Single-tree biomass modeling of Pinus massoniana based on BP neural network

The purpose of this study was to explore and verify the applicability of BP neural network model on the singletree biomass estimation for Pinus massoniana. The optimal model had been built after the topology was determined through screening 12 algorithms and choosing the number of inputs, outputs and hidden nodes. To explain the impact of input variable number on the accuracy, double input BP model was compared with single one. Also，to explain the impact of output variable number on the accuracy, multiple output BP model was compared with single one. And to verify the feasibility, the optimal BP model was compared with allometric equation. The results showed that: 1) the algorithm of optimal model LM-DH-8-WtWaWr was LevenbergMarquardt algorithm，with DBH and height as input variables, total weight, weight of above ground and weight of root as output variables, and the number of hidden nodes was 8 2) Adding input and output variables would not decrease the accuracy of BP neural network model. 3) The optimal BP model LM-DH-8-WtWaWr had a good performance in estimating the biomass of P. massoniana and its accuracy was higher than the relative growth model. The BP model can be used to estimate several quantities at once, which makes the estimation of single-tree biomass more simply.