We turn satellite data, hydrological models, and machine learning into actionable drought and flood risk analytics — so that investors, insurers, governments, and food-supply chains can act before the crisis hits.
👋 Hi, I'm your AI guide.
AI4Water helps companies modernize their climate risk assessment. We connect satellite data, hydrological models, and AI to show you — in plain language — how drought and flood risk affects your supply chain, your portfolio, and your bottom line. Click the map below to see it in action.
Droughts and floods don't just damage ecosystems — they disrupt supply chains, move commodity prices, and change insurance liabilities overnight. Our models connect hydrological events to financial exposure across crops, regions, and sectors.
A drought is not a single event. It propagates through layers: meteorological deficit → soil-moisture depletion → agricultural stress → hydrological shortage → socioeconomic impact. Each layer has its own onset, duration, and recovery — and they don't move in lockstep. Our models track the cascade, not just the headline.
The same atmospheric patterns that cause drought in one region often drive flooding in another. Cross-region synchrony — what we call Concurrent Drought Analysis — reveals the teleconnections that matter for portfolio risk. A drought in Brazil and a flood in Vietnam may share the same ENSO trigger.
We don't just run black-box ML. Our Hydroinformatics Foundation Architecture (HFA) couples satellite foundation models (TerraMind, Clay, Hydro FM, Prithvi) with verified physics simulators (Delft3D, SWAN, XBeach) and Lagrangian moisture tracking — so every prediction has a physical explanation.
Every market signal passes through evidence gates: causal attribution → cross-region synchrony → cascade-layer verification → regional teleconnection → market confirmation. We track 23 companies across 11 sectors, with 112K trade records and 31K crop-price observations. No signal ships without evidence.
Select any city below. Our AI analyst cross-references satellite precipitation, soil moisture, ENSO indices, and crop-price databases to tell you: what's the risk, which crops are exposed, which companies feel it, and where else in the world the same pattern is appearing.
Data pipeline: GPM IMERG v7 (30-min precip, 0.1°) → ERA5-Land (soil moisture, temp, 0.1°) → ONI/MEI ENSO (NOAA CPC) → CHIRPS anomaly (0.05°) → FAOSTAT crop prices → Company financials. AI analysis runs on-demand with the latest available indices. LIVE · refreshed hourly FFG: SIMULATION · using synthetic QPE Research output — not financial advice.
Click any marker to load its AI forecast card above. Red = active warning, orange = watch, green = normal.
Every forecast card on this page is backed by a six-layer system that ingests, models, and refines its understanding of water risk continuously.
Every 30 minutes, the system pulls satellite precipitation (GPM IMERG 0.1°), hourly soil moisture and temperature (ERA5-Land 0.1°), ENSO indices (NOAA CPC), and rainfall anomaly fields (CHIRPS 0.05°). Crop prices update daily from FAOSTAT. Company financials update quarterly. No single source is trusted alone — cross-validation against historical climatology flags sensor drift before it reaches the model.
The Hydroinformatics Foundation Architecture (HFA) couples geospatial foundation models (Hydro FM, TerraMind, Prithvi) with verified physics simulators (SWAN, Delft3D, XBeach) and Lagrangian moisture tracking. For flash flood guidance, a basin-scale SAC-SMA-style soil moisture accounting model computes dynamic FFG thresholds at 1h/3h/6h durations per catchment. The ensemble runs multiple models in parallel; disagreement triggers deeper investigation.
Every signal passes through five evidence gates before surfacing: causal attribution → cross-region synchrony → cascade-layer verification → teleconnection confirmation → market impact correlation. A drought signal in Colombia must show consistent patterns across IMERG, soil moisture, reservoir levels, and ENSO before triggering a crop exposure alert. The system currently tracks 23 companies across 11 sectors with 112K trade records.
Raw numbers don't help decision-makers. The AI analyst generates a concise natural-language explanation for each city: what the risk is, why the models flagged it, which crops are exposed, how markets are reacting, and where else the same pattern is emerging. Every statement is traceable to its source data — no black-box claims. For flash flood basins, it explains which catchments are close to exceedance and why upstream conditions matter.
The system learns from every event. When a forecasted drought materialises, the model records which signals preceded it correctly and which were noise. When a flash flood threshold is exceeded, the actual streamflow (where gauges exist) is compared to the FFG prediction. Over time, the system refines its per-basin Thresh-R values, soil moisture scaling curves, and exceedance probability distributions — becoming more accurate for each specific catchment without losing global transferability.
All benchmark results are public. Code is on GitHub. Forecast summaries are available as structured JSON for integration into your own dashboards. The FFG pipeline outputs GeoJSON with per-basin FFG, QPE, exceedance ratios, and tier assignments — ready for any GIS or web mapping tool. Research methodology is documented and reproducible. We learn in the open.
The system does not require manual calibration per site — it starts from global defaults and improves through continuous validation against observations. This is the statistical-distributed approach: run the model on historical data, establish flood/drought frequency per basin, then compare real-time output to those baselines.
Systematic comparison of 9 geospatial foundation models (Hydro FM, TerraMind, Clay, Prithvi, Sapiens) on water-body segmentation. Interactive dashboard with per-model IoU, F1, precision-recall across 17 Colombian coastal sites. Open benchmark methodology, reproducible in any region.
benchmark (contact for access) ATMOSPHERIC INTELLIGENCECombined Eulerian-Lagrangian platform for atmospheric river detection and moisture source attribution. ERA5 back-trajectories, object-based rainfall feature clustering, and moisture convergence fields. Real-time monitoring for Colombia, with CONUS and Mekong expansions planned.
orbita (tailnet — contact for access) FINANCIAL INTELLIGENCEEvidence-gated commodity risk signals connecting hydrological events to financial exposure. 23 companies across 11 sectors (fertilizer, grain, equipment, protein, CAT bonds). 112K trade records, 31K crop-price observations. CDA → cross-region synchrony → cascade layers → market confirmation pipeline.
trading (contact for access) OPERATIONAL SHELLDelft-FEWS-class operational shell for countries and river basins. Sector-of-use analytics across 5 system layers × 7 dimensions. Agent protocol for governed forecast triggers, risk classification, confirm/reject warning pipeline. First deployment: El Salvador. Built in Rust.
watria (contact for access)ORBITA tracks atmospheric moisture from source to sink: ERA5 back-trajectories, atmospheric river detection, and moisture-source attribution for Colombia — with CONUS and Mekong expansions planned. Explore the live platform below.
Live interactive dashboard: trajectory analysis, moisture convergence fields, and rainfall feature clustering. Open it in a new tab for the full experience.
Embedded view of ORBITA (Firebase-hosted, repo: github.com/corzogac/ORBITA). If the frame appears blank, the platform may need to load scripts — use the Open ORBITA button instead.
We design every system to be consortium-ready: open protocols, documented APIs, reproducible pipelines, and GDPR-compliant data handling. If you're building a Horizon Europe, UKRI, or national research proposal that needs AI-for-water expertise, we're ready.
Production ML pipelines on Google Cloud Run. Geospatial foundation model benchmarking. Real-time ERA5 + IMERG data ingestion. CARAVAN hydrological dataset (16K+ gauges). PostgreSQL + pgvector knowledge infrastructure.
Coupling satellite foundation models with verified physics simulators. Flash Flood Guidance system from IMERG + ERA5-Land + HydroSHEDS. Lagrangian moisture tracking. Basin-scale exceedance thresholds at 1h/3h/6h durations.
Evidence-gated drought-to-market signals. 23 companies across 11 sectors. Flood↔drought symmetry for portfolio hedging. 112K trade records and 31K crop-price observations. CDA → cross-region synchrony → market confirmation pipeline.
We contribute to open science: all benchmarks are public, code is on GitHub, and our training materials are freely available. Looking for partners in drought finance, climate adaptation, and AI-for-science.
Whether you manage an agricultural supply chain, an insurance portfolio, or a research consortium — we build the intelligence layer you need.
195–197 Wood Street, Suite RA01
London, England, E17 3NU
Company No. 15070420
Netherlands +31 6 4533 1596
UK +44 7500 382312