The 2025 Timor-Leste Land Cover Statistics pilot report presents the methodology, implementation, and preliminary results of Timor-Leste's first Land Cover Classification pilot program, conducted by the National Institute of Statistics (INETL) through the Directorate of Social Demographic Statistics (DESD) in the Administrative Post of Aileu Village, Aileu Municipality. The initiative aims to establish a systematic land monitoring framework based on updated data to inform policy decisions regarding natural resource management, climate change mitigation, urban planning, and sustainable economic development, while also supporting indicators for Sustainable Development Goals (SDGs) 1, 2, 6, and 15.
The methodology employed a two-stage cluster sampling approach combining Simple Random Sampling (SRS) and Probability Proportional to Area (PPA), using agricultural parcel data from the 2019 Agricultural Census as the sampling frame. Selected parcels were divided into 50m × 50m grids for ground truth validation. Field data collection was conducted using the Open Foris Ground application developed by the Food and Agriculture Organization (FAO), with 931 coordinate points collected across ten days by three field teams.
Land Cover Classification was performed using Google Earth Engine (GEE) with a Random Forest machine learning model, utilizing satellite imagery from 2019 to 2025. The model incorporated spectral indices (including NDVI), cloud masking, and crop growth calendars to classify seven land cover classes: Ai-Farina (cassava), Floresta (dense forest), Batar (maize), Natar (rice), Rai Maran (bare land), Area Infraestrutura (built-up area), and Be'e (water bodies). Training and testing samples were split 70:30, with accuracy assessment yielding an overall accuracy of 98% and a Kappa statistic of 0.98. Spatial analysis was conducted in QGIS, including manual feature editing to correct misclassifications, area calculations, and land cover change analysis using the Geoprocessing Intersection tool.
Results indicate that forest area (Floresta) remained the dominant land cover class, increasing from 33.4% in 2019 to 39.6% in 2025, representing a 6.2% increase. The largest proportional increase was observed in maize (Batar) at 347%, while Cassava experienced the largest decline at −39% (43 km² decrease). The land cover change matrix revealed significant conversions, including 30 km² of Cassava transitioning to Forest and 7.4 km² converting to Infrastructure Area. The pilot demonstrated the feasibility of using machine learning and cloud-based geospatial platforms for land cover monitoring in Timor-Leste, while identifying areas for methodological improvement, including proportional sample distribution, data augmentation, image resolution enhancement, and phenological studies using Sentinel-1 data. The report recommends continued collaboration between Directorate of Social and Demographic Statistics and the Directorate of Methodology to refine the methodology and produce Land Cover Statistics aligned with international standards.
