**1. Introduction**

In recent decades, observed climate trends have shown an increase in temperature worldwide so that extreme precipitation has increased in some specific areas (e.g., eastern half of North America, Eastern Europe, Asia, and South America) [1–5]. Due to rising temperatures, particularly over the Arctic, the sea-ice retreat allows for increased transport of heat and momentum from the ocean up to the tropo- and stratosphere. In the upper atmosphere, these waves deposit the momentum transported, disturbing the stratospheric polar vortex, which can lead to a breakdown of this circulation with the potential to also significantly impact the troposphere in mid- to late-winter and early spring [6,7].

Iran's climate is generally semi-arid and is subject to frequent flooding, causing major damage to people and society. In spring 2019, major floods occurred almost concurrently in different parts of the country. The first flood event occurred in late winter to early spring 2019 in the northeastern provinces due to heavy precipitation over the March 17–22 period. In at least one station, over 280 mm of precipitation was recorded over the six-day period. The subsequent second and third flood events occurred in the March 24–26 and March 31–April 2 periods, respectively, where most of the precipitation fell in the southwest, causing widespread damage to the people and infrastructure while filling/causing an overflow of most reservoirs. The total economic cost of these floods is estimated to be \$3.5 billion U.S. dollars. Studies on the causes of the March–April 2019 severe floods are still ongoing, although exceptional precipitation and climate change attribution are on the minds of most experts.

Given the heavy negative impacts imposed by the 2019 flood events, the monitoring and forecasting of precipitation remain major challenges for hydrologists and reservoir managers. The availability of global ensemble forecast models in the THORPEX Interactive Grand Global Ensemble (TIGGE) database [8] as well as high-resolution satellite estimates creates new opportunities for flood monitoring/forecasting. Extensive research has been conducted on the application of satellite precipitation estimates for motoring purposes and numerical weather prediction (NWP) models for forecast objectives. In terms of the latter, using TIGGE forecasts for a flood alarm system in China [9], flood early warning with European Centre for Medium-Range Weather Forecasts (ECMWF) model forecasts under Global Flood Awareness System (GLoFAS) in global scale, European Flood Awareness System (EFAS) for Europe projects [10], and African Flood Forecasting System (AFFS) [11] are some examples of NWP applications in flood forecasting.

Numerous studies have been conducted to evaluate the estimated precipitation from NWP models and satellite-based precipitation estimates (SPEs). For example, the results of TIGGE precipitation forecasting in flood-prone areas of China showed that the ensemble forecast model is more proficient than the single forecast models [12]. He et al. [13] showed that TIGGE ensemble forecasts are suitable for forecasting flood events. The rapid alert system was developed from four operational NWP models: UKMO (United Kingdom Met Office), NCEP (National Centers for Environmental Forecast), ECMWF, and JMA (Japan Meteorological Agency). The probability of severe weather events was forecasted based on the climatological probability density function in each model. Numerous case studies have shown that these products successfully forecasted severe events such as the Russian heat wave in 2010, the Pakistan flood in 2010, and Hurricane Sandy in 2012 [14]. In West Africa, evaluating the forecasts of seven TIGGE meteorological databases against the Tropical Precipitation Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) product assumed as the observed precipitation over the 2008 to 2012 period showed that ECMWF and UKMO performed better than the other models [15]. In Iran, the forecasts of ECMWF, UKMO, and NCEP centers for thirteen synoptic stations in eight different precipitation regions over a 1–3 day lead time over the 2008 to 2016 period showed that ECMWF in most regions, UKMO in mountainous areas, and NCEP along the Persian Gulf coast performed the best, while, as expected, the model skill decreased with increasing lead time [16].

The accuracy of SPEs is influenced by their spatiotemporal resolution, which in turn impacts the prediction of natural hazards. Therefore, the assessment of new precipitation products is often recommended before the product can be employed in research and decision-making. Several studies have been conducted to evaluate the SPEs over Iran. Moazzami et al. [17] and Javanmard et al. [18] examined different SPEs products at a daily time-scale over diverse climate conditions in Iran. Overall, the results showed that 3B42V7 outperformed other SPEs. Sharifi et al. [19] evaluated the first version of the integrated multi-satellite retrievals for global precipitation measurement (IMERG) version-03 in comparison with the TMPA and ERA-Interim products across different parts of Iran and found that IMERG generally outperformed the other products. They later improved the spatial resolution and accuracy of the SPEs through downscaling and bias-correction techniques [20,21]. In another study, Beck et al. [22] evaluated 26 precipitation datasets and compared them with gauge-radar data over the United States. Among the gauge-corrected products, the best overall performance was obtained

by Multi-Source Weighted-Ensemble Precipitation (MSWEP)-V2.2, followed by IMERGDF-V05 and MERRA-2. However, IMERG real-time V05 performed substantially better than TMPA-3B42-real-time V7 and ERA5-HRES, particularly, in regions dominated by convective storms. In another study, Sharifi et al. [23] examined the accuracy of six SPEs and gridded precipitation models against a dense network of 872 stations over Austria, in terms of extreme events and different altitude categories. They also found that the latest version of IMERG-V06A performed better than the other products (except MSWEP-2.2), which was consistent with the study results by Beck et al. [22]. With respect to extreme precipitation events, Fang et al. indicated that although IMERG well captured the spatial pattern of extreme precipitation over China, the topography and climate condition had a significant influence on its performance [24]. In another study, Sunikumar et al. demonstrated the ability of IMERG to follow the intraseasonal variability with minor differences observed in the maximum values of precipitation during the rainy season over Japan, Philippines, and Nepal [25]. Mazzoglio et al. improved an extreme precipitation detection system using IMERG data and stated that this product guarantees good results when the precipitation aggregation interval is equal to or greater than 12 h [26]. However, no comparison has been made so far between the ensemble forecasts with IMERG product for extreme flood events over Iran.

The purpose of this study was to evaluate the performance of ensemble precipitation forecasting models and IMERG products for three severe 2019 flood events in Iran to determine whether these products have potential in (major) flood warning applications over Iran. According to ground precipitation data, the highest precipitation occurred in Gorganrud, Karkheh, and Karun Basins. These three basins constitute the study area.
