Dynamic Forest Management Plan Selection and Optimization Based on Improved NLP, LSTM, and XGBoost

Abstract To account for both economic benefits and carbon sequestration, we use a carbon sequestration prediction model to quantify and predict economic benefits. Then, we present the model to predict the economic benefits of the different forest management plans collected, based on which the best forest management plan is selected and optimized. To achieve the balance, we use three types of models in this paper. Firstly, we collect the current state-of-the-art forest management plans using natural language processing. Secondly, we compare and optimize the SAEs model from LSTM, GRU, and SAEs. The best R2 reached 0.952707 with optimizing. Finally, we compare four ML models, which are Artificial Neural Network (ANN), XGBoost, Random Forest (RF), and Logistic Regression (LR), and decided to model using XGBoost. Constrained by the principles of financial economics, we optimized the XGBoost model, and based on the model, we found the top 5 factors and calculated the carbon sequestration to find the optimal forest management model and felling method. The felling rate should be 71.29%, the best balance between economic benefits and carbon sequestration. According to the three models trained in this paper, in the future 100 years, about 195.950 million tons of carbon will be stored by each forest with an area size of 10,000 km2.