Estimation of above-ground biomass in tropical afro-montane forest using Sentinel-2 derived indices

Abstract

Empirical analyses were common methods for forest biomass estimation. Lately, satellite images are popularly used to study different attributes of forest vegetation. Sentinel-2 image provides a significant improvement in spectral coverage, spatial resolution and temporal frequency in assessing forest biomass. This study examined the potential use of multispectral (MS) bands, vegetation indices and biophysical variables derived from Sentinel-2 images in modeling above-ground biomass (AGB) in tropical afro-montane forest of the Yayu biosphere reserve. A coupled method of remote sensing and statistics was applied to establish a biomass estimation model using spectral data generated from Sentinel-2 image and AGB data measured from the field. Multispectral bands, vegetation indices and biophysical variables were extracted from the Sentinel-2 image. Forest stand parameters such as DBH and tree height were measured from sampling plots to calculate AGB using allometric equations. The strength of correlation between the measured biomass and the MS bands, indices and biophysical variables were examined using Pearson’s product-moment correlation coefficients. A regression analysis was iteratively applied to identify the determinant variables for predicting AGB. The prediction results were validated based on the magnitude of coefficients of determination between the observed and the predicted values and the magnitude of the Root Mean Square Error (RMSE). A strong correlation (r ranging from 0.65 to 0.74) was observed between the biophysical variables from Sentinel-2 image and the measured AGB from the field. The MS Band 4 (red band), vegetation variables LAI, FCOVER and FAPAR, and band combination index IRECI yielded better results and are good predictor variables for forest AGB. The model goodness of fit between the observed and predicted AGB showed a coefficient of determination (r2) of 0.74 and RMSE of 0.16 ton C/pixel, which shows strong performance of the prediction model. Vegetation indices derived from Sentinel-2 imagery are good predictors of AGB in tropical afro-montane forests. Sentinel-2 image has improved the reliability of biomass estimation from remotely sensed data. Since field sampling plots were few in this study, the level of accuracy will likely improve with more number of field sample measurements.