Geo-ML Enabled Above Ground Biomass and Carbon Estimation for Urban Forests

The study of the carbon cycle and climate change in the worldwide terrestrial ecosystem relies heavily on forest aboveground biomass (AGB). Remote sensing based AGB estimate is an excellent solution for regional scale. Urban trees play an important part in carbon cycling because they sequester carbon. Above Ground Biomass (AGB) quantification is thus critical for better understanding the function of urban trees in carbon sequestration. This work was carried out for geospatial modelling of AGB and carbon with the aid of field-based data and their correlations with spectra and textural variables derived from Landsat-8 OLI data for urban forests in Jodhpur city, Rajasthan, India, using a Random Forest (RF) based machine learning (ML) approach. A total of 198 variables were retrieved from the satellite image including bands, Vegetation Indices (VIs), linearly transformed variables and Grey Level Co-occurrence textures (GLCM) were taken as independent input variables for RF regression. The RF regression model has been evaluated independently for spectral and texture variables and their integration together. Best prediction accuracy noted for the integrated model. Using RF regression a Boruta feature selection method has been applied to extract important variables from the list of variables to get a more accurate prediction. A total number of 18 variables have been identified as most important for the integrated model. Highest considerable accuracy given by the integrated model and values noted were R2 of 0.83, MAE of 11.86 t/ha and RMSE of 16.22 t/ha while for individual bands R2 value of 0.69, MAE of 16.37 t/ha and RMSE of 22.20 t/ha. For indices R2 value noted was 0.81, MAE value of 13.17 t/ha, RMSE of 17.27 t/ha. For individual textures R2 of 0.71, MAE value of 16.56 t/ha and RMSE of 21.67 t/ha have been noted. Results of the study indicate the potential efficiency of RF regression algorithm for modelling AGB and assessing the role of urban forests for carbon sequestration.