Content Type
Program
Two independent reports on AI in humanitarian action reach the same conclusions from different angles. This piece compares ITU's new early warning report with NASA Lifelines' own GeoAI findings on trust, co-design, and connectivity.
In February 2026, NASA Lifelines published GeoAI for Humanitarian Action, a report built on interviews, focus groups, and a community survey to understand how open source AI and satellite data are being used across the full humanitarian crisis lifecycle. A few months later, the ITU released its own report, Leveraging AI to Enhance Multi-Hazard Early Warning Systems, produced through the Early Warnings for All initiative. The two reports were written independently, by different teams, for different audiences. Read side by side, they land on nearly identical conclusions.
Where the Lifelines report takes a wide view across anticipatory action, response, and recovery, the ITU report narrows in on a single slice of that lifecycle: early warning systems. That narrower focus turns out to be useful. It gives us five live, deployed case studies (in Tonga, the United States, Liberia, Colombia, and China) that put evidence behind some of the more abstract barriers our own community survey identified. Here is where the two reports overlap, and what each one adds to the other.
Our GeoAI report found that Strategic Decision Makers, the humanitarian leaders responsible for approving and standing behind a tool’s use, are held back less by what the models can do and more by whether they can trust the outputs. Lack of transparency, unclear model behavior, and reputational risk topped their list of concerns.
The ITU report reaches the same place from a different angle. Its recommendations call for strong AI governance, national focal points, and human oversight built into any life safety decision. Both reports describe the same gap. The Lifelines survey captured it as a lived frustration among decision makers. The ITU report translates it into a governance framework.
Across our stakeholder personas, one requirement showed up repeatedly regardless of role: humanitarian partners need to be involved from the start of a tool’s development, not brought in after the fact to validate something built without them.
The ITU report states this almost word for word as a core recommendation, calling for AI design that is human-centered, equity-driven, and co-designed with affected communities from initial development through deployment. Colombia’s Sketch Map Tool, featured in the ITU report, is a working example of this in practice. Communities in the Río Atrato Basin lead workshops where participants hand-draw conservation project locations, with AI reading the sketches into usable digital data. It is a small-scale but concrete answer to a barrier our own survey respondents described in the abstract.
Our report identified poor connectivity as one of the top operational barriers facing Data Driven Humanitarians, the people using these tools in the field. It is a barrier that shows up constantly in survey responses but is hard to illustrate without a specific example.
The ITU report gives us that example. Its Liberia case study describes the Early Warning Connectivity Map, built by ITU with Microsoft, IHME, and Planet Labs, which identifies exactly where populations are at risk of flooding but sit in mobile network coldspots and cannot receive an alert at all. It is a direct, mapped version of the connectivity gap our own community described, and a reminder that the most sophisticated forecasting model in the world does not help anyone if the warning cannot reach them.
Both reports draw from the same open-source community that Lifelines works alongside every day. Our GeoAI report specifically highlights the Humanitarian OpenStreetMap Team’s fAIR tool, an AI assisted mapping platform that HOT uses to localize open source models like RAMP with volunteer mappers on the ground. It is one more thread connecting the two reports back to the partners already active in our own network.
Anticipatory action and early warning were already priorities for the Data Studios, and this ITU report is a strong piece of source material for that work. If you are digging into flood-specific tools and workflows, the Flood Early Warning Data Studio package is the most directly relevant place to start layering in some of these findings.
Two reports, written apart, arriving at the same conclusions is a sign of where the real work is right now. It sits in building the trust, partnerships, and infrastructure that let these models actually reach the people who need them.
