Extended Abstract
Introduction
Accurate precipitation estimation, particularly in mountainous and semi-arid regions, is a fundamental prerequisite for water resources management, drought monitoring, flood forecasting, and hydrological modeling. Although satellite precipitation products provide continuous spatial coverage in regions with limited ground-based observations, their accuracy can be affected by precipitation regime, seasonality, rainfall intensity, topography, and spatial heterogeneity. The Run Final 07V IMERG-GPM product, as one of the latest versions of the IMERG precipitation product, has undergone improvements compared with previous versions; however, its performance is not necessarily uniform across complex climatic and topographic environments. Therefore, this study aimed to evaluate the performance of the Run Final 07V IMERG-GPM product and develop a regional, conditional, and precipitation-intensity-dependent framework for bias correction over northwestern Iran.
Materials & Methods
The study area covered northwestern Iran, characterized by considerable climatic and topographic heterogeneity and a broad elevation range. Daily precipitation observations from 188 rain-gauge stations were compared with the Run Final 07V IMERG-GPM product over the 2021–2023 period. Following quality control and exclusion of stations with more than 10% missing observations, satellite and ground-based precipitation data were temporally harmonized, and the nearest IMERG pixel was extracted for each rain-gauge station.
To investigate the spatial structure of bias, longitude, latitude, and elevation were considered as explanatory variables. Analysis of variance (ANOVA) was employed to identify influential variables and determine the appropriate model structure. Considering the dynamic and context-dependent nature of IMERG errors, precipitation observations were classified into six intensity categories, including light, low-to-moderate, moderate, heavy, extreme, and torrential precipitation. In addition, the stations were divided into four elevation classes to investigate the dependence of product performance on topographic conditions.
Based on the observed error structure, additive, multiplicative, and combined bias-correction relationships were developed separately according to month, precipitation intensity, and spatial characteristics. Model coefficients were estimated using a Monte Carlo cross-validation framework, and the most appropriate model was selected based on the minimum RMSE. Model performance before and after bias correction was evaluated using continuous statistical metrics, including the correlation coefficient (CC), root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), relative bias (rBias), Kling–Gupta efficiency (KGE), and Willmott-type agreement indices, as well as categorical precipitation-event indices including false alarm ratio (FAR), probability of detection (POD), critical success index (CSI), and Heidke skill score (HSS).
Results & Discussion
The results demonstrated that the bias structure of Run Final 07V IMERG-GPM over northwestern Iran was neither constant nor spatially homogeneous. Instead, the magnitude and structure of the bias varied substantially with month, precipitation intensity, and topographic conditions. Accordingly, different additive, multiplicative, and combined correction relationships were developed for different temporal, intensity, and spatial conditions. The inclusion of elevation, longitude, and latitude in several correction equations further confirmed the spatial heterogeneity of IMERG errors. These findings indicate that applying a single correction factor across the entire study region cannot adequately represent the complex structure of satellite precipitation errors.
At the overall scale, the proposed bias-correction framework substantially improved the performance of the satellite product. Based on 205,860 paired observations, CC increased from 0.1659 before correction to 0.8532 after correction, while RMSE decreased from 3.7207 to 1.6288 mm day⁻¹. NSE improved from −0.5436 to 0.7042, KGE increased from 0.1076 to 0.7114, and WAI increased from 0.4961 to 0.8607. In terms of precipitation-event detection, FAR decreased from 0.7685 to 0.0000, whereas POD increased from 0.4331 to 0.6370, CSI from 0.1776 to 0.6370, and HSS from 0.1785 to 0.7564. These changes indicate that the proposed correction framework not only reduced the magnitude of precipitation errors but also substantially improved the temporal agreement and detection capability of the satellite product.
The precipitation-intensity analysis revealed that the magnitude of improvement was not uniform across all precipitation classes. The most favorable performance was generally obtained for moderate to heavy precipitation events. For moderate precipitation, for example, rBias approached zero after correction, while POD and CSI reached 1.000. For heavy precipitation, RMSE decreased from 23.1140 to 8.5640 mm day⁻¹, and POD reached 0.9943. In contrast, although RMSE was substantially reduced for light precipitation, the corrected product showed a considerable reduction in relative bias and limited event-detection capability. For extreme precipitation, improvements in RMSE and relative bias were observed; however, KGE remained negative, indicating the difficulty of completely reproducing the statistical structure of very intense precipitation events. The torrential precipitation class contained only four events, resulting in considerable statistical uncertainty; therefore, these results cannot be interpreted as conclusive evidence regarding the ability of the model to reproduce rare extreme events.
The elevation-based analysis further confirmed the dependence of IMERG performance on topographic conditions. In the raw product, RMSE increased along the elevation gradient, indicating greater estimation errors in higher-altitude areas. Following bias correction, CC values across the elevation classes increased to approximately 0.84–0.86, while NSE became positive in all four classes. For example, after correction, CC reached 0.8532, 0.8447, and 0.8570 for the plains, mid-elevation, and high-elevation classes, respectively, while KGE reached 0.6639, 0.6988, and 0.7250. Despite these improvements, the highest RMSE remained in the very-high-elevation areas, indicating that part of the topography-related error persisted after correction.
The combined temporal, intensity-based, and elevation analyses indicate that the remaining IMERG errors cannot be attributed to a single source. They are likely associated with limitations in satellite precipitation retrieval algorithms, spatial and temporal variability of precipitation, complex topography, and the scale mismatch between gridded satellite estimates and point-based rain-gauge observations. This scale mismatch can become particularly important in mountainous regions, where precipitation may vary considerably over short distances.
Conclusion
The findings demonstrate that the bias of the Run Final 07V IMERG-GPM product over northwestern Iran is dynamic, heterogeneous, and strongly dependent on temporal conditions, precipitation intensity, and topographic characteristics. Therefore, a regional and conditional bias-correction approach is more appropriate than applying a single correction relationship or factor across the entire study area. The proposed framework substantially improved the overall performance of IMERG in terms of both continuous statistical and categorical event-based metrics, with particularly favorable results for moderate-to-heavy precipitation events and across the elevation gradient.
Nevertheless, the persistence of underestimation under certain conditions, the limited ability to detect very light precipitation, and the high uncertainty associated with rare torrential events indicate that future bias-correction frameworks should not focus exclusively on minimizing RMSE or maximizing CC. Preservation of precipitation volume and distribution, reduction of relative bias, and improved detection of rare and weak precipitation events should also be incorporated into model development and optimization.
Overall, the corrected Run Final 07V IMERG-GPM dataset can provide a more reliable complementary source for hydroclimatological studies, water-resources assessment, drought monitoring, extreme-precipitation analysis, and hydrological modeling in northwestern Iran. However, independent validation and longer observation periods are required to assess the generalizability and robustness of the proposed framework.