Huerta et al. (2025) Enhancing daily precipitation reconstruction: An improved version of the reddPrec R package

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-10-10
• Authors: Adrian Huerta, Stefan Brönnimann, Martín de Luis, Santiago Beguerı́a, Roberto Serrano‐Notivoli
• DOI: 10.1016/j.envsoft.2025.106717

Research Groups

• Institute of Geography, University of Bern, Bern, Switzerland
• Oeschger Centre for Climate Change Research, Bern, Switzerland
• Departamento de Geografía y Ordenación del Territorio. Instituto Universitario de Ciencias Ambientales (IUCA), University of Zaragoza, Zaragoza, Spain
• Estacion Experimental de Aula Dei, Consejo Superior de Investigaciones Científicas (EEAD-CSIC), Zaragoza, Spain

Short Summary

This paper introduces an improved version of the reddPrec R package for daily precipitation reconstruction, featuring enhanced quality control, homogenization, and flexible machine learning models with dynamic covariates. Case studies in Switzerland and Spain demonstrate its superior accuracy in gap-filling and grid creation, and its effectiveness in detecting and adjusting data inhomogeneities.

Objective

• To present an improved version of the reddPrec R package that enhances the reconstruction of high-quality daily precipitation series by introducing greater flexibility in spatial modeling, dynamic covariates, and new modules for enhanced quality control and homogenization.

Study Configuration

• Spatial Scale: Daily precipitation grids at approximately 1 km (0.009°) resolution for Switzerland and the province of Valencia, Spain; station-level time series.
• Temporal Scale: Daily precipitation series; gap-filling evaluation for 2010–2015; extreme precipitation events on July 24–28, 2014, and October 29, 2024; homogenization evaluation and long-term trend analysis for 1960–2015.

Methodology and Data

• Models used: reddPrec R package (Version 3.0.0); Reference Values (RVs) computed via classification and regression functions; Statistical/Machine Learning models including Generalized Linear Models (glm), Support Vector Machines (SVM), Random Forests (rf), Extreme Gradient Boosting (xgboost), and Neural Networks (nn); Homogenization using five univariate breakpoint tests (Student’s t-test, Mann–Whitney, Buishand R, Pettitt, Standard Normal Homogeneity Test) and a quantile-matching adjustment technique.
• Data sources: Weather station data from the Swiss National Basic Climatological Network (Swiss NBCN) and the National Meteorological Service of Spain (AEMET); MODIS Aqua/Terra surface reflectance bands 1 and 2 (dynamic covariates); EarthEnv topographic data (static covariates: latitude, longitude, elevation, principal components of topographic features); Reference products for evaluation: RhydchprobD (Switzerland) and CSIC reference grid (Spain); Synthetically corrupted Swiss NBCN dataset for homogenization testing.

Main Results

• Random Forest (rf) and Extreme Gradient Boosting (xgboost) models consistently outperformed Generalized Linear Models (glm) in gap-filling and grid creation, showing an average increase of approximately 0.05 in Matthews Correlation Coefficient (mcc) for dry/wet day classification and modest improvements in the refined index of agreement (dr) for wet-day precipitation amounts.
• The enhanced quality control framework effectively detected subtle data issues such as truncation, small gaps, weekly cycles, and precision/rounding patterns, classifying most Swiss NBCN stations as high quality (Level 0) while revealing more frequent issues in the Aragón (Spain) dataset.
• The homogenization method demonstrated strong performance in detecting artificial breakpoints (Break Detection Accuracy > 0.7, high Timing Accuracy within ±1 year) and significantly improved agreement with original uncorrupted data (mean mcc and dr > 0.7), while achieving moderate to high success in preserving short-term trends (mcc around 0.7 for PRCPTOT and 0.5 for R1 mm; dr around 0.5).

Contributions

• Introduces a new version of the reddPrec R package with enhanced flexibility in spatial modeling through user-defined machine learning models and the integration of dynamic covariates.
• Develops a comprehensive, modular, and automated enhanced quality control system to detect systematic errors and temporal inconsistencies in daily precipitation series.
• Incorporates a novel homogenization framework specifically adapted for daily precipitation datasets, combining relative and absolute approaches to detect and adjust inhomogeneities while preserving trends.
• Provides a robust and reliable tool for reconstructing high-quality daily precipitation series, applicable in data-sparse environments and heterogeneous terrain, supporting the creation of reproducible datasets for climate research.

Funding

• Swiss Government Excellence Scholarships for Foreign Scholars (ESKAS-Nr: 2023.0404)
• Grant RYC2021-034330-I funded by MCIN/AEI/10.13039/501100011033 and by ‘‘European Union NextGenerationEU/PRTR’’

Citation

@article{Huerta2025Enhancing,
  author = {Huerta, Adrian and Brönnimann, Stefan and Luis, Martín de and Beguerı́a, Santiago and Serrano‐Notivoli, Roberto},
  title = {Enhancing daily precipitation reconstruction: An improved version of the reddPrec R package},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106717},
  url = {https://doi.org/10.1016/j.envsoft.2025.106717}
}