Inggit Lolita Sari
National Research and Innovation Agency

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Random Forest-Based Vegetation and Non-Vegetation Classification in Siak Regency Using Landsat 8 and Google Earth Engine Faiza Aulia Supriatna; Elvira Nurrahma; Inggit Lolita Sari; Emilia Roza; Harry Ramza
Jurnal Teknoif Teknik Informatika Institut Teknologi Padang Vol 14 No 2 (2026): TEKNOIF OKTOBER 2026
Publisher : ITP Press

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.21063/jtif.2026.V14.2.932

Abstract

Monitoring vegetation degradation is essential for supporting sustainable land resource management through accurate and continuous land cover assessment. This quantitative study classifies vegetation in Siak Regency, Riau Province, using Landsat 8 Surface Reflectance (Level-2) imagery from 2024 and the Random Forest (RF) algorithm in the Google Earth Engine (GEE) environment. Fifty purposively selected ground truth samples, categorized as vegetation and non-vegetation, were used. Spectral bands were combined with three vegetation indices, namely NDVI, SAVI, and EVI, to improve classification performance. The dataset was randomly split into 70% training data (33 samples) and 30% validation data (17 samples) to maintain reproducibility. The model achieved an Overall Accuracy of 88.24% and a Kappa coefficient of 0.757, indicating substantial agreement between predicted and reference classes. The integration of the three indices provided complementary spectral information, improving the model’s ability to distinguish vegetation density, atmospheric influence, and soil background sensitivity. However, the small validation dataset and binary classification scheme limit the statistical reliability of the results and their comparability with multi-class studies. Overall, the findings show that combining Random Forest and Google Earth Engine offers an efficient and scalable approach to vegetation mapping. Future studies should use larger datasets, more robust validation strategies, and finer land cover classification schemes to improve reliability in tropical environmental monitoring.