Modeling Aboveground Biomass Using Non-Redundant Vegetation Indices from PlanetScope Imagery via Multiple Linear Regression in Planted Forests Sarono (a*), Erna Kurniati (a)
a) Faculty of Geography, Gadjah Mada University, Yogyakarta,Indonesia
*sarono90[at]mail.ugm.ac.id
Abstract
Aboveground biomass (AGB) estimation using single Vegetation Indices (VI) derived from satellite imagery has been widely implemented, yet often yields varying levels of accuracy due to limitations in representing vegetation heterogeneity. This study aims to develop an AGB estimation model based on a combination of non redundant vegetation indices derived from high resolution PlanetScope imagery in the Wanagama Forest area, Yogyakarta, Indonesia. A total of 19 VIs were calculated and analyzed using Pearson Correlation Matrix (PCM) on 404 correlation sample points. Index combinations were selected based on low inter-index correlation values (R) under 0.3, assuming that lower correlation represents more diverse spectral information while minimizing redundancy. The best combinations identified were EGCV (EVI,GNDVI,CIVE,VREI1) and EGGV (EVI,GNDVI,GLI,VREI1). Biomass estimation was performed through seven modeling scenarios: five single-index models and two combination models (EGCV and EGGV). A Multiple Linear Regression (MLR) model was applied for the combination schemes, while Simple Linear Regression (SLR) was used for the single-index models. The models were trained using 137 training samples and validated with 63 test samples derived from field-measured tree diameter and height. Results show that total biomass estimation across the seven scenarios ranged from 20.9 to 3,562 tons, with the highest correlation value 0.56 obtained from the EGGV model. The EGGV model outperformed both the EGCV combination and all single index models, which only achieved R values ranging from 0.17 to 0.55. RMSE, MAE, and R values were consistently aligned, confirming that the EGGV combination model provided the most accurate results. This study demonstrates that selecting VIs based on PCM analysis can improve AGB estimation accuracy and minimize spectral information redundancy in high resolution remote sensing applications.
