Bhuiyan et al. (2018) A nonparametric statistical technique for combining global precipitation datasets: development and hydrological evaluation over the Iberian Peninsula

Identification

• Journal: Hydrology and earth system sciences
• Year: 2018
• Authors: Md Abul Ehsan Bhuiyan, Efthymios I. Nikolopoulos, Emmanouil N. Anagnostou, Pere Quintana Seguí, Anaïs Barella-Ortiz
• DOI: 10.5194/hess-22-1371-2018

Research Groups

• Department of Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA
• Innovative Technologies Center S.A., Athens, Greece
• Ebro Observatory, Ramon Llull University – CSIC, Roquetes (Tarragona), Spain
• Castilla-La Mancha University, Toledo, Spain

Short Summary

This study develops and evaluates a nonparametric statistical technique, Quantile Regression Forests (QRF), to optimally combine multiple global precipitation datasets (satellite and reanalysis) and characterize their uncertainty over the Iberian Peninsula. The QRF-generated ensemble significantly reduced systematic and random errors in precipitation estimates and led to substantial improvements in streamflow simulations when forcing a distributed hydrological model.

Objective

• To investigate the use of a nonparametric, tree-based model (Quantile Regression Forests, QRF) for combining multiple global precipitation datasets and characterizing the uncertainty of the resulting combined product.
• To evaluate the relative improvements and overall impact of the combined precipitation product on hydrological response by forcing a distributed hydrological model (SURFEX/RAPID) over the Ebro River basin.

Study Configuration

• Spatial Scale: Iberian Peninsula (focus on the Ebro River basin for hydrological evaluation); precipitation products analyzed at $0.25^{\circ}$ resolution; reference data (SAFRAN) at $5\,\text{km}$ resolution.
• Temporal Scale: 11 years (2000–2010); input data at 3-hourly intervals; reference data at 1-hourly resolution (aggregated to match input resolution for QRF training).

Methodology and Data

• Models used:
  • Quantile Regression Forests (QRF): Nonparametric statistical technique for blending precipitation products and generating ensembles.
  • SURFEX (Surface Externalisée) Land Surface Model (ISBA scheme): Used for hydrological simulations.
  • RAPID (River Routing Scheme): Used for river routing within the Eau-dyssée framework.
• Data sources:
  • Reference Precipitation: SAFRAN meteorological analysis system (high-resolution, ground-reference precipitation dataset for Spain, $5\,\text{km}$ resolution).
  • Satellite Precipitation Inputs: CMORPH, PERSIANN, and 3B42 (V7) (gauge-adjusted, quasi-global, $0.25^{\circ}$ resolution, 3-hourly).
  • Atmospheric Reanalysis Inputs: WFDEI (WATCH Forcing Dataset ERA-Interim) precipitation and air temperature ($0.5^{\circ} \times 0.5^{\circ}$ resolution, 3-hourly).
  • Ancillary Inputs: ESA CCI (v02.0) satellite-derived near-surface daily soil moisture ($0.25^{\circ}$ resolution); SRTM terrain elevation dataset ($90\,\text{m}$ resolution, interpolated to $0.25^{\circ}$).
• Evaluation Metrics: Normalized Centered Root Mean Square Error (NCRMSE, random error), Bias Ratio (BR, systematic error), Exceedance Probability (EP), Uncertainty Ratio (UR), Rank Histogram, and Nash–Sutcliffe Efficiency (NSE, hydrological performance).

Main Results

• Precipitation Improvement (QRF Ensemble Mean): The QRF-based combined product significantly reduced both random and systematic errors compared to individual input products.
  • Random Error (NCRMSE): Relative reduction ranged from $53\,\%$ to $99\,\%$. Reduction was greater during the cold season, particularly in low-elevation regions ($75\,\%$ to $99\,\%$).
  • Systematic Error (BR): Relative reduction ranged from $17\,\%$ to $76\,\%$. The combined product exhibited BR values closer to 1, especially for moderate to high rain rates.
  • Uncertainty Characterization: The QRF ensemble successfully encapsulated the reference precipitation dynamics, with acceptable Exceedance Probability (EP) values ($<0.26$ for rain rates below the 95th percentile).
  • Elevation Dependency: The QRF model noticeably reduced systematic error at high elevations, suggesting its utility in complex terrain.
• Hydrological Simulation Improvement (Streamflow): Forcing the hydrological model with the QRF ensemble mean resulted in substantial improvements in streamflow simulations relative to individual forcing datasets.
  • Random Error Reduction (NCRMSE): Relative reduction in streamflow NCRMSE ranged from $44\,\%$ to $78\,\%$ (above the 50th percentile) and $56\,\%$ to $88\,\%$ (low streamflow).
  • Systematic Error Reduction (BR): Relative reduction in streamflow BR ranged from $20\,\%$ to $99\,\%$.
  • Scale Dependency: Larger basins (Ebro at Tortosa and Ebro at Zaragoza) exhibited considerably lower systematic and random error values in streamflow simulations compared to smaller basins, demonstrating the smoothing effect of larger scales.

Contributions

• Development and application of Quantile Regression Forests (QRF), a nonparametric statistical technique, for optimally blending multiple global precipitation and ancillary datasets (soil moisture, temperature, elevation).
• Provision of a consistent formalism for characterizing precipitation uncertainty through the generation of a stochastic ensemble product.
• Comprehensive evaluation of the combined product's performance in both precipitation estimation (using high-resolution ground reference data) and hydrological modeling (streamflow simulation) over a complex region (Iberian Peninsula).
• Demonstration that incorporating dynamic (soil moisture, temperature) and static (elevation) land surface variables significantly improves the blending of satellite and reanalysis precipitation products.

Funding

• FP7 project eartH2Observe (grant agreement no. 603608)

Citation

@article{Bhuiyan2018nonparametric,
  author = {Bhuiyan, Md Abul Ehsan and Nikolopoulos, Efthymios I. and Anagnostou, Emmanouil N. and Quintana‐Seguí, Pere and Barella-Ortiz, Anaïs},
  title = {A nonparametric statistical technique for combining global precipitation datasets: development and hydrological evaluation over the Iberian Peninsula},
  journal = {Hydrology and earth system sciences},
  year = {2018},
  doi = {10.5194/hess-22-1371-2018},
  url = {https://doi.org/10.5194/hess-22-1371-2018}
}

Generated by BiblioAssistant using gemini-flash-latest (Google API)