INTEGRASI CITRA SENTINEL-1 DAN SENTINEL-2 UNTUK PEMETAAN HUTAN TANAMAN EUCALYPTUS MENGGUNAKAN ALGORITMA RANDOM FOREST (STUDI KASUS: DAS SEPAKU)

Tegar Abdul Hafid, Dr. Bowo Susilo, S.Si. MT.; Dr. Prima Widayani, S.Si., M.Si.

2025 | Tesis | S2 Penginderaan Jauh

Pemetaan hutan tanaman merupakan langkah penting dalam mendukung upaya mitigasi perubahan iklim. Namun, pemetaan hutan tanaman Eucalyptus di wilayah tropis menghadapi tantangan utama berupa kemiripan spektral antar kelas vegetasi, tingginya tutupan awan, serta berbagai gangguan atmosfer yang memengaruhi kualitas citra satelit. Penelitian ini bertujuan untuk mengkaji akurasi penerapan algoritma Random Forest dalam pemetaan hutan tanaman Eucalyptus menggunakan citra Sentinel-1 dan Sentinel-2. Data utama yang digunakan meliputi citra Sentinel-1 Level 1 GRD dan Sentinel-2 Level 2, dengan tahapan pra-pemrosesan berupa koreksi geometrik dan radiometrik, reduksi noise, normalisasi reflektansi, dan komposit median bebas awan. Variabel penelitian diekstraksi dari indeks vegetasi (NDVI, EVI, GNDVI, SAVI), fitur tekstur GLCM (Entropy, Variance, Dissimilarity), serta parameter radar (VV, VH, RVI, dan Radar Mean). Analisis dilakukan dengan algoritma Random Forest. Hasil penelitian menunjukkan bahwa model klasifikasi hutan tanaman Eucalyptus memperoleh User Accuracy sebesar 87.43?n Intersection over Union (IoU) sebesar 68.92%. Dalam konteks pemetaan spasial hutan tanaman, nilai IoU dinilai lebih relevan karena merepresentasikan tingkat kesesuaian spasial antara hasil klasifikasi dan data referensi. Dengan nilai IoU sebesar 68,92% yang tergolong kategori menengah, model klasifikasi dinilai belum mampu membedakan areal hutan tanaman Eucalyptus secara tegas. Hal ini terlihat dari masih banyaknya piksel yang terklasifikasi sebagai hutan tanaman Eucalyptus namun secara spasial berada di luar areal hutan tanaman yang sebenarnya, menandakan bahwa model berbasis klasifikasi piksel digital belum sepenuhnya optimal dalam membedakan areal hutan tanaman Eucalyptus.

Mapping plantation forests is an important step in supporting climate change mitigation efforts. However, mapping Eucalyptus plantation forests in tropical regions faces major challenges, including spectral similarity among vegetation classes, high cloud cover, and various atmospheric disturbances that affect the quality of satellite imagery. This study aims to evaluate the accuracy of the Random Forest algorithm in mapping Eucalyptus plantation forests using Sentinel-1 and Sentinel-2 imagery. The primary datasets used consist of Sentinel-1 Level-1 GRD and Sentinel-2 Level-2 imagery, with preprocessing steps including geometric and radiometric correction, noise reduction, reflectance normalization, and cloud-free median composite generation. Research variables were extracted from vegetation indices (NDVI, EVI, GNDVI, SAVI), GLCM texture features (Entropy, Variance, Dissimilarity), and radar parameters (VV, VH, RVI, and Radar Mean). The analysis was conducted using the Random Forest algorithm. The results show that the Eucalyptus plantation classification model achieved a User Accuracy of 87.43% and an Intersection over Union (IoU) of 68.92%. In the context of spatial plantation mapping, IoU is considered more relevant because it reflects the spatial agreement between the classification output and the reference data. With an IoU value of 68.92%, which falls into the medium-accuracy category, the classification model has not yet successfully distinguished Eucalyptus plantation areas with sufficient clarity. This is evidenced by the large number of pixels classified as Eucalyptus plantations that are spatially located outside the actual plantation areas, indicating that the pixel-based digital classification approach is not yet fully optimal in differentiating Eucalyptus plantation areas.

Kata Kunci : Hutan Tanaman Eucalyptus, Random Forest, Sentinel-1 dan Sentinel-2