Optimization of a soil type prediction method based on the deep learning model and vegetation characteristics

Abstract

The structure and composition of forest vegetation plays an important role in different ecosystem functions and services. This study aimed to identifying soil types based on vegetation characteristics using a deep learning model in the High Conservation Value (HCV) area of Central Kalimantan, spanning 632.04 hectares. The data on vegetation were collected using a combination method between line transect and quadratic plots were placed. The development of a deep learning model was based on the results of a vegetation survey and the processing of aerial photos using the Feature Classifier method. The results of applying a deep learning model could provide a relatively accurate and consistent prediction in identifying soil types (Entisols 62%, Spodosols 90%, Ultisols 90% accuracy). The composition of vegetation community in Ultisols was dominated of seedling and tree (closed canopy), meanwhile in Entisols and Spodosols was dominated of seedling and sapling (dominantly open canopy). Ultisols exhibited the highest species richness (57 species), followed by Spodosols (31 species) and Entisols (14 species). Ultisols, Entisols, and Spodosols displayed even species distribution(J' close to 1) without dominance of certain species(D < 0.5). The species diversity index was at a low to moderate level(H' < 3), while the species richness index remained at a very low level(D_mg > 3.5).