LAI Estimation of Pinus densata in Northwest Yunnan Based on UAV and Sentinel–2 Data

The research focused on the typical Pinus densata natural forest in Shangri-La and utilized multi-spectral images obtained by low-altitude aerial photography using a UAV. The canopy gap rate model was applied to calculate leaf area index, and the extracted vegetation index from the Sentinel–2 image was integrated. Subsequently, an estimation model for the leaf area index of P. densata was developed using the random forest (RF), Bayesian optimization algorithm improved RF(BO–RF). At the same time, the leave one out cross-validation method was used to evaluate the fitting accuracy of the estimation model by the determination coefficient (R²), root mean square error(RMSE), prediction accuracy(P) and mean absolute error(MRE). The BO–RF model with the highest accuracy was used for LAI remote sensing inversion at the regional scale. The results showed that the mean LAI calculated by the canopy gap rate model was 4.24, and the standard deviation was 0.96 based on the multi-spectral remote sensing image of UAV. Bayesian optimization algorithm can effectively improve the estimation accuracy of the machine learning model and the R² = 0.82, RMSE=0.41, P=90.03% and MRE=8.78% of the BO–RF model. The R² increased by 20.59%, RMSE decreased by 24.07%, P increased by 2.87%, and MRE decreased by 1.78%, compared to the values before optimization. The BO–RF model was used to estimate LAI and spatial mapping in the study area. The average value was 4.25, which was mainly distributed in the range of [4,6], accounting for 63.15 %. This distribution aligns with the measured and predicted values, displaying a strong correlation coefficient of 0.75 and R² = 0.58. The overall distribution of LAI tends to be high in the center and low in the periphery, with the central and northern regions being the main distribution areas for high LAI values, and low values mainly in the southeast. The research results can provide a reference for rapid and precise remote sensing estimation of forest LAI at a large spatial scale using ultra-high resolution optical data in conjunction with spaceborne medium-resolution satellite data.