Ying Yong Sheng Tai Xue Bao. 2022 Jan;33(1):9-16. doi: 10.13287/j.1001-9332.202201.001.
ABSTRACT
Forest biomass is an important index in forest development planning and forest resource monitoring. In order to provide a more efficient and low-biased method for estimating individual tree biomass, we introduced artificial neural network here. We used the data of aboveground biomass of 101 Larix olgensis trees harvested from the Dongzhelenghe Forest Farm in Heilongjiang Province to develop four aggregation model systems (AMS), based on different combination of the variables (diameter at breast height, tree height, crown width). The weighted functions were used to eliminate heteroscedasticity. Then, we trained artificial neural network (ANN) biomass model based on the optimal combination. The models were tested by the leave-one-out cross-validation method to compare the accuracy of the two biomass estimation methods. The results showed that biomass model based on only one variable, diameter at breast height, could accurately estimate the biomass of L. olgensis. Adding two indices, tree height and crown width, could improve the fitting performance of models, with AMS4 performing the best among the four addictive model systems. The biomass models developed by the two methods both could estimate biomass at tree level accurately, with the coefficient of determination (R2) of each component was higher than 0.87. Compared with the AMS4, R2 of leaf biomass model was about 0.05 higher, and that of other organs were also about 0.01 higher in artificial neural network model system. In addition, the root mean square error (RMSE) and other indicators were also significantly smaller. For example, the RMSE of tree stem and aboveground biomass were smaller by 2.135 kg and 3.908 kg, respectively. The model’s validation statistics mean relative error (MRE) performed better. In general, ANN was a flexible and reliable biomass estimation method, which was worthy consideration when predicting tree component biomass or aboveground biomass.
PMID:35224920 | DOI:10.13287/j.1001-9332.202201.001
