The Amazon makes much of its own rain. Water transpired by the canopy rises, blows west as “flying rivers,” and falls again — over and over across the basin. Clear the forest and that conveyor breaks, opening dry pockets that can spread.
The trouble: rain gauges are sparse. This tool takes real rainfall from 57 stations (ERA5 reanalysis) and interpolates them into a continuous surface, so the pockets hiding between stations become visible.
Rainfall & air temperature are real ERA5 reanalysis (via Open-Meteo) at real station sites; geography is real (Natural Earth + HydroBASINS divide). Forest cover is the one modelled layer. Educational tool — not a climate or land-use study.
What is real observation-based data, and what is modelled — stated plainly.
Geography is real public-domain vector data. Coastline and national borders for the whole continent come from Natural Earth (1:50m); the river network (Amazonas, Negro, Madeira, Tapajós, Xingu, Purus, Japurá, Branco, Ucayali, Marañón…) is Natural Earth, clipped to the basin. The Amazon-basin boundary is the real hydrological divide from HydroSHEDS HydroBASINS (level-3), simplified for size — which is why Bélem and Cuiabá fall outside it (they drain the separate Tocantins and Paraguay basins).
Rainfall and air temperature are real observations — ERA5 reanalysis (2021–2023 monthly climatology) sampled at the 57 real station coordinates via the Open-Meteo Historical Weather API. Every rainfall and temperature surface you see is an interpolation of genuine observation-based data. The only modelled layer is forest cover (no open point source was wired); it is built from a deforestation-arc model, so treat the canopy layer — and the forest term in pocket detection — as illustrative. Gauge values are editable: import your own CSV to interpolate a different network.
The rainfall surface offers two interpolators: IDW (inverse-distance weighting — each cell a weighted average of nearby gauges, weight = 1 / distanceᵖ, with adjustable power p and search radius) and ordinary kriging (a Gaussian variogram model with adjustable range and nugget). Switching between them shows how the choice of method — not just the data — shapes where pockets appear.
Each cell is scored by combining three signals: interpolated rainfall anomaly (vs the basin mean), forest cover, and surface temperature. Wet pockets = wet + intact + cool (the recycling engine worth protecting). Dry pockets = below-norm rain + low forest + hot (where clearing is breaking the cycle). Outlines are iso-contours via marching squares.
The streamlines follow a climatological moisture-flux field — Atlantic easterlies sweeping west, then recurving south against the Andes (the South American Low-Level Jet). Each particle's moisture is a live budget: it bleeds where the interpolated rainfall sits below the basin norm and where forest cover is low, and recharges over wet, intact canopy. So the rivers visibly run dry over the pockets you've detected, and respond to the season, drying trend, and interpolation settings. The flux direction is illustrative climatology, not a reanalysis wind product.