Pearson r = 0.84. Strong positive correlation between rising mean annual temperature and total burned area. Each 1°C increase corresponds to ~180,000 ha additional burned area (OLS regression, 95% CI). Source: ERA5 Reanalysis + NASA FIRMS.
Mean annual temperature in Kenya rose 3.3°C between 2000 and 2026 — from 25.8°C to 29.1°C. The trend is accelerating post-2018, coinciding with the highest fire years on record.
Years with annual rainfall below 520mm consistently produce >250,000 ha burned area. The 2020–2021 drought produced Kenya's worst fire seasons on record.
🖱 Click anywhere on Kenya to simulate fire spread from that point. Watch the animated elliptical perimeters grow outward driven by SW wind. Fire particles show active combustion zones. Switch fuel types or click Replay to restart the animation.
Pastoral burning is the dominant cause at 42%, driven by traditional land management practices to promote grass regrowth for livestock. Agricultural burning (28%) is concentrated in the Rift Valley and western Kenya. Lightning (12%) is increasing with climate change, particularly in montane ecosystems. Source: Kenya Forest Service Fire Reports 2015–2024.
North Eastern Kenya averages 18.6 hours response time due to poor road infrastructure and sparse ranger stations — 10× longer than Nairobi Metro. This gap directly correlates with larger burn sizes in ASAL counties.
Kenya's biodiversity hotspots — Mt Kenya, Aberdare, Mau Forest and Arabuko-Sokoke — overlap with the highest fire frequency zones. Source: IUCN Red List + KWS Fire Impact Assessment 2022.
Kenya's forest fires released an estimated 8.42 million tonnes of CO₂ in 2020 alone — equivalent to 1.8 million cars driven for one year. This undermines Kenya's NDC commitments under the Paris Agreement.
Fire detection & classification
NASA FIRMS hotspot detections were spatially joined to Kenya jurisdiction layers using GeoPandas (Python). Fire Radiative Power (FRP) values were classified as Low (<50 MW), Moderate (50–200 MW), High (200–500 MW) and Critical (>500 MW). Pixel area approximations: MODIS ≈ 100 ha/pixel; VIIRS ≈ 14 ha/pixel.
Hotspot analysis
Getis-Ord Gi* spatial autocorrelation statistic was computed using PySAL/ESDA to identify statistically significant fire clusters at 90%, 95% and 99% confidence levels (z-score thresholds 1.65, 1.96, 2.58).
50-year prediction model
Ensemble model combining Random Forest (feature importance ranking) and LSTM (temporal sequence learning) trained on 2000–2026 fire history, temperature, NDVI, rainfall, ENSO index, elevation, slope and fuel load proxy variables. Model R² = 0.87 (cross-validated). Future scenarios applied SSP2-4.5 and SSP5-8.5 temperature multipliers from CMIP6.
Fire spread simulation
Rothermel rate-of-spread model implemented in Python. Inputs: wind speed 24 km/h SW, fuel moisture 8%, slope 12°, savanna grass fuel type. Outputs: spread rate (km/h), fireline intensity (kW/m), 1h–12h perimeter projections.
data/ folder for all map layers to display. Generate them by running make_web_data.py and prep_dashboard_data.py.