Identifikasi Sebaran Spasial Hutan Mangrove Berbasis Algoritma Machine Learning dengan Citra PlanetScope di Kabupaten Langkat

Abstract

Tulisan ini menerangkan tentang identifikasi sebaran spasial hutan mangrove di Kabupaten Langkat. Penelitian ini mengintegrasikan data citra PlanetScope dengan data spasial sosio-geo-biofisik untuk mengembangkan model algoritma decision tree of machine learning guna mengidentifikasi sebaran spasial hutan mangrove di Kabupaten Langkat. Model diuji dengan tiga skenario kombinasi variabel: (1) variabel spektral (NDVI, NRGI, NDWI, GNDVI, GARI, ARVI, VDVI, SAVI, CMRI); (2) variabel sosio-geo-biofisik (substrat, salinitas, elevasi, jarak dari sungai, jarak dari jalan); dan (3) kombinasi keduanya. Hasil menunjukkan algoritma terbaik diperoleh dengan menggunakan skenario ketiga dengan kombinasi variabel spektral (NDVI, SAVI, ARVI, VDVI, NRGI) dan variabel sosio-geo-biofisik (substrat, elevasi, jarak dari jalan) menghasilkan performa overall accuracy 94,5% dan kappa accuracy 93%.

This paper describes the identification of the spatial distribution of mangrove forests in Langkat Regency. The study integrates PlanetScope image with socio-geo-biophysical spatial data, for developing a decision tree algorithm model of machine learning to identify the spatial distribution of mangrove forests in Langkat Regency. The model was tested with three variable combination scenarios: (1) spectral variables (NDVI, NRGI, NDWI, GNDVI, GARI, ARVI, VDVI, SAVI, CMRI); (2) socio-geo-biophysical variables (substrate, salinity, elevation, distance from rivers, distance from roads); and (3) a combination of both. The study results show that the best algorithm was obtained by using third scenario with a combination of spectral variables (NDVI, SAVI, ARVI, VDVI, NRGI) and socio-geo-biophysical variables (substrate, elevation, distance from roads) produced an overall accuracy of 94.5% and a kappa accuracy of 93%.