| Based on the effective accumulated temperature,the prediction model of plant height,stem diameter,LAI and aboveground biomass of Xinluzhong 81 cotton and the prediction model of plant height,stem diameter and LAI of Tahe 2 cotton were established,in order to provide theoretical basis for using effective accumulated temperature to simulate the growth and development dynamics of cotton in Alar reclamation area.The plant height,stem diameter,LAI,aboveground biomass and effective accumulated temperature were normalized to establish a traditional regression model.At the same time,the machine learning method was applied to the cotton growth simulation and compared with the traditional regression model.The main research of this paper is as follows:(1)Using the traditional regression model method,three theoretical equations of Logistic,Gompertz and Richards were used to construct models for plant height,stem diameter,LAI,aboveground biomass of Xinluzhong 81 cotton and plant height,stem diameter and LAI of Tahe 2 cotton.The results showed that the relative superior plant height model of Xinluzhong 81 was Logistic,with R~2=0.9912 and RMSE=0.0321.The relative superior plant height model of Tahe 2 was Richards,with R~2=0.9922 and RMSE=0.0294.The relative superior stem diameter model of Xinluzhong 81 was Logistic,with R~2=0.9690,RMSE=0.0556,and the relative superior stem diameter model of Tahe 2 was Richards,with R~2=0.9426,RMSE=0.0685.The relative optimal LAI model of Xinluzhong 81 was Richards,R~2=0.9904,RMSE=0.0358,and the relative optimal LAI model of Tahe 2 was Gompertz,R~2=0.9912,RMSE=0.0338.The relative superior leaf biomass model of Xinluzhong 81 was Logistic,R~2=0.9480,RMSE=0.0745;the relative optimal stem biomass model of Xinluzhong 81 was Logistic,R~2=0.8719,RMSE=0.1174;the biomass model of Xinluzhong 81 was Logistic,R~2=0.9589,RMSE=0.0706.(2)The decision tree machine learning method based on CART was used to construct the models of plant height,stem diameter,LAI,aboveground biomass of Xinluzhong 81 cotton and Tahe 2 cotton by using CART tree,random forest and gradient boosting decision tree.The results show that among the three decision tree models,the ensemble learning model is generally higher than the non-ensemble learning model,and the random forest model performs relatively well.The R~2 of the Xinluzhong 81 plant height random forest model was 0.9963,which was 0.51%higher than the logistic model in the traditional regression model,and the MAE on the test set was 3.3424.The R~2 of the Tahe 2 plant height random forest model was 0.9947,which was 0.25%higher than the Richards model in the traditional regression model,and the MAE on the test set was 3.5161.Among them,the R~2 of Xinluzhong 81 stem diameter random forest model is 0.9901,which is 2.18%higher than the logistic model in the traditional regression model,and the MAE on the test set is 0.5316.The R~2 of Tahe 2 stem diameter random forest model is 0.9713,which is 3.04%higher than the logistic model in the traditional regression model,and the MAE on the test set is 0.4930.The R~2of Xinluzhong 81 LAI random forest model is 0.9974,which is 0.71%higher than the Richards model in the traditional regression model,and the MAE on the test set is 0.0498.The R~2 of Tahe 2LAI random forest model is 0.9979,which is 0.68%higher than the Gompertz model in the traditional regression model,and the MAE on the test set is 0.1050.The R~2 of the Xinluzhong 81 leaf biomass random forest model was 0.9960,which was 5.06%higher than the logistic model in the traditional regression model,and the MAE on the test set was 1.0400.The R~2 of the Xinluzhong 81 stem biomass random forest model was 0.9985,which was 14.52%higher than the logistic model in the traditional regression model,and the MAE on the test set was 0.1930.Among them,the R~2 of the random forest model of cotton boll biomass of Xinluzhong 81 was 0.9981,which was 4.09%higher than the logistic model in the traditional regression model,and the MAE on the test set was 2.1789.The experimental study found that the random forest algorithm in the machine learning method is superior to the traditional regression and the other two machine learning models in prediction performance,which provides a new idea for predicting the growth and development dynamics of cotton based on effective accumulated temperature. |
PDF Full Text Request