Mohammadiigder et al. (2025) Evaluation of differences between gridded precipitation products in the Southern Prairies of Manitoba

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-10-07
• Authors: Omid Mohammadiigder, Chandra Rupa Rajulapati, Ricardo Mantilla, Fisaha Unduche
• DOI: 10.1016/j.ejrh.2025.102786

Research Groups

• Department of Civil Engineering, University of Manitoba, Winnipeg, MB, Canada
• Executive Director of Hydrologic Forecasting and Water Management, Government of Manitoba, Winnipeg, MB, Canada

Short Summary

This study evaluates the discrepancies among seven daily gridded precipitation datasets against 103 independent gauge observations in Southern Manitoba from 2018 to 2023. It finds that MRMS and CaPA are the top-performing products with the lowest errors and highest correlations across seasons, while other products exhibit varying seasonal biases and accuracy.

Objective

• To evaluate the accuracy and error characteristics of seven state-of-the-art gridded precipitation products (MRMS, CaPA, ERA5-L, NLDAS-2, PERSIANN, GSMAP, and GPM) against an independent rain gauge network in the Southern Prairies of Manitoba, Canada.
• To analyze the spatial, seasonal, monthly, and quantile variations of errors in these products, particularly focusing on systematic biases, non-stationary temporal variations, and event-related discrepancies relevant to hydrological modeling.

Study Configuration

• Spatial Scale: Southern Manitoba, Canada, covering approximately 100,000 square kilometers (49.00–52.35°N, 95–101°W).
• Temporal Scale: Daily precipitation data from 2018 to 2023.

Methodology and Data

• Models used:
  • MRMS (Multi-Radar Multi-Sensor)
  • CaPA (Canadian Precipitation Analysis)
  • ERA5-Land (Fifth Generation European Centre for Medium-Range Weather Forecasts)
  • NLDAS-2 (North American Land Data Assimilation System)
  • GPM-IMERG V06 (Integrated Multi-satellitE Retrievals for Global Precipitation Measurement)
  • PERSIANN-CDR (Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks-Climate Data Record)
  • GSMaP-V6 (Global Satellite Mapping of Precipitation)
• Data sources:
  • Reference: Observations from 103 independent Pluvio2 weighing rain gauges managed by Manitoba Agriculture.
  • Evaluation metrics: Root Mean Square Error (RMSE), Relative Bias (RBias), Mean Error (ME), Correlation Coefficient (CC), Probability of Detection (POD), False Alarm Ratio (FAR), and Critical Success Index (CSI).
  • Evaluation approach: Point-to-pixel comparison at native spatial resolutions of the gridded products, using the nearest-neighbor matching technique.

Main Results

• Overall Performance: MRMS and CaPA consistently demonstrate superior performance with the lowest RMSE (e.g., median RMSE for CaPA ~3.18 mm/day, MRMS ~3.03 mm/day) and highest correlation coefficients (median CC ~0.76) across all seasons.
• Seasonal Variability:
  • Winter (December-February): Most products show high positive relative bias (80 % to 300 %), indicating systematic overestimation (e.g., ERA5-L 199 %, GPM 161 %, CaPA 137 %), despite low RMSE values (around 1 mm/day) due to low observed precipitation. GSMAP consistently underestimates precipitation.
  • Summer (June-August): Products exhibit the largest RMSE values (up to 9.78 mm/day for GSMAP), driven by intense convective storms. Relative biases are generally lower (e.g., MRMS 11.2 %, ERA5-L 9.15 %), but MRMS, GPM, GSMAP, PERSIANN, and ERA5-L tend to overestimate, while CaPA and NLDAS-2 show underestimation in northern regions.
  • Spring (March-May) and Fall (September-November): Moderate RMSE and RBias values, with most products performing more consistently.
• Product-Specific Biases:
  • ERA5-L and GPM: Tend to overestimate precipitation, especially in winter and fall.
  • NLDAS-2: Shows spatial variability, overestimating in the north during winter and underestimating in summer.
  • PERSIANN: Exhibits poor temporal consistency (CC < 0.57) and overestimates, particularly in winter.
  • GSMAP: Consistently underestimates in winter and significantly overestimates in summer (RBias > 40 %, median RMSE ~10 mm/day).
• Categorical Metrics: CaPA and MRMS consistently outperform other products in detecting precipitation events across most intensity thresholds (e.g., CSI values 0.5-0.65 for events < 10 mm/day). Products struggle to accurately capture extreme events (≥50 mm/day).
• Spatial Resolution Impact: Finer spatial resolution products (MRMS at 0.01°, CaPA at 0.02°) generally correlate with improved performance across error metrics (higher CC, lower RMSE).

Contributions

• Provides a comprehensive, region-specific evaluation of seven widely used gridded precipitation products in the Southern Prairies of Manitoba, a region characterized by sparse gauge data and high hydrological importance.
• Utilizes an independent network of 103 rain gauges for robust validation, addressing a gap in existing literature where products are often evaluated against data used in their calibration.
• Offers detailed insights into the spatial, seasonal, monthly, and intensity-dependent error characteristics of these products, highlighting their strengths and limitations for hydrological applications.
• Emphasizes the critical need to account for precipitation uncertainty and preserve temporal structure in hydrological modeling, suggesting advanced techniques beyond standard bias correction methods like Q-Q mapping.

Funding

• Manitoba Transportation and Infrastructure (all authors)
• Natural Sciences and Engineering Research Council of Canada (NSERC Discovery Grant: RGPIN-2024-06417 for C.R.R.)
• Natural Sciences and Engineering Research Council of Canada (NSERC Discovery Grant: RGPIN-2023-04724 for R.M.)

Citation

@article{Mohammadiigder2025Evaluation,
  author = {Mohammadiigder, Omid and Rajulapati, Chandra Rupa and Mantilla, Ricardo and Unduche, Fisaha},
  title = {Evaluation of differences between gridded precipitation products in the Southern Prairies of Manitoba},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.102786},
  url = {https://doi.org/10.1016/j.ejrh.2025.102786}
}