WILDFIRE RISK SITUATION REPORT Andalucía (Spain) | GFS Short-Term (≤504 h) | Legacy Simple | 10 July 2026
WILDFIRE RISK SITUATION REPORT
Andalucía (Spain) | GFS Short-Term (≤504 h) | Legacy Simple | 10 July 2026
Special focus: concurrent active fires validating the posterior | Includes media triangulation (BOJA, INFOCA, La Voz del Sur, El Mira, Libertad Digital, Sierra de Baza)
| HORIZON | GFS short-term (≤504 h · 21 days) | ENSO BASELINE | All years (2019–2025) |
| MODEL CLASS | Legacy simple (PML×SMAP + GFS × FIRMS) | RES | 500 m fine / 1 km exposure · Spain/Andalucía |
1. Executive summary
The Capybara legacy-simple wildfire model places a moderate but operationally meaningful 21-day fire-risk surface across Andalucía. The absolute exposure percentages are low relative to previous regional runs — 2.14 percent of urban land, 2.41 percent of protected areas, 2.90 percent of built-up footprint, and 4.57 percent of cropland [MODEL:table]. But the internal validation diagnostic is the story of this run.
Cropland exposure at 4.57 percent maps to 132,933 hectares of Andalusian agricultural land inside the high-risk band [MODEL:table]. Roads and grid exposure sit just above 3 percent (658 km of roads, 504 km of grid), reflecting a dispersed pattern rather than corridor-concentrated risk. Solar farms at 8.13 percent (one of ten mapped sites) is the highest per-asset exposure figure — worth checking against the specific site's operator. Wind farms at 0.45 percent register no sites in the flagged zones.
2. Model diagnostics — read this before the exposure table
Two internal diagnostic panels are attached to the run. They are not operational forecasts. They tell you whether the operational forecast is trustworthy.
Diagnostic 1 · Ignition-density ratio (high vs rest): 5.76. The chart at top-right cross-tabulates the model's risk bands (0–0.25, 0.25–0.5, 0.5–0.75, 0.75–1.0) against FIRMS active-fire detections from the concurrent 7-day window. Active fires appear in every band — as they always do; ignitions are noisy — but the density in the high band (0.75–1.0) is 5.76 times what it would be if fires were uniformly distributed across the risk surface. This is a consistency check, not a predictive-skill measure (the fires and the model share overlapping input timeframes), but a ratio this high means the posterior is landing where fires are actually starting.
Diagnostic 2 · Efficient climatology — 2026 vs baseline (2019–2024). The chart at bottom-right compares the model's monthly area-at-risk trajectory this year against the 2019–2024 average. The 2026 curve tracks below the baseline through the first half of the year but climbs sharply into July, matching (in shape and magnitude) the historical July–August peak. What this tells the operator: the current-year signal is not a mechanical replay of climatology. The model is picking up on the current dryness / soil-moisture / GFS-forecast conditions and translating them into a plausible seasonal envelope.
3. Exposure analysis
| Variable | % of total | Expected (absolute) | Total (denominator) |
|---|---|---|---|
| Population (expected exposed) | N/A — WorldPop100m did not resolve for this AOI | ||
| Cropland area (expected exposed) | 4.57% | 132,933 ha ⚠ | 2,907,314 ha |
| Urban area | 2.14% | 2,892 ha | 135,249 ha |
| Oil palm area | N/A — not present in AOI | ||
| Protected areas | 2.41% | 63,082 ha ⚠ | 2,616,647 ha |
| Built-up footprint | 2.90% | 253,282 ha ⚠ | 8,723,530 ha |
| Electric grid length | 3.91% | 504 km ⚠ | 12,875 km |
| Roads length | 3.13% | 659 km | 21,065 km |
| Oil & gas sites | N/A — not present in AOI | ||
| Wind farms | 0.45% | 0 of 37 sites | 37 sites |
| Solar farms | 8.13% | 1 of 10 sites ⚠ | 10 sites |
Reading note. The exposure percentages in this run are structurally lower than in the Cundinamarca or Peru runs published earlier this month. Two reasons: (1) the AOI is very large relative to the fire-prone cores, so the whole-Andalucía denominator dilutes any single hotspot; (2) the legacy-simple flow produces a lower-contrast posterior than the Bayesian variant. The relevant number for operational read is the ignition-density ratio (5.76), not the absolute percentages. The percentages tell you the scale of what is exposed; the ratio tells you the model has correctly localised it.
4. Where the risk concentrates — four provincial zones
Zone A · Huelva coastal + Doñana flank (Moguer, Palos de la Frontera, Almonte). The most concentrated pixel cluster on the map, with FIRMS active-fire pixels present [MODEL:map]. The Moguer fire on Camino del Tresmal is inside this zone [MEDIA:INFOCA]. Doñana's dune-forest interface remains a chronic hotspot after the 2024 dry pine mortality event. Priority: Doñana perimeter and Huelva–Palos industrial buffer.
Zone B · Sierra Morena flank (northern Sevilla, Córdoba, Jaén). Dispersed pixel cluster along the Sierra Morena–Guadalquivir depression transition. Historical repeat-burn geography; the Sierra de Cardeña–Montoro Natural Park and Sierra de Andújar sit inside the flagged surface [MODEL:map]. Priority: rural municipal fire brigades and coordination with Junta INFOCA staging points.
Zone C · Málaga hinterland (Sierra de las Nieves, Ronda, Estepona hinterland). The Estepona fire on 9 July 2026 activated Plan INFOCA in emergency phase, cut the AP-7, and forced preventive evacuations [MEDIA:El Mira]. The pixel cluster on the map extends inland from the Costa del Sol tourist strip into the sierra. Priority: coastal-urban interface, tourism-corridor evacuation planning, AP-7 corridor traffic contingency.
Zone D · Almería coastal and eastern sierra (Los Gallardos, Cabo de Gata hinterland). The Los Gallardos fire on 10 July 2026 is reported as tragic in national coverage [MEDIA:Libertad Digital] — details still emerging at the time this report was drafted. The pixel cluster on the map extends from the Cabo de Gata semi-desert to the Sierra de los Filabres. Priority: verification of casualty and containment status, coordination with the joint Málaga–Almería INFOCA deployment already reported at 300+ personnel [MEDIA:La Voz del Sur].
5. Media triangulation
- INFOCA (concurrent, 10 July 2026). Camino del Tresmal fire (Moguer, Huelva) active with 184 personnel, 13 fire trucks, 1 medical unit, 12 helicopters, 14 aircraft. Confirms Zone A.
- El Mira (9 July 2026). "Un incendio forestal en Málaga activa el Plan Infoca: desalojan viviendas y obliga a cortar la AP-7." Confirms Zone C, Estepona.
- Libertad Digital (10 July 2026). "Última hora del trágico incendio de Los Gallardos." Confirms Zone D, Almería. Detail on casualties pending.
- La Voz del Sur. "Andalucía sigue ardiendo: más de 300 efectivos del Plan Infoca luchan contra el fuego en Málaga y Almería." Confirms joint Zone C + Zone D deployment scale.
- BOJA (1 July 2026, Acuerdo del Consejo de Gobierno). The 8 June 2026 fire in Huelva formally declared a natural disaster with impact on agrarian productive potential. Establishes that the 2026 season began early and severe in Zone A.
- Sierra de Baza (26 June 2026). "Arranca la temporada de riesgo alto de incendios 2026, que se mantendrá hasta el 15 de noviembre." Declares the current high-risk season officially open.
- BOJA (20 April 2026, Decreto-ley 5/2026). Amendment to the 2023 emergency and forest-fire management framework; INFOCA 2026 mobilisation covers ~5,000 professionals. Establishes the institutional posture heading into the fire season.
- El Tiempo (national). National fire-risk map remains red across much of eastern and southern Spain, with eastern Andalucía specifically named.
- Junta de Andalucía · Portal Ambiental. Índice de Riesgo de Incendio Forestal — AEMET-derived FWI adapted to Andalusian territory. Operational reference.
The triangulation is unusually direct. Three concurrent active fires, all inside the posterior's flagged zones, sit alongside a formal BOJA disaster declaration for a June fire and an INFOCA plan mobilising 5,000 personnel. The model's diagnostic (5.76 ignition-density ratio) is empirically consistent with the observed fire distribution.
6. Who should act, and on what
Junta de Andalucía · Consejería de Sostenibilidad, Medio Ambiente y Economía Azul. INFOCA is already in full mobilisation. The Capybara posterior offers a 21-day horizon add-on to the existing FWI-based daily risk index, at 500 m resolution and per-asset quantification. Priority use: resource pre-positioning for the four flagged zones over the coming three weeks.
Provincial civil-protection · Málaga, Almería, Huelva, Sevilla, Córdoba, Jaén. Zones A, C, D have active fires now. Zone B (Sierra Morena flank) is dispersed but historically a repeat-burn geography and warrants pre-emptive resource staging.
Agricultural extension · IFAPA, Cajas Rurales, ADRs. 132,933 hectares of cropland inside the flagged zone. The Andalusian agricultural mix means olive groves (northeast), vineyards (Málaga hinterland, Jerez), horticulture (Almería), and dry-farm cereal (Sevilla campiña) are all represented. Coordinate with cooperatives on harvest scheduling and stored-fuel management.
Protected-area authorities · Parques Nacionales + Junta de Andalucía. 63,082 hectares of protected area inside the flagged zone. Doñana, Sierra de Cardeña–Montoro, Sierra de Andújar, Sierra de las Nieves, and Cabo de Gata are all named zones. Ranger patrols and firebreak verification warranted.
Renewable-energy operators. One solar farm (of ten mapped) inside the flagged zone. Site verification and vegetation-management review at the specific asset warranted.
7. Method, uncertainty, and falsifiability
Method. The run uses the Capybara legacy-simple flow: PML/V2 GPP anomaly (recent vs 2016 baseline) × SMAP surface-wetness sum + GFS 2 m temperature departure ÷ 100, all multiplied by FIRMS T21 fire-density history over the last three years, percentile-stretched to 0–1. SMAP, GFS, and FIRMS roll forward to the run date; PML GPP is frozen at 2020 in the source catalogue. ENSO baseline is "all years" (2019–2025) for this run.
Uncertainty. Three main sources: (1) GFS forecast skill at 504 h is significantly lower than at 168 h — treat weeks 2 and 3 of the horizon as directional, not point estimates; (2) the frozen PML GPP baseline means the vegetation-productivity signal reflects 2020 conditions, not 2026 land-cover; (3) the "all years" ENSO baseline dilutes the ENSO conditioning — a targeted el_niño or la_niña baseline would sharpen the seasonal comparison.
Falsifiability. Two operational tests over the coming three weeks: (a) the model's July–August area-at-risk trajectory (rising sharply per the seasonal diagnostic) should coincide with rising INFOCA fire counts in Zones A, B, C, D. If July–August fire counts stay flat or fall relative to the 2019–2024 baseline, the model's peak-season forecast for 2026 is not validated. (b) Concurrent active-fire concentration in the high-risk band should remain elevated (ratio > 2.0) throughout the horizon. A drop toward 1.0 would mean the posterior is no longer localising fire correctly.
8. AI-use disclosure and JANUS provenance
This report was drafted under the JANUS framework for AI-assisted qualitative analysis in disaster risk reporting (Feged 2025, DOI 10.21203/rs.3.rs-7580516/v1). Specifically:
- Quantitative claims tagged [MODEL] are traceable to the Earth Engine model run of 2026-07-10T19:07Z, exposure table, and diagnostic panels.
- Qualitative claims tagged [MEDIA:source] are traceable to the named source and were verified by targeted media triangulation on the day of drafting.
- Interpretive claims (zone characterisations, operational recommendations, uncertainty framing) are the analyst's read of the combined evidence, produced with LLM-assisted drafting under the JANUS claim–evidence contract.
- AI-generated content was reviewed by the analyst before publication. No claim was published without an underlying data source or verified media citation. The falsifiability statement in Section 7 is binding.
9. Sources
- BOJA (1 July 2026) — Acuerdo del Consejo de Gobierno: desastre natural, incendio 8 junio 2026, Huelva
- BOJA (20 April 2026) — Decreto-ley 5/2026 — reforma marco emergencias e incendios forestales
- Junta de Andalucía / INFOCA — Incendios en tiempo real
- Junta de Andalucía · Portal Ambiental — Índice de riesgo de incendio forestal
- El Mira (9 July 2026) — Incendio forestal en Málaga activa Plan Infoca, corta AP-7
- Libertad Digital (10 July 2026) — Última hora del trágico incendio de Los Gallardos
- La Voz del Sur — 300+ efectivos INFOCA en Málaga y Almería
- Sierra de Baza (26 June 2026) — Arranca la temporada de riesgo alto de incendios 2026
- El Tiempo · nacional — Mapa de riesgo de incendios en España
- AEMET — Predicción meteorológica de incendios
- Visor incendios 2025–2026 EMA — Dashboard operativo
Method reference. JANUS analytic framework for AI-assisted disaster-risk analysis: Feged (2025), DOI 10.21203/rs.3.rs-7580516/v1. This report was produced under JANUS with AI-assisted drafting and adversarial media triangulation. Quantitative claims are traceable to the model run of 10 July 2026 (run_utc=2026-07-10T19:07Z); qualitative claims are traceable to the media sources cited inline; interpretive claims are the analyst's read. Falsifiability tests in Section 7 are binding.
Contact. Alejandro Feged-Rivadeneira · alejandrofeged@gmail.com · Capybara Center
Note: Only a member of this blog may post a comment.