A comparison of missing values imputation methods applied to precipitation of two semi-arid and humid regions of México

Abstract

Climatological data with unreliable or missing values is an important area of research, and multiple methods are available to fill in missing data and evaluate data quality. Our study aims to compare the performance of different methods for estimating missing values that are explicitly designed for precipitation and multipurpose hydrological data. The climate variable used for the analysis was daily precipitation. We considered two different climate and orographic regions to evaluate the effects of altitude, precipitation regime and percentage of missing data on the Mean Absolute Error of imputed values and using a homogeneity evaluation of meteorological stations. We excluded from the analysis meteorological stations with more than 25% missing data. In the semi-arid region, ReddPrec (optimal for 9 stations), and GCIDW (optimal for 8) were the best performing methods for the 23 stations, with average MAE values of 1.63 mm/day and 1.46 mm/day, respectively. In the humid region, GCIDW was optimal in ~59% of stations, EM in ~24%, and ReddPrec in ~17%, with average MAE values of ~6.0 mm/day, 6.5 mm/day and ~9.8 mm/day, respectively. This research makes an important contribution to identifying the most appropriate methods to impute daily precipitation in different climatic regions of Mexico based on efficiency indicators and homogeneity evaluation.

Influence of climate variability on change in storage of overexploited aquifers in a semi-arid region

“…An analysis was performed to discard outliers from the time series. The missing data were imputed following a methodology proposed by Morales et al (2019) and the results of comparing different methods in different climate regions can be found in Navarro Céspedes et al (2023). From monthly precipitation values, the standardized precipitation index was calculated for a time scale of 6 months (SPI6).…”