assess-climate-risk-kenya

name	assess-climate-risk-kenya
description	Assess and rank climate risk (flood, drought, rainfall) for Kenyan counties from a natural-language question, grounded in Earth Engine data, population exposure, and retrieved Kenyan documents.

Skill: Assess Climate Risk in Kenya

A reusable skill in the I/O 2026 SKILL.md format. It encodes the standard procedure the agent follows to answer a climate-risk question for Kenya, so the same workflow can be invoked consistently and extended to new hazards.

Inputs

Input	Required	Description
question	yes	Plain-English question about climate risk in Kenya.
counties	no	Specific county names to focus on. Omit for all 47 counties.
start_date / end_date	no	Analysis window (ISO dates). Defaults to the long-rains season.
hazard	no	Focus hazard: flood, drought, or rainfall. Inferred from the question if omitted.

Process

1. Understand the question. Identify the hazard, the geographic scope (specific counties or all of Kenya), and what the user wants ranked.
2. Retrieve context. Query the knowledge base for relevant Kenyan documentation (historical floods/droughts, county profiles, authority bulletins).
3. Gather geophysical signals. Pull flood (Sentinel-1 SAR), rainfall (CHIRPS), and elevation (SRTM) layers per county from Earth Engine.
4. Quantify exposure. Retrieve WorldPop population per county.
5. Score and rank. Combine the signals with the transparent weighted model into a per-county risk score and rank the counties.
6. Fuse and explain. Produce a grounded answer: ranked counties, factor-by factor explanations, source citations, confidence, and recommendations.

Tool sequence

search_knowledge(query=<question>, top_k=4)
        │   retrieved Kenyan context (with sources)
        ▼
get_earth_engine_layers(counties=<scope>, start_date=…, end_date=…)
        │   flood / rainfall / elevation per county
        ▼
get_population_exposure(counties=<scope>)
        │   exposed population per county
        ▼
compute_risk_score(counties=<scope>)
        │   ranked, transparent risk scores
        ▼
   (fusion) → final grounded answer

For a pure-context question ("what does the NDMA say about Tana River?"), the skill may stop after search_knowledge.

Outputs

• A ranked list of counties by climate-risk score (0–1), highest first.
• Per-county components (flood, rainfall, population, elevation) and their weighted contributions to the score.
• Exposed population per county.
• A natural-language explanation with source attribution.
• A confidence indicator and recommendations.

Example usage

Question:

"Which counties in Kenya face the highest flood risk this season, and who is most exposed?"

Tool calls (plan):

1. search_knowledge(query="highest flood risk counties Kenya long rains", top_k=4)
2. get_earth_engine_layers(counties=[], start_date="2026-03-01", end_date="2026-05-31")
3. get_population_exposure(counties=[])
4. compute_risk_score(counties=[])

Result (shape):

Ranked counties (e.g. Tana River, Garissa, Busia…) with risk scores and exposed-population figures; an explanation tying the top ranks to high flood signal + low elevation + large exposed population, citing the NDMA bulletin and the Tana flood-vulnerability study; medium confidence (note if running on fallback data); and recommendations to pre-position resources in the top counties and validate against the latest field reports.