Annual rain erosion (R) in Brazil (Q13088)

The erosivity data in Brazil. It has a spatial resolution of 30 seconds (~ 1 km). The data set grid is in GeoTIFF format and corresponds perfectly to WorldClim. It uses the geographic coordinate reference system, with WGS84 projection (EPSG: 4326). Soil is a most important non-renewable natural resource for sustaining life. The rates of soil loss have been increasing. The strength of storms can become a disturbing factor, this water energy is known as rain erosivity, and is a major cause of the loss of sediment and nutrients worldwide. The method of obtaining these values is not simple and is usually one-off and uses the USLE or RUSLE equation. Point values cannot be applied in areas that need to estimate soil losses. And traditional spatialization techniques like kriging, IDW or Thiessen polygons do not represent the variability that actually occurs. Thus, the objective of this article was to model a map of rainfall erosivity for Brazil, with spatial resolution of 30 seconds of arc (~ 1 km). Using products made available by other articles, GIS techniques and machine learning modeling. Of the 31 pre-selected covariates 8 were used in the modeling, in order of importance, they were: Longitude, Solar Radiation, Annual precipitation (BIO12), Precipitation of the coldest quarter (BIO19), Wind speed, Precipitation of the warmest quarter (BIO18 ) and the annual reference evapotranspiration. After 400 trainings and validations, the model with the best performance indicators was the Random Forest, using the medians, the indices were: NSE of 0.5823, RMSE of 1567.17 MJ.mm/ha.h.ano, MAE of 1135.90 MJ.mm / ha.h.year, nRMSE of 58.50%, ME of -17.76 MJ.mm/ha.h.year and D of 0.8487. The article was submitted for publication. Dados_Erosividade_BR.csv - Data used to model the models. eros_cubist.tif - Erosivity image generated by the cubist model eros_gbm.tif - Image of erosivity generated by the gbm model eros_lm.tif - Erosivity image generated by the linear model eros_rf.tif - Erosivity image generated by the random forest model