Research On GPP Estimation Based On Machine Learning Coupled With Mechanism Process Model

Global warming,as the most significant global environmental issue in recent years,has gradually affected various aspects of the global ecosystem,altering the balance and stability of ecosystems,thereby attracting attention from all sectors of society.Vegetation,as the most climate-sensitive group in terrestrial ecosystems,has also undergone significant changes under this circumstance.Total Gross Primary Productivity(GPP)of vegetation is one of the key indicators for measuring changes in vegetation growth activities in ecosystems and is crucial for understanding carbon cycling and ecosystem functionality.The Biome-BGC model,as the most commonly used GPP estimation model,exhibits high simulation accuracy.However,it is limited by challenges such as difficulty in collecting input data and high costs,resulting in limited application in current ecosystem research.To address this issue,this study determined the feasibility of replacing the water-heat calculation module in the Biome-BGC model with an Artificial Neural Network(ANN)model from machine learning through methods like sensitivity analysis and accuracy verification.The study establishes an ANN-Biome coupled model to estimate GPP data in the QinghaiTibet Plateau region between 2010 and 2019 and compares the accuracy and applicability of the two models before and after coupling.Subsequently,the simulated results of the models are analyzed for temporal and spatial comparisons to comprehensively determine the accuracy of the coupled model in spatial distribution.The main conclusions are as follows:(1)Through validation with measured data,the estimation accuracy of the coupled model is higher than that of the original model.The simulation results of the Biome-BGC model are poor,with an R2 of only 0.58,while the coupled ANN-Biome model achieves an R2 of 0.75.Similarly,the RMSE of the Biome-BGC model is 3.17 g C·m-2·a-1,whereas the RMSE of the ANN-Biome model is 1.75 g C·m-2·a-1.During the simulated period from 2010 to 2019,the Biome-BGC model calculates an average annual GPP of 242.7 g C·m-2·a-1 and a total of 0.615 Pg C·a-1 for the Qinghai-Tibet Plateau.The ANN-Biome model calculates an average annual GPP of 191.3 g C·m-2·a-1 and a total of 0.485 Pg C·a-1.(2)Comprehensive analysis of model usage indicates that the ANN-Biome model outperforms the Biome-BGC model in terms of simulation accuracy,model usability,input data requirements,with only slightly longer usage time compared to the Biome model.(3)In terms of spatial distribution,the estimated results of the two models exhibit similar spatial patterns.From the perspective of the entire Qinghai-Tibet Plateau,the trend of GPP shows a general decrease from east to west,gradually declining from southeast to northwest.The Biome-BGC model estimates the annual GPP in the Qinghai-Tibet Plateau to range between 0 and 3250 g C·m-2·a-1,while the ANN-Biome model estimates it to range between 0 and 2796 g C·m-2·a-1.(4)In terms of temporal variation,the two simulation models for the Qinghai-Tibet Plateau show similar trends from 2010 to 2019.Overall,vegetation productivity in the southern region exhibits an increasing trend,while the northern region shows a decreasing trend.The simulation results of the ANN-Biome model indicate that the highest rate of change is 30.24 g C·m-2·a-1.