Comprehensive Global Assessment of 23 Gridded Precipitation Datasets Across 16,295 Catchments Using Hydrological Modeling

Ather Abbas, Yuan Yang, Ming Pan, Yves Tramblay, Chaopeng Shen, Haoyu Ji, Solomon H. Gebrechorkos, Florian Pappenberger, JongCheol Pyo, Dapeng Feng, George J. Huffman, Phu Nguyen, Christian Massari, Luca Brocca, Jackson Tan, Hylke E. Beck2025 · DOI

TL;DR

This study presents the most comprehensive global evaluation to date of 24 gridded precipitation datasets by using hydrological modeling across over 18,000 catchments worldwide. The multi-source weighted ensemble precipitation dataset MSWEP V2.8 emerged as the top performer, highlighting the benefits of data fusion and daily gauge corrections.

What Problem Does This Paper Address?

Accurate precipitation data is crucial for a wide range of applications, from water resource management to climate change studies. However, numerous gridded precipitation (P) datasets exist, each with its own strengths and weaknesses derived from satellite, gauge, or reanalysis sources. Selecting the most suitable dataset for a specific application and region is a significant challenge for users. Previous evaluations have often been limited in scope, regional coverage, or the number of datasets assessed, and some have used flawed validation methods. This paper aims to provide a robust, global assessment to help users make informed decisions.

Methods

The core of this study is a large-scale hydrological modeling approach. The authors used the conceptual hydrological model HBV, which is known for its versatility and efficiency. For each of the 18,428 catchments, they calibrated the HBV model using daily streamflow observations and driving it with precipitation data from each of the 24 gridded P datasets. The Kling-Gupta Efficiency (KGE) was used as the primary performance metric, with the calibration score serving as a proxy for the P dataset's ability to accurately represent precipitation dynamics. To ensure fairness, the HBV model was re-calibrated for each P dataset. They also explored different calibration scenarios for bias correction parameters to assess model sensitivity.

Datasets

The study evaluated 24 gridded precipitation (P) datasets, categorized by their data sources:

  • Satellite-only (S): IMERG-Early V7, IMERG-Late V6, IMERG-Late V7, PERSIANN-CCS, PDIR-Now, GSMaP-std V7, GSMaP-std V8, SM2RAIN-ASCAT, SM2RAIN-CCI, GPM + SM2RAIN, CMORPH-CDR, and CMORPH-RT.
  • Reanalysis/Analysis-only (R/A): ERA5, GDAS, and JRA-3Q.
  • Gauge-only (G): CPC Unified and REGEN V1.
  • Satellite and Gauge (S+G): IMERG-Final V7, GPCP V3.2, and PERSIANN-CCS-CDR.
  • Satellite, Reanalysis, and Gauge (S+R+G): CHIRPS V2.0 and MSWEP V2.8.
  • Satellite and Reanalysis (S+R): CHIRP and MSWEP-ng V2.8.

Streamflow observations were compiled from 29 national and international datasets, resulting in a database of 18,428 catchments worldwide that met specific criteria (area < 10,000 km², sufficient record length, and minimal reservoir influence).

Key Insights and Ideas

  • Data fusion reigns supreme: The study strongly supports the idea that combining data from multiple sources (satellite, reanalysis, gauge) and applying daily gauge corrections, as done by MSWEP V2.8, leads to superior hydrological performance. This challenges the notion that a single, high-resolution satellite product is always best.
  • Satellite datasets aren't always superior: Contrary to a common misconception, purely satellite-based datasets performed the worst overall, suggesting that high spatial resolution doesn't automatically translate to better hydrological simulation.
  • Spatial resolution isn't everything: While higher resolution can be beneficial in mountainous regions for capturing orographic precipitation (e.g., GDAS outperforming ERA5), it doesn't guarantee better performance globally, especially when catchment-averaged inputs are used.
  • Climate matters: Precipitation dataset performance varies significantly across different climate zones. Datasets excel in polar regions but struggle in arid regions, where storm characteristics and runoff generation are more sensitive to small errors.
  • Catchment characteristics are key predictors: Aridity index, potential evaporation, and P occurrence were identified as strong predictors of how well a precipitation dataset will perform in a given catchment.

Key Results

  • MSWEP V2.8 was the top performer, achieving a median Kling-Gupta Efficiency (KGE) of 0.78.
  • Among purely satellite datasets, GPM + SM2RAIN performed best with a median KGE of 0.64.
  • The GDAS reanalysis dataset achieved the highest median KGE among (re)analyses at 0.72, slightly outperforming ERA5 (0.71) and JRA-3Q (0.71).
  • Performance was highest in polar (E) regions (median KGE of 0.76) and lowest in arid (B) regions (median KGE of 0.53).
  • Gauge-based datasets generally performed better in flat catchments, while non-gauge-based datasets performed better in steep catchments.

Limitations and Future Work

  • Model Simplification: The HBV model, while widely used, is a parsimonious conceptual model that doesn't explicitly represent human influences like dam operations or groundwater withdrawals, which can significantly alter streamflow.
  • Data Gaps: Despite using a large dataset, some highly populated and vulnerable regions remain underrepresented in streamflow observations.
  • Potential Bias in Evaluation: The overlap between streamflow and meteorological monitoring networks might slightly favor gauge-based datasets and reanalyses over satellite-only datasets.
  • Daily Time Scale: The daily time scale of the assessment might obscure critical sub-daily dynamics, particularly in small catchments and arid regions prone to flash floods. Future research could expand to sub-daily scales.
  • Calibration Masking: The calibration process might mask certain systematic issues in P datasets, such as consistent under- or overestimation of peaks. However, the authors note that these biases are often easily mitigated through post-processing.

The authors suggest that future work should focus on developing P datasets that balance long-term homogeneity, latency, and spatial-temporal coverage to meet diverse user needs. Expanding the analysis to sub-daily time scales is also a promising direction.

Summary updated

Abstract

Numerous gridded precipitation (P) datasets have been developed to address a variety of needs and challenges. However, selecting the most suitable and reliable dataset remains a challenge for users. We conducted the most comprehensive global evaluation to date of gridded (sub-)daily P datasets using hydrological modeling. A total of 23 datasets, derived from satellite, model, gauge sources, or their combinations thereof, were assessed. To evaluate their performance, we calibrated the conceptual hydrological model HBV against observed daily streamflow for 16,295 catchments (each &lt; 10, 000 km2) world- wide, using each P dataset as input. The Kling-Gupta Efficiency (KGE) was used as the performance metric and the calibration score served as a proxy for P dataset performance. Overall, MSWEP V2.8 demonstrated the highest performance (median KGE of 0.75), highlighting the value of merging P estimates from diverse data sources and applying daily gauge corrections. Among the purely satellite-based P datasets, the soil moisture- and microwave-based GPM+SM2RAIN dataset performed best (median KGE of 0.60), while the JRA-3Q reanalysis ranked highest among the purely model-based datasets (median KGE of 0.67), outperforming the widely used ERA5 reanalysis (median KGE of 0.59). Performance varied across Köppen-Geiger climate zones, with the best results in polar (E) regions (median KGE of 0.74 across datasets) and the lowest in arid (B) regions (median KGE of 0.33 across datasets). We further examined the spatial relationships between catchment attributes and KGE scores, identifying potential evaporation, air temperature, solid P fraction, and latitude as the strongest predictors of performance. Our analysis revealed significant regional differences in dataset performance and heterogeneity in P error characteristics, underscoring the critical importance of careful dataset selection for water resource management, hazard assessment, agricultural planning, and environmental monitoring.
BibTeX
@misc{abbas2025comprehensive,
  author = {Ather Abbas and Yuan Yang and Ming Pan and Yves Tramblay and Chaopeng Shen and Haoyu Ji and Solomon H. Gebrechorkos and Florian Pappenberger and JongCheol Pyo and Dapeng Feng and George J. Huffman and Phu Nguyen and Christian Massari and Luca Brocca and Jackson Tan and Hylke E. Beck},
  title = {{Comprehensive Global Assessment of 23 Gridded Precipitation Datasets Across 16,295 Catchments Using Hydrological Modeling}},
  year = {2025},
  doi = {10.5194/egusphere-2024-4194},
}
Figure