Laporkan Masalah

Pemetaan Kerawanan Kebakaran Hutan di Taman Nasional Gunung Merbabu Menggunakan Metode Random Forest Berbasis Google Earth Engine

Miftah Desma Syahputra, Dr. Barandi Sapta Widartono, S.Si., M.Si., M.Sc

2026 | Tugas Akhir | D4 SISTEM INFORMASI GEOGRAFIS

Kebakaran hutan merupakan permasalahan lingkungan serius dengan dampak luas terhadap ekosistem, perekonomian, dan kesehatan manusia. Penelitian ini memetakan kerawanan kebakaran hutan di Taman Nasional Gunung Merbabu menggunakan algoritma Random Forest berbasis data penginderaan jauh pada Google Earth Engine, dengan delapan variabel prediktor (kelerengan, arah hadap lereng, elevasi, NDVI, NBR, LST, jarak ke jalur pendakian, dan tutupan lahan) dari citra Sentinel-2, Landsat 9, NASADEM, dan ESA WorldCover 10m. Model dilatih dari 453 titik presence (dNBR ? 0,1 periode 2016–2025) dan 435 titik absence menggunakan spatial block cross-validation (730 data latih, 158 data uji) yang menghasilkan peta probabilitas kerawanan yang diklasifikasikan menjadi lima kelas dengan Equal Interval. Model mencapai nilai overall accuracy 0,68; Kappa 0,356; dan AUC 0,7637. Elevasi merupakan variabel dengan importance tertinggi (99,50), diikuti aspect atau arah hadap lereng(89,39), NDVI (82,26), dan jarak ke jalur pendakian (81,94), menegaskan dominasi faktor topografi dan vegetasi. Sebanyak 53,29% AOI termasuk kelas kerawanan sedang hingga sangat tinggi, terkonsentrasi di lereng tengah – atas sisi Utara – Tenggara. Verifikasi lapangan pada 24 titik di jalur pendakian Selo menunjukkan konsistensi antara kondisi fisik teramati dan logika spasial model. Hasil pemetaan divisualisasikan melalui aplikasi web interaktif Google Earth Engine Apps.

Forest fires are a serious environmental problem with far-reaching impacts on ecosystems, the economy, and human health. This study maps forest fire susceptibility in Mount Merbabu National Park using Random Forest algorithm based on remote sensing data from Google Earth Engine, with eight predictor variables (slope, aspect, elevation, NDVI, NBR, LST, distance to hiking tracks, and land cover) derived from Sentinel-2, Landsat 9, NASADEM, and ESA WorldCover 10m imagery. The model was trained using 453 presence points (dNBR ? 0.1 for the 2016–2025 period) and 435 absence points via spatial block cross-validation (730 training data points, 158 test data points), resulting in a probability map of fire risk classified into five classes using the Equal Interval Model, achieving an overall accuracy of 0,68; a Kappa of 0,356; and an AUC of 0,637. Elevation was the variable with the highest importance (99.50), followed by aspect (89.39), NDVI (82.26), and distance to the hiking track (81.94), confirming the dominance of topographic and vegetation factors. A total of 53.29% of the AOI falls into the moderate-to-very-high risk classes, concentrated on the middle-to-upper slopes of the north-eastern side. Field verification at 24 points along the Selo hiking track demonstrated consistency between the observed physical conditions and the model’s spatial logic. The mapping results were visualized using the interactive web application Google Earth Engine Apps.

Kata Kunci : Bencana Kebakaran Hutan, Pembelajaran Mesin, Google Earth Engine, Sistem Informasi Geografis, Taman Nasional

  1. D4-2026-505796-abstract.pdf  
  2. D4-2026-505796-bibliography.pdf  
  3. D4-2026-505796-tableofcontent.pdf  
  4. D4-2026-505796-title.pdf