ESTIMASI BIOMASSA VEGETASI KEBUN RAYA BOGOR MENGGUNAKAN KOMBINASI CITRA WORLDVIEW-2 DAN ALGORITMA PEMELAJARAN MESIN

Abstract

An effective and efficient measurement method is required for estimating the vegetation biomass of an area with high canopy density. The combination of WorldView-2 imagery and machine learning algorithms can be an alternative approach for estimating vegetation biomass in Bogor Botanic Gardens. The research aimed to determine the variables from WorldView-2 imagery that can be used to estimate the vegetation biomass of Bogor Botanic Gardens, to identity several types of machine learning algorithms that produce the best prediction in estimating vegetation biomass in the field, to estimate and to map vegetation biomass in Bogor Botanic Gardens. The variables that had a significant correlation with biomass were NIR-reflectance, Blue-Correlation, Green-Correlation, NIR-Mean, and NIR-Variance. The NIR-Mean variable was the most important variable for estimating vegetation biomass. The random forest algorithm produced the best model for estimating vegetation biomass with r, PBIAS, RMSE, MAE and RSR values of 0,83, -11,51%, 185,47 Mg/ha, 139,43 Mg/ha, and 0,56 respectively. The estimated vegetation biomass of the Bogor Botanic Gardens had a range from 6,27 to 1.576,90 Mg/ha with an average of 183,96 Mg/ha and total biomass of 13,23 Gg. The combination of WorldView-2 imagery and the Random Forest algorithm produced a good predictive model compared to Artificial Neural Network and Support Vector Machine for estimating the vegetation biomass of Bogor Botanic Gardens. Bogor Botanic Gardens has a very important role in climate change mitigation, especially for the Bogor City.