TROPICAL FOREST BIOMASS ESTIMATION AND MAPPING USING K-NEAREST NEIGHBOUR (KNN) METHOD

Estimation and mapping of tropical forest biomass is important for periodic carbon accounting, as tropical deforestation is one of the major sources of terrestrial carbon emission in the recent decades. K-nearest neighbour (kNN) method is recently introduced for the estimation of boreal and temperate forest variables from satellite sensors and sample based inventory data. The current study is a first attempt to extend its application to the tropical forest regions. The number of neighbours and feature weighting parameters in the kNN estimation procedure were varied to obtain the optimal precision. The study area was located in the tropical evergreen and semi-evergreen forests of south-eastern Bangladesh. Orthorectified Landsat ETM+ satellite imagery was procured from United States Geological Survey. Atmospheric effect on the image was removed using appropriate correction procedure and digital numbers (DNs) were converted to surface reflectance. Seventy sample plots were laid out in the forests. Diameter at breast height (dbh) and heights of all the trees inside the sample plots were measured and later converted to biomass using allometric relations. Forest biomass map was prepared using kNN method entering the optimal parameters and validation was performed using another thirty sample plots. The method is finally recommended for estimation and wall-to-wall mapping of tropical forest biomass