Switanek et al. (2026) Leveraging normalized data to improve point-scale estimates of precipitation–temperature scaling rates

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-03-31
• Authors: Matthew B. Switanek, Jakob Abermann, Wolfgang Schöner, Michael L. Anderson
• DOI: 10.5194/hess-30-1719-2026

Research Groups

• Department of Geography and Regional Science, University of Graz, Graz, Austria
• California Department of Water Resources, Sacramento, California, United States

Short Summary

This study proposes and evaluates a methodology using normalized precipitation and dew point temperature data to improve point-scale estimates of precipitation-temperature scaling rates, demonstrating that normalization effectively accounts for spatio-temporal climatological variability and enhances prediction skill in the Upper Colorado River Basin.

Objective

• To develop and evaluate a methodology that more accurately estimates point-scale precipitation-temperature (P-T) scaling rates, particularly by addressing challenges related to pooling raw data and statistical independence, and to use these improved estimates for skillful predictions of extreme precipitation.

Study Configuration

• Spatial Scale: Point-scale station data from the Upper Colorado River Basin (UCRB). Data pooling considered at: individual station, 50 km radius, 100 km radius, and the entire UCRB region.
• Temporal Scale: Hourly and daily extreme precipitation (Rx1hr, Rx1day) and dew point temperature data from 1 January 1951 to 31 December 2024. Temporal pooling windows considered: 1-month, 3-month, 5-month, and all 12 calendar months.

Methodology and Data

• Models used:
  • Exponential regression model (y = a * b^x) fitted to monthly average dew point temperature anomalies (x) and normalized/non-normalized precipitation anomalies (y).
  • Comparison models: Climatology (100% of normal), fixed Clausius-Clapeyron (C-C) scaling rate (7% °C⁻¹), non-normalized data model, and normalized data model.
  • Cross-validation frameworks: Leave-one-year-out and two-fold cross-validation.
  • Model performance evaluated using the Root Mean Squared Error Skill Score (SS_RMSE).
• Data sources:
  • Global Historical Climatology Network – hourly (GHCN-hourly or GHCNH) dataset for hourly precipitation and concurrent dew point temperature.
  • Global Historical Climatology Network – daily (GHCN-daily or GHCND) dataset for daily precipitation.
  • ERA5 Reanalysis dataset for daily dew point temperatures (nearest grid cell to GHCND stations).
  • Data quality controlled to remove spatial and temporal outliers.
  • Normalized anomalies of dew point temperature and precipitation (Rx1hr, Rx1day) computed relative to station/month climatological means.

Main Results

• Pooling raw (non-normalized) data across multiple stations and/or months leads to inaccurate P-T scaling rate estimates due to spatio-temporal climatological differences and inconsistent sampling.
• Hourly precipitation data exhibits significant temporal autocorrelation, with events separated by less than 4 hours not being statistically independent. For extreme events, statistical independence is approached at approximately 12 hours for the top 0.1% and 24 hours for the top 1%.
• Using monthly average dew point temperatures as a predictor for Rx1hr and Rx1day precipitation shows a stronger statistical relationship and predictive power compared to concurrent hourly dew point temperatures.
• The normalized data model significantly outperforms non-normalized models, climatology, and fixed C-C scaling in cross-validated predictions of extreme precipitation.
• Optimal model performance for Rx1hr predictions (SS_RMSE = 0.71) was achieved using normalized data with a 100 km spatial window and a 3-month temporal window.
• Optimal model performance for Rx1day predictions (SS_RMSE = 0.81 for leave-one-year-out; 0.76 for two-fold cross-validation) was achieved using normalized data with a 50 km spatial window and a 3-month temporal window.
• Non-normalized models show best performance with very narrow spatial and temporal extents (e.g., single station/month) and degrade significantly when data is pooled.
• Estimated P-T scaling rates exhibit substantial spatio-temporal variability across the UCRB:
  • Median scaling rates for both Rx1hr and Rx1day in winter months are typically sub-Clausius-Clapeyron (< 7% °C⁻¹).
  • Median scaling rates in summer months are typically super-Clausius-Clapeyron (> 7% °C⁻¹), with some stations showing rates approximately triple the C-C value.
  • Median Rx1hr scaling rates range from 4.6% °C⁻¹ (January) to 17.0% °C⁻¹ (June).
  • Median Rx1day scaling rates range from 5.2% °C⁻¹ (December) to 12.9% °C⁻¹ (June).

Contributions

• Identifies and quantifies two primary challenges in estimating point-scale P-T scaling rates: the adverse impact of pooling raw data due to spatio-temporal climatological variability and the lack of statistical independence in high-resolution data.
• Proposes and validates a methodological improvement by using normalized precipitation and monthly average dew point temperature data within an exponential regression framework.
• Demonstrates that data normalization allows for more effective leveraging of pooled data, leading to significantly improved and more robust estimates of P-T scaling rates and predictions of extreme precipitation.
• Provides quantitative evidence (SS_RMSE values) of the superior predictive skill of the normalized data approach compared to non-normalized models, fixed C-C scaling, and climatology.
• Reveals the substantial spatio-temporal variability of P-T scaling rates in the UCRB, highlighting the necessity for localized and seasonally-specific estimations for accurate climate impact assessments.

Funding

• California Department of Water Resources (contract no. 4600015149)
• University of Graz

Citation

@article{Switanek2026Leveraging,
  author = {Switanek, Matthew B. and Abermann, Jakob and Schöner, Wolfgang and Anderson, Michael L.},
  title = {Leveraging normalized data to improve point-scale estimates of precipitation–temperature scaling rates},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-1719-2026},
  url = {https://doi.org/10.5194/hess-30-1719-2026}
}