Query
Agent Pipeline
1
Interpret the query
LLM → JSON plan {analysis, zone, days}
2
Fetch context from DataHub (MCP)
search + schema + lineage for NDVI
3
Fetch data (synthetic ↔ GEE)
Sentinel-2 over ~2 km bbox, auto-generated code
4
Compute indices and trend
NDVI, % below threshold, Δ, state
5
Write-back with lineage
7 datasets + tendencia, upstreamLineage
6
Report + Telegram alert
formatted report + map + dashboard
Console
Context details (DataHub MCP)
search: db=ndvi, platform=sentinel229 datasets
series detected (Lima)—
schema: bandndvi (float32)
registered lineage—
NDVI Map — Lima day 7/7
Low NDVIHealthy NDVI
Trend of the series
% area below threshold (τ)
day 1 → day 7—
state—
Analysis math
NDVI = NIR − REDnumerator
/ NIR + REDdenominator
Pt = degraded fraction (NDVI < τ)—
Δ = Pt0 − PtN (percentage points)—
ruleALERT if P≥50 or Δ>+5 · WATCH if Δ>0
DataHub lineage (write-back)
raster_ndvi_lima_2026-08-04urn:datahub:dataset:terra.raster_ndvi_lima
raster_ndvi_lima_2026-08-05urn:datahub:dataset:terra.raster_ndvi_lima
… +5 more days7-day time series
anaalysis_ndvi_tendencia_2026-08-10urn:datahub:dataset:terra.anaalysis_ndvi_tendencia
informe_ndvi_2026-08-10visualization + Telegram
Simulated Telegram alert — @ERRADICADORBOT
📡 Terra Cognita — Daily report
🔔 Agent report — waiting for execution…
Window: 2026-08-04 → 2026-08-10 (7 days) ⏳
Window: 2026-08-04 → 2026-08-10 (7 days) ⏳