evidoria

← Back to browse

Good practice Imported

Southeast Asia's First Live AI Weather Forecasting System — DOST-PAGASA & ASTI (Philippines)

Philippines · Quezon City · See the Philippines profile · See the Quezon City profile

Evidence: Descriptive / self-reported Top 50% 60/100 · Ask Evidence Copilot about this practice

In October 2024, the Philippines' PAGASA and ASTI deployed Southeast Asia's first live AI weather system with US firm Atmo: forecast runtime fell from 3 hours to 15 minutes, lead time grew from 2 to 14 days, resolution sharpened from 3km to 2km.

3 km to 2 km km
Forecast horizontal resolution improvement (as of October 2024 deployment)
2 to 14 days
Forecast lead time extension (as of October 2024 deployment)
~3 hours to ~15 minutes time per forecast
Forecast computation (runtime) reduction (as of October 2024 deployment)
Southeast Asia's First Live AI Weather Forecasting System — DOST-PAGASA & ASTI (Philippines)

Details

Maturity
Scaling
Promoter
DOST-PAGASA (Philippine Atmospheric, Geophysical and Astronomical Services Administration) and DOST-ASTI, with Atmo Inc.
Period
October 2024–present
Keywords
weather forecasting, disaster risk reduction, machine learning, meteorology, early warning systems

Context

The Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) and the Department of Science and Technology's Advanced Science and Technology Institute (DOST-ASTI) signed a commercial agreement with US AI-meteorology firm Atmo Inc. in October 2024, describing the result as Southeast Asia's first live, operational AI-based weather forecasting system, in a country facing roughly 20 tropical cyclones a year.

Objectives

The system was intended to improve forecast resolution, extend forecast lead time, and cut computation time, feeding into the Philippines' broader hazard and impact-based warning systems including flood and tropical-cyclone early warning.

Activities

Machine-learning models were trained on satellite imagery, radar, ground-station and environmental-sensor data to generate high-resolution forecasts, integrated into national early-warning tools. A second iteration of the partnership was reported in 2026, indicating continued operational use.

Results

According to the partners, horizontal resolution improved from 3 km to 2 km, forecast lead time extended from 2 to 14 days, and computation time to produce a forecast fell from roughly 3 hours to about 15 minutes.

Conclusions

These are vendor- and agency-reported technical benchmarks rather than an independently audited evaluation of forecast accuracy against observed weather outcomes.

Implementation
Indicative cost
Medium (€50k–€500k) — Commercial agreement with a private US AI-meteorology vendor (Atmo Inc.); no public cost figures found in source.
Time to results
Medium (1–3 years) — Deployed operationally from October 2024, with a second partnership iteration reported in 2026.
Staffing & skills
DOST-PAGASA (national weather agency) and DOST-ASTI (science/technology institute) as government partners, Atmo Inc. (US AI-meteorology firm) as commercial technology vendor

Conditions for success

  • Integration with existing satellite, radar, ground-station and environmental-sensor data feeds
  • Feeding AI forecasts into established national hazard and early-warning systems (flood, tropical cyclone)
  • Sustained multi-year commercial partnership, with a second iteration reported in 2026

Common failure modes

  • Reported performance figures originate from the vendor and implementing agencies rather than independent third-party verification against observed outcomes
  • Underlying proprietary ML model and validation methodology are not publicly disclosed

Where it fits

Governance type
national government science agencies with private vendor partnership
Scale
national
Income level
lower-middle-income (Philippines)

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.

Attachments

Similar practices you may find useful