Optimized Climate Dataset for Natal and São Paulo (OCD-NSP)
Description
This database comprises time series of maximum temperature, minimum temperature, and precipitation for Natal and São Paulo for the period from 1991 to 2020 by integrating data from the Climate Hazards Center InfraRed Precipitation with Station data (CHIRPS), Climate Hazards Center InfraRed Temperature with Stations (CHIRTS), and the Brazilian National Institute of Meteorology (INMET) using Linear Regression with Moving Windows (LReg-MW). The method adjusts the local relationship between satellite-derived and station-observed data and uses this relationship to fill missing values. For the quantitative evaluation of the LReg-MW method, this study employs six main statistical metrics: Bias, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Scaled Error (MASE), Pearson correlation coefficient (r), and Standard Deviation (SD). The model achieved its best results for temperatures in São Paulo. Maximum and minimum temperatures exhibited very high correlations (0.947 and 0.903, respectively) and high coefficients of determination (R² = 0.896 and 0.816). The MASE for maximum temperature (0.443) demonstrates excellent predictive accuracy. The model also performed very well in Natal, particularly for maximum temperature, which recorded the lowest Mean Absolute Error (MAE = 0.523 °C) and a strong correlation (r = 0.800). Because precipitation is a spatially heterogeneous phenomenon that is inherently difficult to model, the errors (RMSE and MAE) are naturally higher compared with those obtained for temperature, while the correlations are moderate (r = 0.612 in Natal and r = 0.527 in São Paulo). Nevertheless, MASE values below 1 indicate that the model-based imputation outperforms simple conventional methods. The strongest result across all statistical metrics was observed for Bias. All values were extremely close to zero, ranging from −0.059 to 0.015, demonstrating that the resulting database is well balanced and exhibits minimal systematic bias. Each table contains the columns referring to Station (meteorological station identification), X (sequential identifier of the observation), date (Date of the observation), variável_inmet (It is the reference/observation variable), latitude (Latitude of the station), longitude (Longitude of the station), altitude (Altitude of the station), variável_chirps or variável_chirts (daily variable from satellite data), ym (Year-month index in YYYYMM format), pred_rolling (Value of the variable estimated by LReg-MW for each day in which CHIRPS/CHIRTS is available and there is a window with at least 30 paired observations) and variável_filled (Final series of the variable. It keeps the variável_inmet when it exists and uses pred_rolling only to fill in the missing INMET values).