Trunk volume estimation of irregular shaped Populus euphratica riparian forest using TLS point cloud data and multivariate prediction models

Background: Trunk volume (Vt) is an essential parameter for estimating forest stand volume, biomass, and carbon sequestration potential. As the dominant tree species in desert riparian forests, Euphrates poplar (Populus euphratica) has a high proportion of irregularly shaped tree trunks along the Tarim River, NW China, where the habitat is very fragile owing to long-term water stress. This causes uncertainty in estimation accuracy as well as technical challenges for forest surveys. Our study aimed to acquire P. euphratica Vt using terrestrial laser scanning (TLS) and to establish a species-specific Vt prediction model.Methods: A total of 240 individual trees were measured by TLS multiple-station in 12 sampling plots in three sections along the lower reaches of the Tarim River. Vt was calculated by a definite integration method using trunk diameters (Di) at every 0.1-m tree height obtained from TLS, and all data were split randomly into two sets: 70% of data were used to estimate the model parameter calibration, and the remaining 30% were used for model validation. Sixteen widely used candidate tree Vt estimation models were fitted to the TLS-measured Vt and tree structural parameter data, including tree height (H), diameter at breast height (DBH), and basal diameter (BD). All model performances were evaluated and compared by the statistical parameters of determination coefficient (R2), root mean square error (RMSE), Bayesian information criterion (BIC), mean prediction error (ME), mean absolute error (MAE), and modeling efficiency (EF), and accordingly the best model was selected.Results: TLS point cloud reflection intensity (RI) has advantageous in the extraction of data from irregular tree trunk structures. The P. euphratica tree Vt values showed obvious differences at the same tree height (H). There was no significant correlation between Vt and H (R2 = 0.11, P < 0.01), which reflected the irregularity of P. euphratica trunk shape in the study area. Among all the models, model (14): Vt = 0 909DBH1 184H0 487BD0 836 (R2 = 0.97, RMSE = 0.14) had the best prediction capability for irregularly shaped Vt with the highest R2, BIC (-37.96), and EF (0.96), and produced a smaller ME (0.006) and MAE (1.177) compared to other models. The prediction accuracy was 93.18%.Conclusions: TLS point cloud RI has a potential for nondestructively measuring irregularly shaped trunk structures of P. euphratica and developed Vt prediction models. The multivariate models more effectively predicted Vt for irregularly shaped trees compared to one-way and general volume models.