Dry Weight Prediction of Wedelia trilobata and Wedelia chinensis by Using Artificial Neural Network and MultipleLinear Regression Models

Abstract

In China, Wedelia trilobata (WT) is among the top most invasive plant species. The prediction of its growth, using different efficient methods under different environmental conditions, is the optimal objective of ecological research. For this purpose, Wedelia trilobata and its native plant species Wedelia chinensis (WC) were grown in mixed cultures under different levels of submergence and eutrophication. The multiple linear regression (MLR) and artificial neural network (ANN) models were constructed, with different morphological traits as the input in order to predict dry weight as the output for both plant species. Correlation and stepwise regression analysis (SWR) were used to find the best input variables for the ANN and MLR models. Plant height, number of nodes, chlorophyll content, leaf nitrogen, number of leaves, photosynthesis, and stomatal conductance were the input variables for WC. The same variables were used for WT, with the addition of root length. A network with the Levenberg–Marquart learning algorithm, back propagation training algorithm, Sigmoid Axon transfer function, and one hidden layer, with four and six neurons for WC and WT, respectively, was created. The best ANN model for WC (7-4-1) has a coefficient of determination (R2) of 0.98, root mean square error (RMSE) of 0.003, and mean absolute error (MAE) of 0.001. On the other hand, the ANN model for WT (8-6-1) has R2 0.98, RMSE 0.018, and MAE 0.004. According to errors and coefficient of determination values, the ANN model was more accurate than the MLR one. According to the sensitivity analysis, plant height and number of nodes are the most important variables that support WT and WC growth under submergence and eutrophication conditions. This study provides us with a new method to control invasive plant species’ spread in different habitats.