|
Evaluation of GPM-IMERG Precipitation Estimate Errors Using Empirical Quantile Mapping for Runoff Management in the Caspian Coastal Region
|
Bromand Salahi *   , Ali Shahi |
| University of Mohaghegh Ardabili |
|
|
Abstract: (14 Views) |
EXTENDED ABSTRACT
Introduction: Precipitation serves as the fundamental driver of the hydrological cycle and acts as a critical input variable for rainfall-runoff modeling, flash flood simulation, drought monitoring, and strategic water resources management. Despite its importance, obtaining accurate and continuous long-term precipitation data remains a significant challenge. In the coastal region of the Caspian Sea, the complex interplay between the Alborz mountain range and the sea results in high spatiotemporal variability in precipitation. Consequently, existing ground-based rain gauge networks are often insufficient for precise hydrological modeling. In this context, satellite-based remote sensing products—particularly next-generation platforms such as GPM-IMERG, which offer superior spatiotemporal resolution—have emerged as a promising alternative. However, these products are inherently prone to systematic errors (bias) and uncertainties stemming from sensor limitations, topographic complexities, and retrieval algorithm deficiencies. Therefore, a rigorous evaluation and systematic bias correction are imperative before these data can be reliably utilized in hydrological studies. The primary objective of this research is to comprehensively evaluate the accuracy of GPM-IMERG precipitation estimates in the Caspian Sea coastal region and to investigate the efficacy of the Empirical Quantile Mapping (EQM) method in mitigating these biases, thereby enhancing the quality of hydro-meteorological inputs.
Methodology: To achieve a robust assessment, this study analyzed GPM-IMERG (Final Run) satellite data over a ten-year statistical period (2005–2014). This timeframe was selected to ensure continuous data availability and to facilitate meaningful comparisons with ground-truth observations. Satellite datasets were extracted via the Google Earth Engine (GEE) cloud computing platform and cross-referenced with selected meteorological station records. The performance assessment was conducted across daily, monthly, and annual temporal scales to elucidate the effects of temporal aggregation on data accuracy. Calculations were performed on a station-by-station basis to capture performance variances across different local climatic conditions (ranging from semi-humid to humid). To address systematic biases, the Empirical Quantile Mapping (EQM) method was employed. This approach identifies and corrects intensity-dependent biases by mapping the cumulative distribution function (CDF) of satellite estimates to that of observed data. Model performance was evaluated using a comprehensive suite of statistical metrics: the Pearson correlation coefficient (R) to assess linear correlation; Root Mean Square Error (RMSE) to quantify absolute error; Nash-Sutcliffe Efficiency (NSE) to measure simulation capability; and Percent Bias (PBIAS) to identify overestimation or underestimation tendencies. Furthermore, the capability of the satellite products to detect precipitation events was examined using the Probability of Detection (POD) and the False Alarm Ratio (FAR). Beyond these statistical indices, Taylor diagrams were utilized for the visual synthesis of correlation, standard deviation, and centered root-mean-square difference. Finally, spatial interpolation of both observed and satellite data was executed using the Inverse Distance Weighting (IDW) method within a GIS environment to delineate the spatial distribution patterns of precipitation across the coastal strip.
Results and Discussion: The results of the evaluation indicate that raw GPM-IMERG estimates consistently exhibit an “overestimation” tendency across all temporal scales, with this bias being particularly pronounced in daily (short-term) observations. Temporal analysis demonstrates that as the time scale increases from daily to monthly and annual, random errors decrease and data stability improves significantly. The application of the EQM method yielded a transformative improvement in data quality; statistical metrics such as NSE and RMSE showed substantial enhancement post-correction. This improvement was especially evident at the daily scale, particularly for moderate precipitation events. Regarding spatial analysis, the influence of topography on satellite data quality was evident. Stations located in relatively flat terrain, such as Babolsar and Gorgan, exhibited higher agreement with ground observations. In contrast, the Ramsar station—characterized by steep slopes and high-elevation topography—showed higher error rates and lower agreement, likely due to complex orographic precipitation phenomena that challenge current satellite retrieval algorithms. These findings confirm that topographic complexity remains a primary constraint for satellite-based precipitation estimation. Overall, the bias correction process successfully elevated the accuracy of satellite estimates to a level deemed reliable for operational applications in water resources management and flood analysis.
Conclusion: The present study demonstrates that while GPM-IMERG satellite data possess significant potential to bridge the gaps in sparsely monitored ground-based networks, their direct application in hydrological studies is associated with significant risks due to inherent systematic biases. The Empirical Quantile Mapping (EQM) method proved to be an efficient tool for structurally adjusting these errors, effectively bringing the data accuracy to a satisfactory level across all analyzed temporal scales. The findings suggest that corrected GPM-IMERG data, owing to their seamless spatial coverage, can serve as a reliable, complementary, or alternative source to ground-based stations for decision-making, drought monitoring, flood disaster management, and water resource planning in the Caspian Sea coastal zone. Future research is recommended to explore the integration of alternative correction frameworks, such as machine learning algorithms (e.g., Random Forest or Neural Networks), in conjunction with EQM to further optimize precipitation estimation accuracy. |
Article number: 7 |
| Keywords: Statistical Evaluation, Satellite Data, Remote Sensing, Northern Iran, Empirical Quantile Mapping |
|
|
|
Type of Study: Applicable |
Subject:
Special
Received: 2026/06/8 | Revised: 2026/07/24 | Accepted: 2026/07/30 | ePublished ahead of print: 2026/07/31 | ePublished: 2026/07/31
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
