Mapping Canopy Height from ICESat-2 and Landsat-9 using Machine Learning in the Himalayan Corbett Tiger Reserve, India

Canopy height (CH) is an important parameter for better managing forests, biomass assessment and biodiversity conservation. The study's goal is the spatial mapping of CH by combining ICESat-2 and optical data information from Landsat-9 and Sentine1-2 using a support vector machine (SVM) and random forest (RF). Further, the most to least important predictors were identified for CH prediction. This assessment was performed in the Corbett Tiger reserve (CTR), Himalayan Uttrakhand state of India. The result showed that the mean CH in the CTR is 32.61 m. Root mean square error (RMSE) (5.339 m and 5.456 m), mean absolute error (MAE) (4.048 m and 4.166 m), and R-squared (R2) (0.552 and 0.531) were the optimal training values for SVM and RF, respectively. Models testing between observed and predicted CH showed the RMSE is 5.42 m and 5.53 m, MAE is 4.10 m and 4.20, and R2 is 0.55 and 0.53 for SVM and RF, respectively. Canopy profiles and metrics at percentiles height (PH) are dominant predictors. Landsat-8 derived vegetative indices (VI's) have moderate importance. Such an integrated approach is helpful in managing CTR and CH mapping of other protected forests.