Pengembangan Algoritma Machine Learning Untuk Deteksi Kesehatan Mangrove Berbasis Citra Sentinel-2 di Kepulauan Natuna

Abstract

Penelitian ini mendeskripsikan pengembangan algoritma deteksi kesehatan mangrove dengan mengintegrasikan data penginderaan jauh dan peubah sosio-geo-biofisik menggunakan pendekatan machine learning di Kepulauan Natuna, Indonesia. Penelitian ini berfokus pada pengujian kombinasi peubah spektral yang diturunkan dari citra Sentinel-2 dan peubah sosio-geo-biofisik, meliputi substrat, elevasi, jarak dari jalan, jarak dari permukiman, jarak dari sungai, dan jarak dari garis pantai. Model dikembangkan menggunakan algoritma decision tree melalui seleksi peubah dan optimasi parameter untuk memperoleh model klasifikasi terbaik. Hasil penelitian menunjukkan model terbaik menghasilkan overall accuracy sebesar 95,4%, koefisien kappa sebesar 0,95, user's accuracy berkisar antara 91,6-99,9%, serta producer's accuracy antara 90-100%. Enhanced Vegetation Index (EVI) merupakan peubah yang paling berpengaruh dalam algoritma. Model yang dikembangkan mampu mengidentifikasi kondisi kesehatan mangrove dengan akurasi yang tinggi, sehingga berpotensi mendukung kegiatan pemantauan dan pengelolaan ekosistem mangrove di Kepulauan Natuna.

This research describes the development of a mangrove health detection algorithm by integrating remotely sensed and socio-geo-biophysical data using a machine learning approach in the Natuna Islands, Indonesia. The research focused on examining the combination of spectral variables derived from sentinel-2 imagery and socio-geo-biophysical variables, including substrate, elevation, distance from roads, settlements, rivers, and the coastline. A decision tree algorithm was developed using feature selection and parameter optimization to identify the optimal classification model. The research showed that the best-performing model provided an overall accuracy of 95.4%, a Kappa coefficient of 0.95, user's accuracy ranging from 91.6% to 99.9%, and producer's accuracy ranging from 90.0% to 100%. The Enhanced Vegetation Index (EVI) was the most important variable in the developed algorithm. The developed model was able to identify mangrove health conditions with high classification accuracy, demonstrating the potential of integrating spectral and socio-geo-biophysical variables to support mangrove health monitoring and management in the Natuna Islands.