Mixed tropical forests canopy height mapping from spaceborne LiDAR GEDI and multisensor imagery using machine learning models

Spatial mapping of forests canopy height (Hcanopy) provides an opportunity to assess above-ground biomass, net primary productivity, carbon dioxide (CO2) sequestration, biodiversity conservation and forest fire risks. This study incorporated a continuous coverage of multi-spectral optical and synthetic aperture radar (SAR) along with sparsely global ecosystem dynamics investigation (GEDI) spaceborne Light Detection and Ranging (LiDAR) data in the machine learning (ML) models for mapping Hcanopy in the mixed tropical forests of Shoolpaneshwar wildlife sanctuary (SWLS), Gujarat, India. We trained seven ML models, including quantile random forest (QRF), support vector machine (SVM), Bayesian regularization for feed-forward neural networks (BRNN), conditional inference random forest (Cforest), Extreme gradient boosting (Xgbtree), multivariate adaptive regression splines (MARS), and k-nearest neighbors (KNN) using GEDI_02A extracted Hcanopy as training data. We used predictors which were extracted from LiDAR (GEDI metrics), multispectral optical (Landsat-8, Sentinel-2), and SAR (ALOS-2/PALSAR-2, Sentinel-1). A 10-fold cross-validation (CV) resampling was used to avoid overfitting or underfitting. The comparison of the models performances shows that the BRNN model has the highest satisfactory accuracy metrics, such as root mean square error (RMSE) of 4.686 m, R-squared (R-2) of 0.49 and mean absolute error (MAE) of 3.66 m. Low training samples of tall canopies (> 25 m), presence of mixed vegetation, geometric and structural variability and sloppy terrain of SWLS possibly restricted models from performing well. Field validation shows an R(2 )of 0.55, satisfactory for mixed tropical forests using spaceborne LiDAR. The present work provides insights into using spaceborne LiDAR GEDI data with optical and SAR data for Hcanopy mapping through ML models, which help to manage SWLS and further implications of forest Hcanopy mapping over large spatial scales.