1. Introduction
Satellite-based maps of quantitative precipitation estimation (QPE) are available globally in near–real time (Sorooshian et al. 2000; Smith et al. 2007; Skofronick‐Jackson et al. 2018). This presents an opportunity for streamflow, and thus flood, forecasting everywhere, even in regions without weather radars and with only sparse rain gauge networks. Early attempts demonstrate the viability of this approach from the computational point of view (e.g., Wu et al. 2012, 2014), however, a comprehensive evaluation of the skill of a satellite-based streamflow forecasting system is currently lacking.
This paper makes a partial contribution in filling this gap. First, a meaningful evaluation requires the existence of an alternative system serving as reference. Such systems exist in developed regions that do not need to rely on satellite products. An example is the central United States, which is well covered by many weather radars and rain gauges, but which also have some limitations (e.g., Villarini and Krajewski 2010; Ehsani et al. 2021; Ehsani and Behrangi 2022), and that has an extensive network of stream gauges. In particular Iowa, located in the upper Midwest between the Mississippi and Missouri Rivers, is well suited for the purpose. Its view from space by satellite sensors is free of the complications caused by coastlines, mountains, and variable land use and cover. Therefore, satellite rainfall retrieval algorithms are expected to perform best over the cultivated plains and low relief terrain of the state.
Depending on the type of satellite sensors used to detect atmospheric properties that are translated into precipitation rate, satellite-based precipitation estimation is primarily divided into two approaches (visible/infrared and microwave sensing). These approaches have their own strengths and weaknesses: 1) geostationary satellites with visible/infrared measurements offer higher temporal resolution data than polar orbiting satellites carrying microwave sensors (e.g., 30 min versus 3 h) and 2) observations using active microwave (e.g., reflectivity) have more reliable relation with precipitation rate than those using infrared channels (e.g., brightness temperature). More detailed aspects of the two approaches are documented in Scofield and Kuligowski (2003) and Levizzani and Cattani (2019). For an estimation of cold precipitation (e.g., snow and ice), passive microwave sensors are often used (e.g., Adhikari et al. 2020). However, the retrieval of cold precipitation is challenging with large uncertainties mainly due to the intricate radiative properties associated with nonspherical shape of snowflakes and ice crystals (e.g., Levizzani et al. 2011; for additional uncertainty sources, refer to Skofronick‐Jackson and Johnson 2011; Liu and Seo 2013; You et al. 2017). By taking advantage from both type products and data from the active dual-frequency precipitation radar (Hamada and Takayabu 2016), the National Aeronautics and Space Administration (NASA) has developed a new satellite-based product called the Integrated Multi-satellitE Retrievals for GPM (IMERG). Recently, NASA has done IMERG retrospective processing to cover the Tropical Rainfall Measuring Mission (TRMM) era (Huffman et al. 2020).
Numerous studies have assessed the state-of-the-art IMERG precipitation estimates using reference data from ground radars or a cluster of rain gauges (e.g., Tan et al. 2016; O et al. 2017; Li et al. 2020; Moazami and Najafi 2021). These assessment efforts have focused mostly on 1) the evaluation of IMERG’s improvement against the TRMM products (e.g., Chen and Li 2016; Liu 2016; Khodadoust Siuki et al. 2017) and 2) the intercomparison among three different IMERG (Early, Late, and Final Run) products (e.g., O et al. 2017; Li et al. 2021). While numerous studies have examined satellite precipitation products (e.g., TRMM) for hydrologic modeling (e.g., Nikolopoulos et al. 2010; Maggioni et al. 2013; Habib et al. 2014; Sperna Weiland et al. 2015), the hydrologic utility of the IMERG products has been barely explored with respect to basin scales associated with IMERG’s spatial resolution. This study contributes and adds values to the effort on the hydrologic evaluation of IMERG by employing more comprehensive validation datasets (see section 2) than those used in previous studies (e.g., Tang et al. 2016; Wang et al. 2017; Jiang and Bauer-Gottwein 2019; Stephens et al. 2022). In this study, we evaluate IMERG-Early and IMERG-Final Run products over Iowa in the United States and examine the streamflow prediction skill driven by the two IMERG forcing products. The main focus of this study is to assess the utility of the IMERG-Early Run (e.g., near-real-time) product for flood prediction, compared to that of the Final Run product, regarding a wide range of hydrologic (e.g., basin) scales.
2. Study domain and data
To comprehensively assess the hydrologic utility of IMERG products, a study domain for which a well-established model and modeling data resources (e.g., precipitation, streamflow, and model components and parameters) are readily accessible is mandatory. The Iowa domain shown in Fig. 1 meets these requirements and is recognized through the NASA field experiments known as Iowa Flood Studies and Soil Moisture Active Passive (e.g., Quintero et al. 2016; Seo et al. 2018a; Walker et al. 2019). The Iowa Flood Center (IFC) has fulfilled operational weather monitoring and streamflow forecasting over this domain for over 10 years and has devoted their effort to extensive flood research and outreach activities (Krajewski et al. 2017). In this section, we briefly describe the IMERG products, as well as point- and grid-based reference precipitation data and streamflow observations used to evaluate the IMERG precipitation estimates.
(a) The study area and USGS streamflow gauging stations in Iowa and (b) NWS COOP rain gauge locations. The solid mesh in (b) indicates the IMERG grids over Iowa.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
a. Precipitation dataset
IMERG is the unified algorithm that combines multisatellite observations from the GPM satellite constellation [for details regarding satellite sensors and products included in the algorithm, refer to Huffman et al. (2020)]. Precipitation estimates for IMERG are obtained through intercalibration, merging, and interpolation of precipitation estimates from satellite microwave estimates together with microwave-calibrated infrared estimates and rain gauge analysis. IMERG provides three precipitation products depending on the system running time and product latency: Early (IMERG-E), Late (IMERG-L), and Final (IMERG-F) runs. The latency of IMERG-E and IMERG-L is up to 4 and 14 h from observation time, respectively. The primary difference between these two products is that IMERG-E allows forward propagation (temporal extrapolation) only, while IMERG-L applies both forward and backward (interpolation) propagation. There have been some studies to mitigate IMERG’s latency and improve its utility for flood warning systems (e.g., Ehsani et al. 2022). IMERG-F includes a gauge adjustment once the monthly gauge analysis is available, and thus its latency is up to 3.5 months from the observation month. Spatial and temporal resolutions of IMERG are 0.1° (approximately 10 km) and 30 min. We collected IMERG-E and IMERG-F Version 06B products from the NASA Precipitation Processing System data servers (https://arthurhou.pps.eosdis.nasa.gov/) for a period from 2016 to 2020 and retrieved hourly precipitation information for the study domain shown in Fig. 1. We excluded IMERG-L in our analysis because its performance demonstrated in many prior evaluations (e.g., O et al. 2017; Wang et al. 2017) showed little difference from that of IMERG-E, if any, and our focus is to evaluate the utility of the near-real-time product.
We used grid- and point-based reference data to assess the IMERG precipitation estimates: 1) Multi-Radar Multi-Sensor (MRMS) precipitation estimates (Zhang et al. 2016) and 2) rain gauge observations from the National Weather Service (NWS) Cooperative network (COOP; Lawrimore et al. 2020). MRMS is radar-based composite precipitation estimates that cover the entire United States. The MRMS system incorporates radar data with satellite, lightning, and surface and upper air observations, and numerical weather prediction model analyses. While MRMS provides a suite of QPE products with 0.01° (approximately 1 km) and several temporal resolutions, we collected the hourly QPE product that includes a bias correction using data from rain gauges. For a fair comparison between MRMS and IMERG, we spatially resampled the MRMS product by taking an average of corresponding MRMS values within an IMERG grid. As an example, Fig. 2 shows 3-day rain totals of the resampled MRMS, IMERG-E, and IMERG-F products for a flooding event that occurred in September 2016 (see, e.g., Seo et al. 2018b). Given the study period (2016–20), the MRMS estimates were derived from multiple relationships between radar reflectivity and rain rate. The new algorithm update (completed in October 2020; refer to https://inside.nssl.noaa.gov/mrms/past-code-updates/) to “synthetic QPE,” including the specific attenuation method (Zhang et al. 2020), has not been applied retrospectively. The real-time MRMS product is available from the National Centers for Environmental Prediction (https://mrms.ncep.noaa.gov/data/).
Example maps of rain totals on 21–23 Sep 2016: (a) MRMS resampled at IMERG’s spatial resolution (0.1°), (b) IMERG-E, and (c) IMERG-F.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
We collected rain gauge records from the COOP Hourly Precipitation Data (HPD) Network, Version 2.0 available from the NOAA National Centers for Environmental Information (https://www.ncei.noaa.gov/data/coop-hourly-precipitation/v2/). These point records were not directly used to evaluate the IMERG products due to the significant difference (e.g., representativeness error) between their spatial scales (i.e., point versus 10 km). As shown in Fig. 1b, it is rare that multiple COOP gauges are located within an IMERG grid in the study domain. Therefore, we used these data to apply an indirect assessment approach known as the triple collocation (TC) method (e.g., Stoffelen 1998; Alemohammad et al. 2015; Bai et al. 2021). TC can characterize the uncertainty of a variable (i.e., precipitation) when three collocated quantities measured from different instruments or platforms are available without direct information on the truth. Gruber et al. (2016) discussed TC’s representativeness assumption and documented that TC allows to estimate the error variance of grid datasets independent from the representativeness of the point-scale data used in the analysis.
b. Streamflow data
To evaluate the performance of hydrologic simulations driven by precipitation products, we collected quality-assured streamflow measurement data with a 15-min resolution at 140 United States Geological Survey (USGS) gauging stations in Iowa for the study period. In our data collection, we excluded stations located downstream of reservoirs to avoid the effect of regulated flow on our evaluation. The streamflow discharge is obtained from river stage readings by using well-defined rating curves developed through USGS’s periodic stage-discharge measurements. In this study, we do not account for the effect of uncertainties in rating curves on our evaluation because the degree of uncertainty is unknown and site specific. The drainage areas monitored by the USGS stations in Iowa range from 10 to 30 000 km2. The UGGS streamflow data are available from the National Water Information System (https://waterdata.usgs.gov).
3. Methodology
To assess the hydrologic utility of the IMERG products, we perform a direct comparison with the reference precipitation dataset and an indirect comparison using streamflow generated by a hydrologic model driven with multiple precipitation products. Because the hydrologic model used in this study is an important factor for the latter assessment, we briefly introduce its structure in this section. We also provide statistical descriptions and formulas of the performance metrics used in the precipitation and streamflow evaluation analyses.
a. Hillslope Link Model
The IFC has developed and operated a fully automatic real-time streamflow forecasting system based on the Hillslope Link Model (HLM) (Krajewski et al. 2017). The HLM is a continuous distributed hydrologic model that simulates the processes of runoff generation and routing based on hillslopes and channel links as the hydrologic response units for which these processes are modeled (e.g., Quintero et al. 2020a; Mantilla et al. 2022). The HLM simulation includes processes of infiltration of water ponded in the surface to the top layer of soil and percolation to deeper soil. For evapotranspiration (ET), the HLM employs monthly averaged estimates from the past 10-yr Moderate Resolution Imaging Spectroradiometer (MODIS) actual ET (e.g., Mu et al. 2011). Overland- and baseflow processes transport water from the hillslopes to the river channels. The HLM configuration used in this study does not include snowmelt processes. A hydrologic routing module estimates discharge velocity in the channel based on the scaling properties of the river network (e.g., Ghimire et al. 2018). The changes in water storage across the layers of soil are described in terms of ordinary differential equations organized following the river network topology. An asynchronous solver takes advantage of the river network structure to solve the differential equations in parallel, using algorithms suitable for high performance computing (Small et al. 2013). The hillslopes and channels were obtained from a 90-m digital elevation model; the average area of the hillslopes is 0.1 km2, and the average channel length is 600 m (Quintero and Krajewski 2018). This allows streamflow predictions at a wide range of spatial scales, ranging from tens to thousands of square kilometers. Performance of the HML has been documented in several studies (Cunha et al. 2012; ElSaadani et al. 2018; Quintero et al. 2020a; Seo et al. 2021).
In this study, we used the MRMS, IMERG-E, and IMERG-F products as the HLM forcing data and performed 1-month simulation with MRMS to spin up the model states and generate initial conditions. The HLM parameters are determined a priori; the model does not “favor” any particular input product. We avoided adjusting HLM’s parameters because parameter calibration may conceal uncertainties in streamflow prediction propagated from different precipitation forcing products (i.e., MRMS versus IMERG).
b. Precipitation analysis
We compared IMERG-E and IMERG-F products with resampled MRMS at hourly and daily scales for the study period. These hourly and daily comparisons included warm season data only because radar-based QPE (MRMS) for solid and mixed precipitation contains large uncertainties (e.g., Seo et al. 2015; Souverijns et al. 2017). Moreover, satellite-based estimates often lead to significant errors in cold seasons as we discussed in the Introduction section. Another reason to exclude cold months in the analysis is that our main focus is on evaluating the prediction capability of floods that mostly occur in the warm seasons.
c. Streamflow analysis
Contingency table used to estimate hit and false alarm rates.
4. Results
a. Precipitation evaluation
Figures 3 and 4 show hourly and daily scale comparison results represented by the two-dimensional histogram and conditional distribution. In the hourly comparison in Fig. 3, both IMERG products do not show good agreement with MRMS: the matched pairs between IMERG and MRMS are highly scattered and are not clustered along the one-to-one line. We observe that the bias correction of IMERG-F improved the overestimation tendency of IMERG-E (bias: 1.45). The box-and-whisker plots presented in Fig. 3 reveal the conditional features of reference precipitation (MRMS) for given IMERG precipitation ranges. While the conditional distributions for both IMERG products are positively skewed with a long tail, particularly at lower ranges, IMERG-E shows higher variability with wider interquartile ranges and longer tails. We observe that IMERG-F agrees slightly better with MRMS, and the bias correction included in IMERG-F seems to reduce variability and a range of outlier at each given IMERG interval although the conditional bias remains almost the same (or worse). Compared to the result at the hourly scale, the comparison shown in Fig. 4 demonstrates significant improvement at daily scale for IMERG products: discrepancy between IMERG and MRMS was largely reduced, and the correlation coefficients of both products were improved. The conditional distributions at daily scale present similar features to the ones at hourly scale (e.g., higher variability and a longer tail with IMERG-E).
Hourly scale comparison of MRMS vs IMERG using two-dimensional histogram and the box-and-whisker plots that represent conditional distributions regarding IMERG precipitation ranges: (a) IMERG-E and (b) IMERG-F. The B and r indicate bias and correlation coefficient defined in Eqs. (1) and (2), respectively. The MRMS data were resampled at the IMERG spatial scale (i.e., 10 km).
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
As in Fig. 3, but on a daily scale.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
In Fig. 5, we present the statistical evaluation result using four performance metrics defined in Eqs. (1)–(4). These metrics were calculated based on hourly IMERG–MRMS pairs and were integrated to the monthly scale. Unlike Figs. 3 and 4, we included results from cold months in Fig. 5 to disclose IMERG’s performance change over different seasons. Vertically shaded areas shown in Fig. 5 indicate warm months (i.e., April–October) in the study period. The monthly bias analysis presented in Fig. 5 illustrates that IMERG-E overestimates significantly against MRMS, whereas the bias of IMERG-F stays around the unity line, implying that IMERG-F agrees well with MRMS. Based on this result, we recognize that IMERG’s monthly gauge adjustment/correction seems effective. Higher bias corresponds to the larger errors, MAE and RMSE, and their temporal changing pattern is quite similar to that of the bias. Unlike other metrics in Fig. 5, there is no meaningful difference in correlation between IMERG-E and IMERG-F, which indicates, not surprisingly, that the bias correction does not much improve correlation (linear dependence) of IMERG-E. Overall, we observe that major differences of all evaluation metrics between the two IMERG products occur mostly in warm seasons, which may result in significant differences in streamflow generation.
The four statistical performance metrics of bias, correlation, MAE (mm), and RMSE (mm) from the IMERG precipitation evaluation on a monthly scale. These monthly metrics were calculated using hourly IMERG–MRMS pairs within corresponding months. The shaded areas indicate warm months (April–October).
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
Figure 6 shows the result from the TC method. In the analysis, we used a set of hourly IMERG-E (or IMERG-F), MRMS, and COOP data and estimated unbiased RMSE and correlation coefficient at monthly scale. The lines representing MRMS and COOP are fairly close between IMERG-E and IMERG-F panels in Fig. 6, but they are not exactly the same. The observed patterns of RMSE and correlation coefficient between IMERG products look similar to those in Fig. 5: 1) IMERG-E contains larger RMSE, and the error increases during warm seasons for both products; and 2) the values of correlation coefficient are comparable with those in Fig. 5, and the difference between IMERG-E and IMERG-F looks insignificant. Note that the RMSE and correlation of gauge observations are not superior to those of MRMS. This is because of the gauge representativeness error (e.g., Kitchen and Blackall 1992; Ciach and Krajewski 1999; Seo and Krajewski 2011) at IMERG product scale, and we used data from a single rain gauge to represent 10 × 10 km2 IMERG grids. The scale-dependent feature (i.e., magnitude) of this error is demonstrated in Seo and Krajewski (2010) for the same area but using different rain gauge data.
The result (unbiased RMSE and correlation coefficient) from the triple collocation method using MRMS, IMERG, and COOP rain gauge data. The shaded areas indicate warm months (April–October).
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
b. Streamflow evaluation
We examined the performance of HLM streamflow simulations at hourly scale driven by the three different precipitation products (MRMS, IMERG-E, and IMERG-F). Figure 7 shows overall predictive performance with the three forcing products, and their performance represented by the KGE are organized with respect to drainage area covered by the USGS stations. Because the KGE values were calculated from a large sample (i.e., 5-yr simulations), the issue of sampling bias and skewness associated with the KGE (e.g., Lamontagne et al. 2020; Clark et al. 2021) is less important. While the performance of IMERG-E over the entire drainage scale stays very low (e.g., KGE values are smaller than zero), that of MRMS and IMERG-F seems quite comparable. We observe from Fig. 7 that the streamflow predictability improves, particularly for MRMS and IMERG-F, as drainage scale increases because random errors in the forcing products tend to be averaged out at larger scale. On the other hand, an opposite tendency over drainage scale detected from IMERG-E may imply that bias in the forcing data exposed in Fig. 5 plays a key role as basin scale becomes larger.
Overall performance (represented by KGE) of hydrologic simulations at hourly scale driven by the three precipitation forcing products regarding drainage scale. The circles indicate the performance evaluation results (simulated vs observed) for the 5-yr period at 140 USGS stations.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
To further compare the prediction performance at different scales, we split the 140 USGS stations into four groups according to their coverage area and aggregated the performance analysis results (KGE, bias in peak flow, and total runoff volume) for each group in Fig. 8. In Fig. 8a, the box-and-whisker plots at scales greater than 100 km2 show that the central values (i.e., median) of KGE distributions with MRMS are greater than those with IMERG-F. At the smallest sale (<100 km2), MRMS shows superior performance to IMERG-F in terms of higher median and lower variability. This implies that the spatial resolution of IMERG (i.e., 10 km) is unsuitable to describe hydrologic behaviors at that scale. Literature recommends that KGE greater than 0.3, shaded areas in Fig. 8a, is considered acceptable for hydrologic simulations (e.g., Knoben et al. 2019). Based on our results, IMERG-F appears to be an acceptable forcing product to simulate the hydrologic behavior of basins in Iowa larger than about 100 km2. Because we present the analyses of bias in peak flow and total runoff volume in Figs. 9 and 10, respectively, the results of Figs. 8b and 8c will be discussed together with the corresponding figures.
Distributions of KGE, bias in peak flow, and bias in runoff volumes obtained from the three precipitation forcing products. The distributions are represented regarding drainage scale. The shaded areas in gray denote acceptable ranges for each performance metric.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
Performance of peak flow estimation: (a) comparison of observed and simulated annual peak flow, (b) bias in annual peak flow regarding drainage scale, and (c) monthly bias in peak flow.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
As in Fig. 9, but for bias in total runoff volumes.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
In the peak flow analysis, we excluded cases where observed and simulated annual peaks are separated more than 48 h to avoid a situation in which the HLM generated an annual peak from a completely different event. Figure 9 compares observed and simulated annual peaks at the 140 USGS stations and demonstrates a comparable skill in estimating peak flow for MRMS and IMERG-F, but somewhat better performance with MRMS. The percentages of the number of stations included in the analysis shown in Fig. 9 are 35%, 12%, and 30% for MRMS, IMERG-E, and IMERG-F, respectively. IMERG-E yields significant overestimations at most USGS stations because of the observed bias in precipitation forcing. The dependence of simulated annual peaks on drainage scale is illustrated in Fig. 9b: there is no systematic over- or underestimation pattern regarding drainage scale. In Fig. 8b, we present the distribution of bias in annual peaks for the four groups of drainage scale. The shaded area in Fig. 8b shows errors smaller than 20% in the peak estimation as an acceptable error range. While IMERG-E reveals considerable overestimations at all scales, MRMS and IMERG-F show underestimations at scales smaller than 1000 km2. At scales larger than 1000 km2, the central values of peak bias for both MRMS and IMERG-F stays around the unity (those for MRMS seem closer to the unity with higher variability). This result might be subject to the skill of HLM to estimate peak flow at smaller scales with no calibration of HLM parameters. Most IMERG-F’s bias distributions fall within the 20% interval for basins larger than 100 km2, demonstrating skill to simulate annual peak flows at these basin scales in Iowa.
We also assessed monthly peak flow generated by the three forcing products and present a time series of monthly bias averaged over the 140 USGS stations in Fig. 9c. In this analysis, we included simulation results from cold months. The most notorious aspect is the substantial overestimations of IMERG-E for 2016–18, and this overestimation tendency decreases in 2019–20, which may indicate improvement in the IMERG-E algorithm from 2019 (refer to Huffman et al. 2020). The monthly biases of MRMS and IMERG-F seem to be in the same order of magnitude over the 5-yr period although there are some oscillations for each product at different periods. We note that the HLM’s peak flow estimation has a limitation for snowmelt events in early spring.
In Fig. 10, we analyzed annual runoff volumes generated by the three precipitation forcing products and HLM simulations. The observed patterns of annual runoff volumes from Fig. 10 are similar to the ones in Fig. 9: 1) IMERG-E leads to overestimations in estimating total water volumes and 2) the performance of MRMS and IMERG-F looks comparable, and both products yield slight underestimations at all scales as shown in Fig. 8c. Assuming an acceptable error of 20% in volume bias, shaded area shown in Fig. 8c, we observe that most IMERG-F’s bias distributions fall within this range. The HLM’s underestimation tendency in estimating runoff volumes are discussed in Quintero et al. (2020a). The monthly volume bias averaged over the USGS stations is presented in Fig. 10c and shows a similar overestimation pattern for MRMS and IMERG-F to the one in Fig. 9c during the second half of 2017.
Figures 11 and 12 illustrate the maps and distributions of hit, false alarm, and critical success index obtained from the 5-yr HLM simulations driven by the three forcing products. In Fig. 11, IMERG-E shows relatively high hit rates because the overestimations of peak flow demonstrated in Fig. 9 contributed to capturing the event peaks that occurred for the analysis period. These overestimations also led to the higher number of false alarms at the same time. In comparison (see Fig. 12), an average hit rate is about 20% for MRMS and 10% for IMERG-F, whereas an average false alarm rate for both is in a similar range (i.e., about 75%). The critical success index is also slightly higher for MRMS (about 15%) compared to IMERG-F (about 10%). As shown in Fig. 12, the interquartile ranges of MRMS are wider than those of IMERG-F: 1) hit rates of 0%–40% for MRMS versus 0%–20% for IMERG-F; 2) false alarm rates of 35%–100% for MRMS versus 45%–100% for IMERG-F; and 3) critical success index of 0%–25% for MRMS versus 0%–18% for IMERG-F. This indicates that MRMS forcing led to the greater number of stations that show a better hit, false alarm, and critical success index. However, the low hit and high false alarm rates (even for MRMS) were caused by the snowmelt process during early spring, which is not accounted for in the current HLM configuration, and errors in precipitation forcing, particularly for small-scale basins. We also visually inspected the geographic and scale dependence of all metrics and found no clear dependence from Figs. 11 and 12.
Spatial distribution of hits, false alarms, and CSI. “No data” indicates that no metric information is available because there were no observed or simulated flood events at those specific locations as described in section 3c.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
Hit rate, false alarm rate, and critical success index represented by (top) drainage scale and (bottom) their distribution.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
To demonstrate the different propagation effect of the three precipitation products through the river networks on the streamflow generation, we present two example cases (all of June and 22–29 September 2016) in Figs. 13 and 14 for the Cedar River basin in Iowa. While the accumulated precipitation amounts over time and space as represented by the river networks are illustrated on the left panel, the right panel shows the observed and simulated hydrographs at the five locations identified in the accumulated precipitation maps. The color on a particular location of the river networks represents mean areal precipitation for the upstream catchment area of the location. The result shown in Fig. 13 is consistent with the overall tendency we found from this study: IMERG-E tends to overestimate precipitation and leads to significant overestimations in streamflow prediction regardless of drainage scale. Although the streamflow prediction results of MRMS and IMERG-F look comparable, we can recognize that there are differences in the spatial distribution of accumulated precipitation along the river networks, particularly for small-scale basins. On the other hand, the case included in Fig. 14 shows an opposite tendency to the one in Fig. 13. The accumulated precipitation amounts of IMERG-E are smaller at most river network locations than those of MRMS, and its spatial distribution is similar to IMERG-F. This results in underestimations in streamflow prediction at the five locations for both IMERG-E and IMERG-F. The IMERG’s underestimation of precipitation observed from Fig. 14 agrees with the precipitation pattern within the Iowa domain presented in Fig. 2. However, the IMERG-E’s overestimation pattern is observed at the northeastern area of the Iowa border in Fig. 2.
(left) Aggregated MRMS and IMERG precipitation over time (June 2016) and space (the Cedar River basin in Iowa) represented by the river networks and (right) observed/simulated hydrographs at the five different locations.
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
As in Fig. 13, but for a different period (22–29 Sep 2016).
Citation: Journal of Hydrometeorology 24, 6; 10.1175/JHM-D-22-0129.1
5. Conclusions and discussion
This study examined the IMERG-E and IMERG-F products by evaluating their precipitation estimates and streamflow generated from hydrologic simulations forced by these products using IFC’s operational hydrologic model, HLM, for the Iowa domain (see Fig. 1). The precipitation and streamflow evaluations for the 5-yr period (2016–20) employed MRMS gauge-corrected precipitation estimates and USGS streamflow observations as references. The primary objective of this study was to assess the hydrologic utility of the near-real-time product (i.e., IMERG-E) across basin scales, compared to that of IMERG-F that becomes available only after the monthly rain gauge records are collected.
Precipitation evaluation results revealed that IMERG-E does not agree well with MRMS, and the conditional distributions for IMERG-E shown in Figs. 3 and 4 show higher variability and longer tails than those for IMERG-F at hourly and daily scales. We detected IMERG-E’s significant overestimations and increasing errors during warm seasons, whereas IMERG-F seems almost unbiased (Fig. 5) at monthly scale. The analysis based on the TC method demonstrated a consistent result with the analysis at a monthly scale: IMERG-E contains significant errors that increase sharply during warm seasons, which have implications for streamflow prediction (e.g., large overestimations). We also discovered MRMS’s good performance (see correlation in Fig. 6), and this performance decreases during cold months because of winter precipitation we discussed in section 4a.
In the streamflow evaluation, IMERG-E tends to overestimate peak flow and total runoff volumes at all drainage scales, which results in lower skills in hydrologic prediction. The overall performance metric, KGE presented in Fig. 7, demonstrates that significant overestimations in precipitation forcing lead to almost no skill in streamflow prediction and no (basin) scale dependence that is observed from MRMS and IMERG-F simulations. On the other hand, IMERG-F shows comparable performance with MRMS despite the temporal scale difference between their gauge-correction schemes (monthly versus hourly). Overall, the streamflow simulations with both products show slight underestimations in estimating peak flow and runoff volumes at most scales (see Fig. 8), and their KGE values tend to increase as drainage scale becomes larger. However, the performance of MRMS looks superior to that of IMERG-F at a smaller scale (<100 km2), and one possible reason could be that IMERG-F cannot describe the spatial variability of precipitation at scales smaller than its spatial resolution.
The hit and false alarm analysis provides practical insight regarding the flood prediction capability of the three precipitation forcing products. This analysis exposed that IMERG-E yields high hit and false alarm rates at the same time because of its significant precipitation overestimations shown in Fig. 5. The low performance of hit and false alarm rates (even for MRMS) indicates the effect of snowmelt processes (not included in the HLM simulations in this study) during early spring, as well as errors in QPE forcing. Based on our operational experience over 10 years and frequent communication with NWS forecasters in the North Central River Forecast Center, errors in QPE (e.g., MRMS) often lead to significant errors in streamflow prediction, particularly at small-scale watersheds. For a comparison between MRMS and IMERG-F, MRMS forcing seems to yield slightly better hit and false alarm rates based on the distributions presented in Fig. 12. We found no geographic and scale dependence of these rates.
We note that our evaluation results might be limited to the study domain (Iowa) and other areas similar to Iowa’s geographic environment. Iowa is a relatively flat area with no orographic effects, and the estimation of precipitation and its effect on streamflow generation are relatively simple and straightforward, compared to those for areas with complex terrain that may lead to dissimilar results. Some studies for different geographic regions (e.g., Wang et al. 2017; Jiang and Bauer-Gottwein 2019) reported that IMERG-E showed similar (poor) performance to the results demonstrated in this study. While those studies discussed IMERG-E’s potential with the parameter calibration of hydrologic models, we think that model calibration masks the effect of errors in IMERG and obscures our understanding of prediction results derived from forcing errors. We conclude that IMERG-E requires a bias correction for near-real-time hydrologic prediction (e.g., Hartke et al. 2023). The systematic error with an under- or overestimation tendency plays a major role in streamflow generation, and random errors in precipitation forcing tend to average out, particularly at larger scales, if the errors are uncorrelated (e.g., Vivoni et al. 2007; Cunha et al. 2012). The product might be useful, even when corrupted with significant errors, at very large scales where predictive skill is more influenced by the water that is already stored in the river network (e.g., Palash et al. 2018).
One can model the conditional structure of the error (e.g., Wright et al. 2017; Seo and Krajewski 2021) at relevant temporal scale and apply it to correct the IMERG-E product. However, our preliminary modeling analysis at hourly and daily scales demonstrated that an average feature conditioned solely on the magnitude of IMERG estimates cannot describe variability that changes from event to event. This suggests a more rigorous way to characterize the conditional features using environmental variables (e.g., data from numerical weather prediction model analysis) that may affect satellite precipitation estimation. In the near future, we plan to incorporate a large dataset of environmental variables into the error modeling framework and test deterministic and data-driven approach to characterize the conditional structure.
Acknowledgments.
This study was supported by the Iowa Flood Center at The University of Iowa. The authors are grateful to the University Information Technology Services and IFC staff who facilitated the HLM simulations using the clusters of high-performance computing resources.
Data availability statement.
The precipitation (IMERG, MRMS, and COOP) and USGS streamflow data are publicly available through corresponding URLs provided in section 2.
REFERENCES
Adhikari, A., M. R. Ehsani, Y. Song, and A. Behrangi, 2020: Comparative assessment of snowfall retrieval from microwave humidity sounders using machine learning methods. Earth Space Sci., 7, e2020EA001357, https://doi.org/10.1029/2020EA001357.
Alemohammad, S. H., K. A. McColl, A. G. Konings, D. Entekhabi, and A. Stoffelen, 2015: Characterization of precipitation product errors across the United States using multiplicative triple collocation. Hydrol. Earth Syst. Sci., 19, 3489–3503, https://doi.org/10.5194/hess-19-3489-2015.
Bai, X., P. Wang, Y. He, Z. Zhang, and X. Wu, 2021: Assessing the accuracy and drought utility of long-term satellite-based precipitation estimation products using the triple collocation approach. J. Hydrol., 603, 127098, https://doi.org/10.1016/j.jhydrol.2021.127098.
Chen, F., and X. Li, 2016: Evaluation of IMERG and TRMM 3B43 monthly precipitation products over mainland China. Remote Sens., 8, 472, https://doi.org/10.3390/rs8060472.
Ciach, G. J., and W. F. Krajewski, 1999: On the estimation of radar rainfall error variance. Adv. Water Resour., 22, 585–595, https://doi.org/10.1016/S0309-1708(98)00043-8.
Clark, M. P., and Coauthors, 2021: The abuse of popular performance metrics in hydrologic modeling. Water Resour. Res., 57, e2020WR029001, https://doi.org/10.1029/2020WR029001.
Cunha, L. K., P. V. Mandapaka, W. F. Krajewski, R. Mantilla, and A. A. Bradley, 2012: Impact of radar-rainfall error structure on estimated flood magnitude across scales: An investigation based on a parsimonious distributed hydrological model. Water Resour. Res., 48, W10515, https://doi.org/10.1029/2012WR012138.
Ehsani, M. R., and A. Behrangi, 2022: A comparison of correction factors for the systematic gauge-measurement errors to improve the global land precipitation estimate. J. Hydrol., 610, 127884, https://doi.org/10.1016/j.jhydrol.2022.127884.
Ehsani, M. R., A. Behrangi, A. Adhikari, Y. Song, G. J. Huffman, R. F. Adler, D. T. Bolvin, and E. J. Nelkin, 2021: Assessment of the Advanced Very High Resolution Radiometer (AVHRR) for snowfall retrieval in high latitudes using CloudSat and machine learning. J. Hydrometeor., 22, 1591–1608, https://doi.org/10.1175/JHM-D-20-0240.1.
Ehsani, M. R., A. Zarei, H. V. Gupta, K. Barnard, E. Lyons, and A. Behrangi, 2022: NowCasting-Nets: Representation learning to mitigate latency gap of satellite precipitation products using convolutional and recurrent neural networks. IEEE Trans. Geosci. Remote Sens., 60, 4706021, https://doi.org/10.1109/TGRS.2022.3158888.
ElSaadani, M., W. F. Krajewski, R. Goska, and M. B. Smith, 2018: An investigation of errors in distributed models’ stream discharge prediction due to channel routing. J. Amer. Water Resour. Assoc., 54, 742–751, https://doi.org/10.1111/1752-1688.12627.
Ghimire, G. R., W. F. Krajewski, and R. Mantilla, 2018: A power law model for river flow velocity in Iowa basins. J. Amer. Water Resour. Assoc., 54, 1055–1067, https://doi.org/10.1111/1752-1688.12665.
Ghimire, G. R., W. F. Krajewski, T. B. Ayalew, and R. Goska, 2022: Hydrologic investigations of radar-rainfall error propagation to rainfall-runoff model hydrographs. Adv. Water Resour., 161, 104145, https://doi.org/10.1016/j.advwatres.2022.104145.
Gilleland, E., D. Ahijevych, B. G. Brown, B. Casati, and E. E. Ebert, 2009: Intercomparison of spatial forecast verification methods. Wea. Forecasting, 24, 1416–1430, https://doi.org/10.1175/2009WAF2222269.1.
Gruber, A., C.-H. Su, S. Zwieback, W. Crow, W. Dorigo, and W. Wanger, 2016: Recent advance in (soil moisture) triple collocation analysis. Int. J. Appl. Earth Obs. Geoinf., 45, 200–211, https://doi.org/10.1016/j.jag.2015.09.002.
Gupta, H. V., H. Kling, K. K. Yilmaz, and G. F. Martinez, 2009: Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. J. Hydrol., 377, 80–91, https://doi.org/10.1016/j.jhydrol.2009.08.003.
Habib, E., A. T. Haile, N. Sazib, Y. Zhang, and T. Rientjes, 2014: Effect of bias correction of satellite-rainfall estimates on runoff simulations at the source of the Upper Blue Nile. Remote Sens., 6, 6688–6708, https://doi.org/10.3390/rs6076688.
Hamada, A., and Y. N. Takayabu, 2016: Improvements in detection of light precipitation with the global precipitation measurement dual-frequency precipitation radar (GPM DPR). J. Atmos. Oceanic Technol., 33, 653–667, https://doi.org/10.1175/JTECH-D-15-0097.1.
Hartke, S. H., D. B. Wright, F. Quintero, and A. S. Falck, 2023: Incorporating IMERG satellite precipitation uncertainty into seasonal and peak streamflow predictions using the hillslope link hydrological model. J. Hydrol. X, 18, 100148, https://doi.org/10.1016/j.hydroa.2023.100148.
Huffman, G. J., and Coauthors, 2020: NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG). NASA Tech. Rep., 35 pp., https://gpm.nasa.gov/sites/default/files/2020-05/IMERG_ATBD_V06.3.pdf.
Jiang, L., and P. Bauer-Gottwein, 2019: How do GPM IMERG precipitation estimates perform as hydrological model forcing? Evaluation for 300 catchments across Mainland China. J. Hydrol., 572, 486–500, https://doi.org/10.1016/j.jhydrol.2019.03.042.
Khodadoust Siuki, S., B. Saghafian, and S. Moazami, 2017: Comprehensive evaluation of 3-hourly TRMM and half-hourly GPM-IMERG satellite precipitation products. Int. J. Remote Sens., 38, 558–571, https://doi.org/10.1080/01431161.2016.1268735.
Kitchen, M., and R. M. Blackall, 1992: Representativeness errors in comparisons between radar and gauge measurements of rainfall. J. Hydrol., 134, 13–33, https://doi.org/10.1016/0022-1694(92)90026-R.
Knoben, W. J. M., J. E. Freer, and R. A. Woods, 2019: Inherent benchmark or not? Comparing Nash-Sutcliffe and Kling-Gupta efficiency scores. Hydrol. Earth Syst. Sci., 23, 4323–4331, https://doi.org/10.5194/hess-23-4323-2019.
Krajewski, W. F., and Coauthors, 2017: Real-time flood forecasting and information system for the state of Iowa. Bull. Amer. Meteor. Soc., 98, 539–554, https://doi.org/10.1175/BAMS-D-15-00243.1.
Lamontagne, J. R., C. A. Barber, and R. M. Vogel, 2020: Improved estimators of model performance efficiency for skewed hydrologic data. Water Resour. Res., 56, e2020WR027101, https://doi.org/10.1029/2020WR027101.
Lawrimore, J. H., and Coauthors, 2020: Quality control and processing of cooperative observer program hourly precipitation data. J. Hydrometeor., 21, 1811–1825, https://doi.org/10.1175/JHM-D-19-0300.1.
Levizzani, V., and E. Cattani, 2019: Satellite remote sensing of precipitation and the terrestrial water cycle in a changing climate. Remote Sens., 11, 2301, https://doi.org/10.3390/rs11192301.
Levizzani, V., S. Laviola, and E. Cattani, 2011: Detection and measurement of snowfall from space. Remote Sens., 3, 145–166, https://doi.org/10.3390/rs3010145.
Li, Z., D. B. Wright, S. Q. Zhang, D. B. Kirschbaum, and S. H. Hartke, 2020: Object-based comparison of data-driven and physics-driven satellite estimates of extreme rainfall. J. Hydrometeor., 21, 2759–2776, https://doi.org/10.1175/JHM-D-20-0041.1.
Li, Z., and Coauthors, 2021: Two decades of GPM IMERG early and final run products intercomparison: Similarity and difference in climatology, rates, and extremes. J. Hydrol., 594, 125975, https://doi.org/10.1016/j.jhydrol.2021.125975.
Liu, G., and E.-K. Seo, 2013: Detecting snowfall over land by satellite high‐frequency microwave observations: The lack of scattering signature and a statistical approach. J. Geophys. Res. Atmos., 118, 1376–1387, https://doi.org/10.1002/jgrd.50172.
Liu, Z., 2016: Comparison of Integrated Multisatellite Retrievals for GPM (IMERG) and TRMM Multisatellite Precipitation Analysis (TMPA) monthly precipitation products: Initial results. J. Hydrometeor., 17, 777–790, https://doi.org/10.1175/JHM-D-15-0068.1.
Maggioni, V., H. J. Vergara, E. N. Anagnostou, J. J. Gourley, Y. Hong, and D. Stampoulis, 2013: Investigating the applicability of error correction ensembles of satellite rainfall products in river flow simulations. J. Hydrometeor., 14, 1194–1211, https://doi.org/10.1175/JHM-D-12-074.1.
Mantilla, R., W. F. Krajewski, N. Velasquez, S. J. Small, T. Ayalew, F. Quintero, N. Jadidoleslam, and M. Fonley, 2022: The hydrological Hillslope-Link Model for space-time prediction of streamflow: Insights and applications at the Iowa Flood Center. Extreme Weather Forecasting, M. Astitha and E. I. Nikolopoulos, Eds., Elsevier, 200–238.
McColl, K. A., J. Vogelzang, A. G. Konings, D. Entekhabi, M. Piles, and A. Stoffelen, 2014: Extended triple collocation: Estimating errors and correlation coefficients with respect to an unknown target. Geophys. Res. Lett., 41, 6229–6236, https://doi.org/10.1002/2014GL061322.
Moazami, S., and M. R. Najafi, 2021: A comprehensive evaluation of GPM-IMERG V06 and MRMS with hourly ground-based precipitation observations across Canada. J. Hydrol., 594, 125929, https://doi.org/10.1016/j.jhydrol.2020.125929.
Mu, Q., M. Zhao, and S. W. Running, 2011: Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sens. Environ., 115, 1781–1800, https://doi.org/10.1016/j.rse.2011.02.019.
Nikolopoulos, E. I., E. N. Anagnostou, F. Hossain, M. Gebremichael, and M. Borga, 2010: Understanding the scale relationships of uncertainty propagation of satellite rainfall through a distributed hydrologic model. J. Hydrometeor., 11, 520–532, https://doi.org/10.1175/2009JHM1169.1.
O, S., U. Foelsche, G. Kirchengast, J. Fuchsberger, J. Tan, and W. A. Petersen, 2017: Evaluation of GPM IMERG early, late, and final rainfall estimates using WegenerNet gauge data in southeastern Austria. Hydrol. Earth Syst. Sci., 21, 6559–6572, https://doi.org/10.5194/hess-21-6559-2017.
Palash, W., Y. Jiang, A. S. Akanda, D. L. Small, A. Nozari, and S. Islam, 2018: A streamflow and water level forecasting model for the Ganges, Brahmaputra, and Meghna Rivers with requisite simplicity. J. Hydrometeor., 19, 201–225, https://doi.org/10.1175/JHM-D-16-0202.1.
Quintero, F., and W. F. Krajewski, 2018: Mapping outlets of Iowa flood center and national water center river networks for hydrologic model comparison. J. Amer. Water Resour. Assoc., 54, 28–39, https://doi.org/10.1111/1752-1688.12554.
Quintero, F., W. F. Krajewski, R. Mantilla, S. Small, and B.-C. Seo, 2016: A Spatial–dynamical framework for evaluation of satellite rainfall products for flood prediction. J. Hydrometeor., 17, 2137–2154, https://doi.org/10.1175/JHM-D-15-0195.1.
Quintero, F., W. F. Krajewski, B.-C. Seo, and R. Mantilla, 2020a: Improvement and evaluation of the Iowa flood center Hillslope Link Model (HLM) by calibration-free approach. J. Hydrol., 584, 124686, https://doi.org/10.1016/j.jhydrol.2020.124686.
Quintero, F., W. F. Krajewski, and M. Rojas, 2020b: A flood potential index for effective communication of streamflow forecasts at ungauged communities. J. Hydrometeor., 21, 807–814, https://doi.org/10.1175/JHM-D-19-0212.1.
Schaefer, J. T., 1990: The critical success index as an indicator of warning skill. Wea. Forecasting, 5, 570–575, https://doi.org/10.1175/1520-0434(1990)005<0570:TCSIAA>2.0.CO;2.
Scofield, R. A., and R. J. Kuligowski, 2003: Status and outlook of operational satellite precipitation algorithms for extreme-precipitation events. Wea. Forecasting, 18, 1037–1051, https://doi.org/10.1175/1520-0434(2003)018<1037:SAOOOS>2.0.CO;2.
Seo, B.-C., and W. F. Krajewski, 2010: Scale dependence of radar rainfall uncertainty: Initial evaluation of NEXRAD’s new super-resolution data for hydrologic applications. J. Hydrometeor., 11, 1191–1198, https://doi.org/10.1175/2010JHM1265.1.
Seo, B.-C., and W. F. Krajewski, 2011: Investigation of the scale-dependent variability of radar-rainfall and rain gauge error covariance. Adv. Water Resour., 34, 152–163, https://doi.org/10.1016/j.advwatres.2010.10.006.
Seo, B.-C., and W. F. Krajewski, 2021: Multi-scale investigation of conditional errors in radar-rainfall estimates. Adv. Water Resour., 156, 104041, https://doi.org/10.1016/j.advwatres.2021.104041.
Seo, B.-C., B. Dolan, W. F. Krajewski, S. A. Rutledge, and W. Petersen, 2015: Comparison of single- and dual-polarization-based rainfall estimates using NEXRAD data for the NASA Iowa flood studies project. J. Hydrometeor., 16, 1658–1675, https://doi.org/10.1175/JHM-D-14-0169.1.
Seo, B.-C., and Coauthors, 2018a: Comprehensive evaluation of the IFloodS radar-rainfall products for hydrologic applications. J. Hydrometeor., 19, 1793–1813, https://doi.org/10.1175/JHM-D-18-0080.1.
Seo, B.-C., F. Quintero, and W. F. Krajewski, 2018b: High-resolution QPF uncertainty and its implication for flood prediction: A case study for the eastern Iowa flood of 2016. J. Hydrometeor., 19, 1289–1304, https://doi.org/10.1175/JHM-D-18-0046.1.
Seo, B.-C., W. F. Krajewski, F. Quintero, S. Buan, and B. Connelly, 2021: Assessment of streamflow predictions generated using multimodel and multiprecipitation product forcing. J. Hydrometeor., 22, 2275–2290, https://doi.org/10.1175/JHM-D-20-0310.1.
Skofronick‐Jackson, G., and B. T. Johnson, 2011: Surface and atmospheric contributions to passive microwave brightness temperatures for falling snow events. J. Geophys. Res., 116, D02213, https://doi.org/10.1029/2010JD014438.
Skofronick‐Jackson, G., D. Kirschbaum, W. Petersen, G. Huffman, C. Kidd, E. Stocker, and R. Kakar, 2018: The Global Precipitation Measurement (GPM) mission’s scientific achievements and societal contributions: Reviewing four years of advanced rain and snow observations. Quart. J. Roy. Meteor. Soc., 144, 27–48, https://doi.org/10.1002/qj.3313.
Small, S. J., L. O. Jay, R. Mantilla, R. Curtu, L. K. Cunha, M. Fonley, and W. F. Krajewski, 2013: An asynchronous solver for systems of ODEs linked by a directed tree structure. Adv. Water Resour., 53, 23–32, https://doi.org/10.1016/j.advwatres.2012.10.011.
Smith, E. A., and Coauthors, 2007: International Global Precipitation Measurement (GPM) program and mission: An overview. Measuring Precipitation from Space—EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, Eds., Springer, 611–653.
Sorooshian, S., K.-L. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 2035–2046, https://doi.org/10.1175/1520-0477(2000)081<2035:EOPSSE>2.3.CO;2.
Souverijns, N., A. Gossart, S. Lhermitte, I. V. Gorodetskaya, S. Kneifel, M. Maahn, F. L. Bliven, and N. P. M. van Lipzig, 2017: Estimating radar reflectivity—Snowfall rate relationships and their uncertainties over Antarctica by combining disdrometer and radar observations. Atmos. Res., 196, 211–223, https://doi.org/10.1016/j.atmosres.2017.06.001.
Sperna Weiland, F. C., J. A. Vrugt, R. L. P. H. van Beek, A. H. Weerts, and M. F. Bierkens, 2015: Significant uncertainty in global scale hydrological modeling from precipitation data errors. J. Hydrol., 529, 1095–1115, https://doi.org/10.1016/j.jhydrol.2015.08.061.
Stephens, C. M., H. T. Pham, L. A. Marshall, and F. M. Johnson, 2022: Which rainfall errors can hydrologic models handle? Implications for using satellite-derived products in sparsely gauged catchments. Water Resour. Res., 58, e2020WR029331, https://doi.org/10.1029/2020WR029331.
Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103, 7755–7766, https://doi.org/10.1029/97JC03180.
Tan, J., W. A. Petersen, and A. Tokay, 2016: A novel approach to identify sources of errors in IMERG for GPM ground validation. J. Hydrometeor., 17, 2477–2491, https://doi.org/10.1175/JHM-D-16-0079.1.
Tang, G., Z. Zeng, D. Long, X. Guo, B. Yong, W. Zhang, and Y. Hong, 2016: Statistical and hydrological comparisons between TRMM and GPM level-3 products over a midlatitude basin: Is Day-1 IMERG a good successor for TMPA 3B42V7? J. Hydrometeor., 17, 121–137, https://doi.org/10.1175/JHM-D-15-0059.1.
Villarini, G., and W. F. Krajewski, 2010: Review of the different sources of uncertainty in single polarization radar-based estimates of rainfall. Surv. Geophys., 31, 107–129, https://doi.org/10.1007/s10712-009-9079-x.
Vivoni, E. R., D. Entekhabi, and R. N. Hoffman, 2007: Error propagation of radar rainfall nowcasting fields through a fully distributed flood forecasting model. J. Appl. Meteor. Climatol., 46, 932–940, https://doi.org/10.1175/JAM2506.1.
Walker, V. A., B. K. Hornbuckle, M. H. Cosh, and J. H. Prueger, 2019: Seasonal evaluation of SMAP soil moisture in the U.S. Corn Belt. Remote Sens., 11, 2488, https://doi.org/10.3390/rs11212488.
Wang, Z., R. Zhong, C. Lai, and J. Chen, 2017: Evaluation of the GPM IMERG satellite-based precipitation products and the hydrologic utility. Atmos. Res., 196, 151–163, https://doi.org/10.1016/j.atmosres.2017.06.020.
Wright, D. B., D. B. Kirschbaum, and S. Yatheendradas, 2017: Satellite precipitation characterization, error modeling, and error correction using censored shifted gamma distributions. J. Hydrometeor., 18, 2801–2815, https://doi.org/10.1175/JHM-D-17-0060.1.
Wu, H., R. F. Adler, Y. Hong, Y. Tian, and F. Policelli, 2012: Evaluation of global flood detection using satellite-based rainfall and a hydrologic model. J. Hydrometeor., 13, 1268–1284, https://doi.org/10.1175/JHM-D-11-087.1.
Wu, H., R. F. Adler, Y. Tian, G. J. Huffman, H. Li, and J. Wang, 2014: Real-time global flood estimation using satellite-based precipitation and a coupled land surface and routing model. Water Resour. Res., 50, 2693–2717, https://doi.org/10.1002/2013WR014710.
Xiang, Z., and I. Demir, 2020: Distributed long-term hourly streamflow prediction using deep learning – A case study for State of Iowa. Environ. Modell. Software, 131, 104761, https://doi.org/10.1016/j.envsoft.2020.104761.
You, Y., N.-Y. Wang, R. Ferraro, and S. Rudlosky, 2017: Quantifying the snowfall detection performance of the GPM Microwave Imager channels over land. J. Hydrometeor., 18, 729–751, https://doi.org/10.1175/JHM-D-16-0190.1.
Zhang, J., and Coauthors, 2016: Multi-Radar Multi-Sensor (MRMS) quantitative precipitation estimation: Initial operating capabilities. Bull. Amer. Meteor. Soc., 97, 621–638, https://doi.org/10.1175/BAMS-D-14-00174.1.
Zhang, J., L. Tang, S. Cocks, P. Zhang, A. Ryzhkov, K. Howard, C. Langston, and B. Kaney, 2020: A dual-polarization radar synthetic QPE for operations. J. Hydrometeor., 21, 2507–2521, https://doi.org/10.1175/JHM-D-19-0194.1.
