Data and code for: Disentangling climate sensitivity from management and place in Amazonian açaí production

Description

Data and code for "Disentangling climate sensitivity from management and place in Amazonian açaí production: an explainable machine-learning attribution with leakage-safe validation". This dataset supports a climate attribution and sensitivity study of açaí (Euterpe oleracea Mart.) production in the Eastern Amazon. It provides the assembled municipal panel, the climate-projection results, and the full analysis code needed to reproduce the study, which is framed as attribution rather than yield prediction. The production panel combines Brazilian official statistics from IBGE/SIDRA: extractive fruit output (PEVS, table 289; 1986 to 2024) and planted cultivation (PAM, table 5457; 2015 to 2024), covering 466 municipalities and 9,094 municipality-year records. A modeling panel pairs these targets with climate predictors aggregated into four phenological windows relative to the December harvest reference (October to December, July to September, April to June, and January to March), using variables of vapour-pressure deficit, mean, maximum and minimum temperature, relative humidity, solar radiation, wind speed, and precipitation. Municipality centroid coordinates are included for the 359 extractive municipalities. The projection file reports the ceteris paribus water-channel response of extractive production under a five-model NEX-GDDP-CMIP6 ensemble (ACCESS-CM2, EC-Earth3, MPI-ESM1-2-HR, MRI-ESM2-0, NorESM2-MM) for SSP2-4.5 and SSP5-8.5 at the 2041 to 2060 and 2061 to 2080 horizons. The raw sources are public: production from IBGE/SIDRA; climate predictors from TerraClimate, CHIRPS and NASA POWER; and climate projections from NASA NEX-GDDP-CMIP6, accessed through the NASA NCCS THREDDS NetcdfSubset Service. This record redistributes the derived panel and results together with the code so the analysis can be reproduced. The code (Python) covers phenological-window feature construction, baseline models, the nested variance decomposition that isolates the partial contribution of climate against a climate-free control, blocked cross-validation (leave-one-year-out, spatial KMeans blocks, and leave-one-state-out), SHAP attribution, and the CMIP6 extraction and projection pipeline. Data files are released under CC BY 4.0 and code under the MIT License; the underlying public datasets remain under the terms of their original providers.

Institutions

• Universidade Federal Rural da Amazônia

  Pará, Belém
• Universidade Federal de Lavras

  Minas Gerais, Lavras
• Universidade Federal de Goiás

  Goiás, Goiânia

Categories

Licence