the LSTM model can effectively improve the simulation accuracy,and has certain advantages in the simulation analysis of time series problems,which is more suitable for the simulation of forest evapotranspiration.(3)The RMSE and MAE of using CNN-LSTM model of Convolutional Neural Network(CNN)to simulate forest evapotranspiration are 0.116 and 0.059 respectively,and R~2 is 0.813.The RMSE,MAE and R~2 of Attention-LSTM model of Attention mechanism(Attention)were0.110,0.055 and 0.832,respectively.The convolutional structure can effectively extract the characteristic information from the input parameters,which is helpful to improve the simulation accuracy of forest evapotranspiration simulated by LSTM model.However,CNN will lose part of the time series information of the input parameters.But the time attention and feature attention of the attention mechanism can fully dig out feature information and highlight key parameters,and maintaining the timing characteristics of the input parameters,which can more effectively simulate the evaporation of the forest.At the same time,the Attention-LSTM model shows high accuracy in the daily evapotranspiration data set of Taikuanhe Nature Reserve in Shanxi Province,which verifies the universality of the model in forest evapotranspiration simulation.(4)The model accuracy of six environmental factors separately simulating forest evapotranspiration from high to low was Rn,Ta,RH,SHF,WS and Pva.However,the accuracy of the single environmental factor model is not high enough to accurately simulate evapotranspiration.When another environmental variable was introduced,each environmental factor played a positive role in the simulation of forest evapotranspiration.Six environmental factors were introduced to improve the simulation accuracy in order of RH,Ta,Rn,SHF,WS and Pva from high to low.When four environmental factors are input,the model can achieve a relatively reliable accuracy.When Ta,RH and Rn are included in the four environmental factors,the simulation effect of the model is better than that of the SVM model,RF model and LSTM model with all environmental factors.Three environmental factors,Rn,Ta and RH,play a very important role in the simulation of forest evapotranspiration.In the absence of some environmental factors,four environmental factors including these three factors can be used as input parameters of the model to simulate forest evapotranspiration.The research results of this paper have certain practical significance for forest evapotranspiration simulation in Shanxi Province.The Attention-LSTM model constructed can be used as a reliable model for forest evapotranspiration simulation.The analysis of environmental factors affecting forest evapotranspiration can provide some theoretical reference for the study of forest evapotranspiration mechanism.Contributes to a more efficient use of forest ecosystem services and functions.