Precipitation Data Products for South America
South America's diverse precipitation regimes — ranging from the hyper-arid Atacama Desert to rainforests receiving up to 8,000 mm annually — require careful selection of gridded precipitation data products. Widely used global products include CHIRPS, GPM IMERG, ERA5 and ERA5-Land, MSWEP, MERRA-2, GSMaP, PERSIANN, SM2RAIN-ASCAT, TerraClimate, and NASA POWER, alongside regionally tailored products such as MERGE (Brazil/South America), PISCO (Peru), RAIN4PE (Peru and Ecuador), CR2MET (Chile), SQPE-OBS (Argentina), and SC-PREC4SA (continent-wide station-based). The Atlantic Cloud data platform now provides programmatic access to 397 million quality-controlled rainfall records from over 21,000 rain gauges across Brazil spanning 1885 to 2025, via the open-source AtlanticCloud.jl Julia package.[^c5]
Major version upgrades have recently been released or announced. CHIRPS version 3 was released on January 1, 2025, expanding coverage to 60°N–60°S, integrating over 90 station data sources (nearly four times the number in v2), adopting a variance-preserving quadratic CCD algorithm, and applying gauge-undercatch correction.[^c4][^c14] The CHIRPS v3 methodology and validation are documented in a peer-reviewed paper published in Scientific Data in 2026.[^c27] MSWEP version 3 (2026) is the first fully global machine-learning-based gridded precipitation product achieving the highest reported KGE among 19 evaluated products, though a known trend artifact in its input data affects the 2000–2015 period. GPM IMERG V08 targets Q3 2026 as the final major upgrade with ML-based GPROF, surface-type-dependent calibration, and improved CORRA, while V07B (January 2024) corrected 162 defective GPROF orbits.[^c12][^c15] New high-resolution national datasets have also emerged: BRain-D (Brazil, 1961–2024, with automatic quality control classifying stations as high- or low-quality[^c18][^c19]), UNIPLU-BR (Brazil, 1885–2025, 2.2 billion station records), CR2MET and PatagoniaMet (Chile), SQPE-OBS (experimental satellite-gauge blend for southern South America), and METBRA25Y (hourly meteorological archive for Brazil from 616 INMET stations, 2000–2025, with 166 million+ observation rows processed through a two-stage quality-control pipeline). RAIN4PE, developed for Peru and Ecuador, uses random forest fusion and SWAT-based reverse-hydrology correction to achieve a median KGE of 0.81 (published paper) to 0.85 (dissertation) with values ranging from 0.65 to 0.96, and monthly NSE of 0.82 across 72 evaluated watersheds, outperforming all compared global products in streamflow simulation.[^c6] The B-EPICC project that funded RAIN4PE concluded in January 2024 after six years, and no updated version has been released subsequently. SM2RAIN-ASCAT has been extended to December 2022 (v2.1.2n), providing a bottom-up satellite rainfall estimate from ASCAT soil moisture observations. A comprehensive global assessment of 23 gridded precipitation datasets using the HBV hydrological model across 16,295 catchments found that MSWEP V2.8 achieved the highest overall performance (median KGE 0.75), while the soil moisture-based GPM+SM2RAIN product led among purely satellite-based products (median KGE 0.60) and JRA-3Q among reanalysis-based products (median KGE 0.67), outperforming ERA5 (median KGE 0.59).[^c29][^c30][^c31]
Gridded precipitation products differ substantially in their estimates across the continent. A multi-product comparison over Brazil and Colombia found pixel-level differences exceeding 75% between ERA5-Land and CHIRPS.[^c3] CHIRPS demonstrated the best overall performance among six evaluated global products across both countries, while reanalysis products struggled in tropical settings. In extratropical Argentina, the CPC gauge-based analysis outperformed CHIRPS for extreme precipitation indices in the Sierras Australes Bonaerenses, though neither product was suitable for complex terrain extreme event assessment.[^c28] ERA5-Land underestimated dry-season precipitation by up to 40% relative to BR-DWGD in the Brazilian Cerrado, propagating into significant differences in modelled soil moisture and deep drainage.[^c8] A multi-catchment study across 95 South American catchments found that ERA5 departs substantially from observation-based estimates with systematic biases and physically implausible water-storage change patterns.[^c10] MERRA-2 showed the smallest biases among reanalyses in continent-scale water cycle assessments, while ERA5 followed closely.[^c2] MERGE outperformed CHIRPS (r=0.96 vs 0.85) in Northeast Brazil, confirming its value for the region.[^c13] SC-PREC4SA (2025) integrates records from 7,794 weather stations through quality control and machine-learning gap-filling, making it the most comprehensive serially complete daily precipitation dataset for South America.[^c1] In the Paraná III Watershed around the Itaipu Reservoir, CHIRPS showed the highest accuracy for mean precipitation and multi-day extremes, while ERA5-based products better captured long-term drying trends.[^c16][^c17] In Uruguay, GSMaP achieved the highest daily accuracy among five tested products, with R² = 0.73 and RMSE = 5.94 mm, outperforming IMERG, CHIRPS, MERGE CPTEC, and NASA POWER.[^c20] In Ecuador's Ambato River Basin, IMERG was better for daily rainfall detection (suited for flood modeling) while CHIRPS delivered more stable dry-period performance with fewer false alarms (preferred for drought and climatological studies).[^c21][^c22] MSWEP performance is highly region-dependent: it is the top global product at daily scales across Brazil but strongly overestimates in Northeast Brazil's Mearim basin. In the Tambo Basin of southern Peru, PERSIANN-CDR outperformed ERA5 and MERRA-2 in a three-product comparison, showing the highest correlation (CC 0.84–0.94) and lowest bias against nine SENAMHI stations.[^c24]
In the Vilcanota Basin of Peru, a 2025 study evaluating 11 satellite and reanalysis precipitation products found that MSWEP, CHIRPS, TRMM-3B42, and PERSIANN-CDR were the most efficient at representing daily and accumulated precipitation variability in this Andean catchment.[^c25] In Venezuela, CHIRPS v2 validation against 154 rain gauges showed overestimation of low monthly rainfall and underestimation of high values, with best performance in flat open regions (Los Llanos) and significant degradation in complex terrain; station availability for CHIRPS blending declined sharply over the study period.[^c26]
The South Atlantic Convergence Zone (SACZ) is the dominant large-scale rainfall-producing system in Brazil during austral summer, forming a NW-SE band of clouds extending from the Amazon across southeastern Brazil into the Atlantic.[^c7] The South American Monsoon System (SAMS) underwent a significant regime shift in the early 1970s, with mean duration increasing from approximately 170 days (1948–1972) to approximately 195 days (1972–1982) and onset and demise dates shifting correspondingly.[^c11] The four main climatic processes driving precipitation anomalies across Tropical South America are the intensity and position of the SACZ, ENSO, the meridional position of the ITCZ, and the strength of the South American Low-Level Jet. IMERG V07B reduces errors compared to V06B in high-rainfall areas of South America, though it exacerbates underestimation in warm-cloud regions of Northeast Brazil during winter.[^c9] Emerging machine-learning-based bias correction methods — including conditional diffusion models and differentiable parametric frameworks — offer improved treatment of extreme events compared to traditional quantile mapping.
ENSO teleconnections are the dominant driver of interannual precipitation variability across South America. Observational streamflow data over 1979–2023 show flood probabilities increase by more than 120% during El Niño in the La Plata Basin and during La Niña in the northern Amazon. Under CMIP6 projections, warming over all regions very likely exceeds 1.5°C before mid-century, with southeastern South America projected to become wetter (high confidence) and northeastern South America drier (high confidence).