Evaluating the accuracy of spectral indices from Sentinel-2 data for estimating forest biomass in urban areas of the tropical savanna

Over the past decades, there has been significant challenges in the detection and quantification of urban forest biomass due to limitation of traditional methods (e.g. Field survey) which are not only costly but also impractical to monitor large and inaccessible areas. The application of accurate remote sensing data has greatly improved urban biomass estimation and modelling approaches due to development in new satellite sensors to address urban climate dynamics. For example, the launch of Sentinel-2 by the Copernicus programme of the European Union, which provides data down to 10 m, has been quite remarkable. In this study, four vegetation indices (VIs) were derived from the Sentinel-2 data for 2016 (the red and near-infrared bands (10 m)) which were atmospherically corrected by using the "Sen2Cor" tool. The sensitivities of the four VIs to biomass (normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), difference vegetation index (DVI) and enhanced vegetation index (EVI2) were assessed through a predictive modelling approach using thirty sixty (36) plots belonging to eleven (11) different tree species sampled at different locations along the major streets in the study area. All VIs significantly correlate with the field estimated biomass (p < 0.0001) using Pearson's correlation. Based on the correlation between the VIs and field measured biomass, we established a simple linear model using the 64% of the field data for model estimation while retaining the remaining 36% for evaluating the model accuracy. The assessments of model performance using the mean absolute error (NDVI = 8.91%, SAVI = 8.92%, EVI2 = 8.94%) indicated that the DVI (9.43%) had the largest error. Urban biomass maps are required for urban vegetation monitoring, urban cool/heat island modelling, climate change mitigation and adaptation research and model parameterization and validation.