FAO and Timor-Leste's government built an XGBoost-corrected seasonal drought index that triggered the country's first Anticipatory Action protocol in October 2023, matching observed drought conditions over 80% of the time (2011–2022) across five high-risk municipalities.
80%+
Match rate with observed drought conditions (backtest, 2011-2022)
0.82
Heidke Skill Score, no-drought conditions (3-month lead time)
0.65-0.68
Heidke Skill Score, moderate drought (3-month lead time)
40%
Maize production decline in comparable 2016 El Niño
57%
Rice production decline in comparable 2016 El Niño
Details
Maturity
Scaling
Promoter
Food and Agriculture Organization of the UN (FAO) Timor-Leste, with Timor-Leste's Ministry of Agriculture, Livestock, Fisheries and Forestry and National Directorate of Meteorology and Geophysics
Period
2023–2026 (AA protocol activated Oct 2023; ASIS portal launched May 2026)
In 2023, Timor-Leste's Ministry of Agriculture, Livestock, Fisheries and Forestry and National Directorate of Meteorology and Geophysics, working with FAO, built a Combined Drought Index (CDI) to trigger the country's first formal Anticipatory Action protocol for agricultural drought. The index blends the Standardized Precipitation Index, a soil-moisture anomaly index, a vegetation health index and El Niño/Indian Ocean Dipole indicators.
Activities
An XGBoost regression model, trained on monthly data from 1993 to 2023 (n=369, 70/30 train-test split), bias-corrects ECMWF SEAS5 seasonal rainfall forecasts for the country's data-sparse municipalities. The Anticipatory Action protocol was formally activated in October 2023 during an El Niño event, covering five high-risk municipalities: Baucau, Covalima, Liquiçá, Viqueque and Oé-Cusse Ambeno.
Results
Backtesting against 2011-2022 records found the CDI matched observed soil-moisture-based drought classifications more than 80% of the time, with Heidke Skill Scores of 0.82 for no-drought conditions and 0.65-0.68 for moderate drought at a three-month lead time (falling to 0.33-0.46 for milder categories). A comparable 2016 El Niño had cut Timor-Leste's maize and rice production by 40% and 57% respectively, in a country where 79% of households depend on crop production.
Conclusions
FAO's 2024 technical working paper is candid about the system's limits: no historical ground-based rainfall or soil-moisture observations existed in the pilot municipalities, forcing reliance on satellite (CHIRPS) proxies, and the model 'will require regular testing, revision and calibration.' Donor rules requiring formal disaster declarations before funds can be released are flagged as a structural bottleneck. In May 2026 the government and FAO launched a successor Agricultural Stress Index System (ASIS) portal, financed by the Green Climate Fund, extending drought monitoring nationally.
Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
Timor-Leste Ministry of Agriculture, Livestock, Fisheries and Forestry (MALFF), National Directorate of Meteorology and Geophysics, FAO technical team, Civil Protection Authority
Conditions for success
Multi-institutional collaboration between government ministries and FAO
Bias-correction of seasonal forecasts (ECMWF SEAS5) via machine learning for data-sparse regions
Sustained funding pathway (Green Climate Fund) for the successor system
Common failure modes
No historical ground-based rainfall or soil-moisture observations existed in pilot municipalities, forcing reliance on satellite (CHIRPS) proxies rather than ground truth
Donor rules requiring formal disaster declarations before Anticipatory Action funds can be released are a structural bottleneck
Model requires regular testing, revision and calibration per FAO authors
Where it fits
Governance type
national government with UN agency technical assistance
Scale
subnational pilot (5 of 13 municipalities) scaling to national
Income level
lower-middle-income
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
Mongolia's National Statistics Office piloted satellite-and-drone machine-learning models in Arkhangai province, identifying large livestock with 81.6% accuracy and traditional ger …
Sierra Leone's National Medical Supplies Agency and University of Pennsylvania researchers built a decision-aware ML tool that raised essential-medicine consumption …