1-, 2-, and 3-month-ahead disease-incidence risk forecasting at sub-national scale Initial coverage: dengue & malaria in Peru and Brazil — additional countries and diseases are being onboarded.
🌐 Live Dashboard · 🗺️ Using the Dashboard · ℹ️ About · 🔬 Research Basis · 👥 Team · 📜 Roadmap · 🤝 Get Involved · 🆘 Support
🚧 Status: Under active development. This repository is an early, evolving release published in coordination with the DIRE project team at UC San Diego (UCSD) and New Light Technologies (NLT). Its structure, documentation, and included code are being staged and reviewed phase by phase; content may change. It is shared now to enable open review and collaboration — please treat it as work in progress and coordinate with the project team before relying on it.
⚠️ Disclaimer: All forecasts and analytical outputs are provided "as is" for decision-support purposes only. New Light Technologies, UC San Diego, and UNICEF expressly disclaim all liability for decisions made using these outputs. See DISCLAIMER.md for full terms.
Phase 0 — initial release (work in progress). This repository is being populated, in coordination with the project team at UCSD and NLT, with the data-source documentation and Python/R processing pipelines for the DIRE IMPACT Platform. It is the open, community-facing home of the project, released under the Wellcome Trust open-source grant conditions. The forecasting API and interactive dashboard are live at dire.impact.nltglobal.com; their source code will follow in subsequent phases. See docs/ROADMAP.md for the full release plan.
DIRE (Dengue Integrated Risk Estimation) is a machine-learning forecasting platform that produces 1-, 2-, and 3-month-ahead disease-incidence estimates (the current month plus the following two) at the sub-national administrative level. It is designed to support public-health decision-makers, epidemiologists, and humanitarian responders in anticipating disease burden and allocating resources before outbreaks peak.
Coverage is expanding. Peru and Brazil — and the diseases dengue and malaria — are the first countries and diseases being trialed during the platform's development and operations-and-maintenance (O&M) phase. The methodology and pipelines are built to generalize, and additional countries and diseases are being onboarded over time. References to Peru, Brazil, dengue, and malaria throughout this repository describe the current rollout, not the platform's ultimate scope.
The platform integrates:
- Remote-sensing covariates — NASA Earthdata (MODIS land cover + vegetation indices, NASADEM elevation)
- Climate covariates — Copernicus ERA5-Land (temperature, precipitation, humidity)
- Epidemiological surveillance — Brazil DATASUS SINAN and Peru MINSA dengue case records
- An ensemble ML model — CatBoost, SVM, and LSTM models synthesized with a Random Forest, as described in Sebastianelli et al. (2024)
Risk estimates are mapped to five tiers (Lowest → Highest) at GADM Level 3 (department/district) granularity and are refreshed monthly.
Partners: New Light Technologies (NLT) · UC San Diego School of Global Policy and Strategy · Universidad Peruana Cayetano Heredia (UPCH) Funder: Wellcome Trust
DIRE is an open science project. We built this repository so that researchers, public-health agencies, and the broader open-source community can inspect, reproduce, improve, and build on the work — not just read about it. There are several ways to participate, whatever your background.
- Reproduce and validate. Every processing pipeline in
workflows/is documented so you can re-run our covariate engineering and forecasting steps against the bundled aggregate data — no credentials or private data required. See docs/processing/. - Extend the science. Propose new covariates, alternative model architectures, additional diseases, or new countries. Open a discussion to sketch an idea before you build it.
- Cite the work. If DIRE supports your research, please cite both the software and the underlying methodology paper — see Citation.
- Use the dashboard. The live platform at dire.impact.nltglobal.com is free to use for planning and situational awareness.
- Tell us what you need. Resource-estimation and reporting features are shaped by real operational needs. File an issue describing your use case, the decisions you make, and the outputs that would help.
- Report issues — bugs, unclear docs, or reproducibility gaps.
- Improve documentation — clarity fixes and translations are especially welcome, given the platform's Peru/Brazil focus (Spanish and Portuguese).
- Submit pull requests — see CONTRIBUTING.md for the workflow and the Developer Certificate of Origin (DCO) sign-off we require.
- We work with surveillance data from national health authorities under appropriate agreements. If your institution wants to collaborate on data, validation, or a country expansion, reach out through the channels in docs/TEAM.md.
Community standards. Participation in this project is governed by our Code of Conduct. Please read it before engaging.
Need help, found a bug, or have a question? See SUPPORT.md for where each request should go. In short:
- Bugs, errors, reproducibility gaps → open an Issue.
- Questions, ideas, new covariates/diseases/countries → start a Discussion.
- Security vulnerabilities → follow SECURITY.md (please don't open a public issue).
- Live dashboard feedback → use the feedback form on dire.impact.nltglobal.com.
New issues are automatically acknowledged and routed to the maintainers.
platform/
├── docs/
│ ├── data-sources/ # Per-source credential, license & coverage guides
│ ├── processing/ # End-to-end pipeline run instructions
│ ├── ABOUT.md # Platform description, diseases, data, references
│ ├── TEAM.md # Project team and partner organisations
│ └── ROADMAP.md # Phased release plan
├── workflows/
│ ├── predictors/ # Python pipeline: remote-sensing & climate covariates
│ └── dengue_forecasting/ # Python + R pipeline: disease incidence forecasts
├── CITATION.cff # Machine-readable citation metadata
├── CONTRIBUTING.md # PR workflow and DCO sign-off
├── CODE_OF_CONDUCT.md # Community standards
├── LICENSE # Apache-2.0 (source code)
├── NOTICE # Attribution
├── SECURITY.md # Vulnerability disclosure
├── DISCLAIMER.md # Platform disclaimer and liability terms
└── .env.example # Credential template
| Component | Planned phase |
|---|---|
| Prediction API / backend service | Phase 1 |
| Containerized one-command local runner | Phase 1 |
| Automated test suite + CI | Phase 1 |
| Interactive dashboard source code | Phase 2 |
| Cloud deployment / infrastructure-as-code | Phase 3 |
| Zenodo DOI archive | Phase 3 |
External data sources
NASA Earthdata (MODIS, NASADEM) ─────────┐
Copernicus CDS (ERA5-Land) ──────────────┤──► workflows/predictors/
Brazil DATASUS SINAN ────────────────────┤ │ covariate tables
Peru MINSA surveillance ─────────────────┘ ▼
workflows/dengue_forecasting/
├── feature engineering
├── CatBoost + SVM + LSTM ensemble
└── Random Forest synthesizer
│
▼
Forecast outputs (predicted cases, risk tier,
confidence interval) by GADM L3 admin unit × month
│
▼
dire.impact.nltglobal.com (live dashboard)
See docs/ABOUT.md for the full description and references.
| Tool | Version |
|---|---|
| Python | 3.11+ (3.12 recommended) |
| R | 4.5+ |
| uv (Python env manager) | latest |
git clone https://github.com/DIRE-OSI/platform.git
cd platform# Predictors (remote-sensing + climate) pipeline
cd workflows/predictors && uv sync && cd ../..
# Dengue forecasting pipeline
cd workflows/dengue_forecasting && uv sync && cd ../..cp .env.example .env
# Edit .env and fill in your free API credentials — see docs/data-sources/Full instructions live in docs/processing/. The forecasting pipeline can be run against the bundled aggregate data without any external credentials.
Source code: Apache License 2.0 — see LICENSE and NOTICE. Documentation and non-code content: Creative Commons Attribution 4.0 International (CC-BY-4.0).
If you use this software or the data pipelines, please cite both the software release and the underlying methodology paper.
- Software: New Light Technologies, Inc. (2026). DIRE IMPACT Platform. GitHub. See CITATION.cff.
- Methodology: Sebastianelli, A., et al. (2024). A machine learning approach to dengue and malaria incidence forecasting. Scientific Reports. DOI: 10.1038/s41598-024-52796-9
This work was funded by a Wellcome Trust grant to the University of California, San Diego (UCSD). The software was developed by New Light Technologies (NLT), Inc. Scientific collaborators include Universidad Peruana Cayetano Heredia (UPCH). The live DIRE dashboard is operated by NLT, Inc. and is currently in soft-launch with UX/UI and data-quality reviews underway.