| Summary: | This study explores the potential of TDX InSAR data from 2011, 2017, and 2019 for estimating and mapping canopy heights in unique forest and plantations landscape in Sabah, Malaysian Borneo. The findings offer crucial insights for sustainable forest and plantation management. The methodology encompassed the SINC forest height inversion model and two machine learning (ML) models Random Forest (RF) and Symbolic Regression (SR) augmented with diverse predictor variables and height references. Training the ML models with 70% of ICESat-2 ATL08 data and validating with the remaining 30%, we achieved an out-of-bag (OOB) RMSE of 5.4 m for RF and 5.96 m for SR. The overall validation RMSEs were 6.06 m (2011 SR), 10.36 m (2017 SR), and 7.58 m (2019 RF). For specific LULC classes, accuracies ranged from 3.92 m (2011 Mangrove RF) to 6.11 m (2017 Mangrove SR) and 4.35 m (2019 Rubber RF). Field inventory data validation in 2011 and 2019 yielded RMSEs between 4.06 m and 8.69 m, with SR as the top-performing model. Spatial distribution and canopy height classes revealed non-uniform variations in 2011, with SINC overestimating. In contrast, 2017 and 2019 showed uniform height patterns, indicating an increase in canopy heights across forest and plantation LULC, particularly in the 15–20 m range for oil palm, secondary forest, acacia mangium, and rubber. Our findings highlight the potential of InSAR-based canopy height estimation and mapping for tropical forest and plantations, which also can be applied to other areas at local scales considering the LULC landscapes dynamics. © 2024, Indian Society of Remote Sensing.
|
|---|