1. Introduction
Accurate representation of spatial and temporal precipitation patterns and structures plays a critical role in hydrological and climatological studies and water resource management applications (Derin and Yilmaz 2014; Li et al. 2015, 2013; Mohammed et al. 2020; Tang et al. 2020; Wu et al. 2012). To achieve this goal, accurate high spatiotemporal resolution gridded precipitation datasets are needed. These products offer the potential to improve our understanding of the physical processes underlying the storm’s structure (Pepler et al. 2018) and creates an opportunity to address challenges in the study of rainfall extremes from short-duration convective storms (Li et al. 2018; Roca et al. 2019; Westra et al. 2014) to long-lasting, larger-scale storms with complex structures, leading to costly damages (Dowdy et al. 2019).
Gridded precipitation datasets have been produced using three approaches: 1) gauge data interpolation, 2) ground-based radar observations, and 3) satellite precipitation retrievals, each of which has specific characteristics and limitations. The accuracy of interpolated gauge data depends largely on gauge network density and measurement quality (X. Zhang et al. 2018; Prein and Gobiet 2017). In many regions, gauge stations are poorly distributed, particularly over mountainous areas due to the difficulties in installation and maintenance costs. Even in areas with relatively high gauge density it has been shown that most hourly heavy precipitation events are missed by the station network (Lengfeld et al. 2020).
Weather radars can provide precipitation products with high spatiotemporal resolution, though they are affected by factors such as poor coverage, beam blockage in complex regions, and expensive operating and maintenance costs (Derin et al. 2019; Gebregiorgis et al. 2017; Krajewski and Smith 2002; Young et al. 2000). These issues limit the applicability of these products to certain populated regions in some countries (Ayat et al. 2018; Ghaemi et al. 2017; X. Zhang et al. 2018). To overcome some of the above mentioned limitations, efforts have been made to produce multisensor products like Multi-Radar Multi-Sensor (MRMS; United States) or Radar-Online-Aneichung (RADOLAN; Germany) that benefit from multiple sources of data including radar data, model outputs and ground observations. However, their coverage is limited to some specific regions in the world.
Multisensor satellite precipitation products offer a uniform global coverage including the mountainous or less populated regions (Prat and Nelson 2015). The recent progress in producing high spatiotemporal resolution satellite products, like IMERG provided by the recently launched Global Precipitation Measurement (GPM) mission (Huffman et al. 2019), has significantly advanced the ability to better capture the spatial and temporal variability of precipitation. However, evaluation of these datasets is still a challenging task (Gebregiorgis et al. 2018; Tang et al. 2020). Many studies have compared satellite precipitation estimates with gauge-based observations (Derin and Yilmaz 2014; Freitas et al. 2020; Guo et al. 2016; Khodadoust Siuki et al. 2017; Li et al. 2013; Moazami et al. 2013, 2014, 2016; Mohammed et al. 2020; O et al. 2017; Sharifi et al. 2016; Tang et al. 2020; Zhang et al. 2019). In most of these studies, the comparison was carried out on a daily or monthly time scale, which is not sufficient to monitor highly dynamic spatial and temporal processes. Therefore, ground-based radar data plays an important part in the assessment of satellite estimates using coincident samples (Porcú et al. 1999).
A number of studies have compared satellite precipitation estimates to ground radar products. For instance, Gaona et al. (2016) evaluated the 30-min gridded Integrated Multisatellite Retrievals for GPM (IMERG) Final Run (V03D) based on gauge-adjusted radar rainfall over the Netherlands at 30-min, 24-h, monthly, and yearly scales. The results indicated a slight underestimation for IMERG in countrywide rainfall depth. The two products have better agreement at coarser temporal resolutions. Similar findings are presented by Tan et al. (2017) over the United States when comparing IMERG version 3 with MRMS, showing better agreement between the two datasets when scaling up to a coarser spatiotemporal resolution. However, a positive mean difference was observed over the selected domain at high spatiotemporal resolutions when taking MRMS as the reference.
In another study, carried out by Chen et al. (2020), the differences between MRMS and GPM products were investigated during Hurricane Harvey and the results showed that IMERG V6 tended to overestimate the low–moderate precipitation rate but underestimate the extreme rain rates compared to MRMS. This is also reported by Li et al. (2020) in cross examination of three products [MRMS, IMERG version 5, and National Centers for Environmental Prediction (NCEP) gridded gauge-only hourly precipitation], and they employed both traditional metrics and the multiplicative triple collection (MTC) method during Hurricane Harvey and multiple tropical cyclones. Omranian et al. (2018) compared the hourly precipitation in IMERG version 05-B with NCEP Stage IV radar data during Hurricane Harvey. They found that the correlation between the satellite and radar data was generally high; however, it decreased significantly over the storm core. Further studies have compared satellite precipitation retrievals with ground radar estimates (Beck et al. 2019; Chen et al. 2020; ElSaadani et al. 2018; Furl et al. 2018; Gebregiorgis et al. 2018; Khan et al. 2018; Prat and Nelson 2015; Rios Gaona et al. 2017; J. Zhang et al. 2018; X. Zhang et al. 2018). There are also some studies that have shown the dependency of the satellite estimates quality to topography, different precipitation types (e.g., snow or rain; Sadeghi et al. 2019, Wen et al. 2018; Behrangi et al. 2018) and seasons (Derin et al. 2019; O et al. 2017; Panahi and Behrangi 2020). While satellite products may not effectively capture orographic precipitation enhancement (e.g., Shige et al. 2013), radar products may also miss precipitation rates due to radar beam blockage over mountainous regions. All of these studies focused on precipitation detection and magnitude on a pixel-by-pixel basis. Although these verification metrics provide valuable information, they are not able to explicitly quantify errors related to size, spatial patterns, location, and other characteristics of storms (Li et al. 2015). In addition, issues like the double penalty error impact the results when comparing high-resolution datasets (Prein et al. 2013).
An object-based approach can address some of these limitations. Based on this approach, groups of connected pixels satisfying a certain condition (e.g., higher than a specified threshold) are considered to be independent objects, referred to as single storms. Thus, characteristics like size, shape, orientation, translation speed, etc. can be extracted and compared, which is not achievable using the pixel-based approach. There are limited studies that compared satellite and ground radar with an object-based approach. Li et al. (2016) compared multiple high-resolution satellite products including Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN), Climate Prediction Center morphing technique (CMORPH), and Tropical Rainfall Measurement Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) 3B42RT with NCEP Stage IV radar data in the warm season of 2008 and cold season of 2010, and they found that PERSIANN performed best in capturing the orientation of the objects, 3B42RT depict the location of the storms better than the other products and in terms of the object size, CMORPH is the best. However, the objects were not tracked in this research to capture the time-dependent characteristics of storms in the datasets. Cui et al. (2020) investigated the mesoscale convective systems (MCSs) characteristics in the central and eastern United States during a 3-yr study period by implementing an object-based approach over IMERG V06B and NCEP Stage IV datasets. The findings indicated that IMERG agrees reasonably well with Stage IV; however, IMERG tended to overestimate the total precipitation and underestimate the hourly mean precipitation. They found errors related to the evaporation of light rain and misclassification of anvil regions as precipitating pixels or the additional false detection caused by the morphing scheme in the IMERG algorithm.
The goal of this study is to compare a merged satellite dataset (IMERG V06B) with a high-resolution ground-based merged radar product (MRMS) during the period 2015–19 over the eastern United States using an object-based approach. In this study, a wide range of storm scales from small convective storms to large long-lasting hurricanes are tracked and comprehensive geometrical and time-dependent characteristics of storms are compared between the two datasets. A secondary goal of this study is to examine the capability of both products in capturing the object-based characteristics of landfalling hurricane events. An effort is made to assess the products’ differences in relation to the storm characteristics in these events. Finally, both products were compared in capturing some pixel-based storm characteristics and the results were compared with their object-based counterparts.
2. Datasets
The datasets selected for the present study are IMERG V06B and MRMS in order to compare storms from satellite and ground radar points of view, respectively. MRMS is a high-resolution mosaic dataset over the United States and its coverage is large enough to track large storms like hurricanes. IMERG is a global satellite product that has attracted the attention of many researchers in recent years in hydrology and climatology fields of study (Liu 2016).
a. IMERG
In this research, we used the latest release of the IMERG (V06B) product. IMERG provides gridded precipitation maps with high spatiotemporal resolution (0.1° × 0.1° every 0.5 h) within 90°S–N. Details of the retrieval method are described in Huffman et al. (2019) and in brief includes four main steps:
Precipitation estimates from the GPM constellation radiometers are gridded, intercalibrated to the radar–microwave combined product (2BCMB), and combined into half-hourly 0.1° × 0.1° fields (variable name: HQprecipitaiton).
Maps of half-hourly infrared (IR) precipitation rate (IRprecipitation) are calculated using an IR-based precipitation retrieval method (PERSIANN-CCS; Hong et al. 2004).
MW and IR estimates are used to create half-hourly estimates (precipitationUncal) by utilizing the Climate Prediction Center morphing–Kalman filter (CMORPH-KF) Lagrangian time interpolation scheme.
The multisatellite half-hour estimates are adjusted so that they sum to a monthly satellite–gauge combination (precipitationCal).
For this study, the final version of the IMERG algorithm, which has the best overall agreement with ground observation, has been employed. The IMERG product was obtained from the Goddard Earth Sciences Data and Information Services Center (GES DISC, https://disc.gsfc.nasa.gov/).
b. MRMS
MRMS is a high-resolution radar product based on Weather Surveillance Radar-1988 Doppler (WSR-88D) consists of 146 S-band and 30 C-band weather radars at 1-km and 2-min spatial and temporal resolution, respectively, over the conterminous United States (CONUS) and southern Canada (Zhang et al. 2016). MRMS was deployed operationally in 2014 and benefits from 7000 hourly tipping-bucket rain gauges from the Hydrometeorological Automated Data System (HADS) to adjust radar estimate biases. In addition, a numerical weather prediction (NWP) model called the Rapid Refresh model (RAP) has been employed extensively in this product in order to find the location of the melting layer and distinguish the snow and rain regions. A special dataset of monthly precipitation climatology called “Mountain Mapper” is also used to improve precipitation estimates in mountainous regions (Zhang et al. 2016). The MRMS surface precipitation rate product has been used in this study and estimates precipitation based on a sophisticated algorithm using multiple R–Z relationships (Zhang et al. 2016).
c. Key considerations for dataset comparison
Since the temporal and spatial resolution of the two products are different, in order to have a fair comparison the MRMS maps are temporally averaged to the IMERG temporal resolution (30 min), then remapped onto the IMERG spatial grid using an area-weighted conservative method. So, a target cell in the upscaled dataset is the average of source cells over the area intersected by the target cell.
Ground radars and satellites are monitoring storms from different viewing angles. IR sensors use the cloud-top information to estimate surface rainfall (Hsu et al. 1996). Passive microwave (PMW) retrievals use lower frequencies (e.g., 19 GHz) over water bodies where radiation is absorbed and reemitted by water droplets and provide us with the column-integrated liquid water information, while over land radiation at higher frequencies (e.g., 85 GHz) is used though it is strongly affected by ice scattering near the top of the clouds. (Petković and Kummerow 2017). IR sensors contributing to satellite precipitation products use the information of cloud-top temperature to estimate the surface precipitation. Thus, the top-down view of satellites leads to strong consideration of information in upper atmospheric levels to estimate surface rainfall potentially missing evaporation effects in PMW retrievals and misrepresentation of nonprecipitating cold clouds as precipitation in IR estimates (Behrangi et al. 2009a,b), whereas ground radar’s estimates come from a cross section of storms at a given altitude in the atmosphere and may miss phase changes including evaporation below this level.
It should be noted that MRMS is merging multiple sensors in space while IMERG is merging multiple sensors in time. Some of the issues that arise using a single ground-based radar (i.e., restrictions in spatial and temporal coverage, beam blockage in mountainous regions, and refraction due to vertical temperature variations; Langston et al. 2007; Seo et al. 2011) can be addressed by spatially merging multiple radars. However, this is a challenging task that creates uncertainties in the final merged product (Seo et al. 2014). Similarly, combining IR retrievals with PMW estimates is inevitable in global satellite datasets like IMERG due to the poor temporal and spatial coverage of PMW sensors (Derin et al. 2018). Therefore, uncertainty originating specifically from multiple sensors and error characteristics rooted in the merging algorithm is one of the main sources of bias affecting the merged end products (Derin et al. 2018; Gebregiorgis et al. 2017; Tang et al. 2014).
3. Method
a. Object-based technique
Method of Object-based Diagnostic Evaluation (MODE) Time Domain (MTD), a part of the Developmental Testbed Center’s (DTC) Model Evaluation Tools (MET), is an extension for the MODE in order to track the objects identified by the MODE algorithm (Clark et al. 2014). This technique provides us with the possibility of studying the time-dependent storm’s characteristics like storm initiation, evolution, dissipation, lifetime, and speed, in addition to the other characteristics such as area, intensity, concentration, etc., derived from MODE analysis (Wen 2016).
Similar to other object-based techniques, this method uses a thresholding approach to extract objects from data fields, but the threshold is applied to the convolved data not directly on the data field’s variable (here the precipitation rate). A pixel value in the convolved data is derived from the averaging of the equivalent pixel and its surrounding pixels that fall within a certain distance from the pixel in the raw dataset. The resolved objects are the connected pixels (in the convolved data field) that are higher than a user-defined threshold (Bullock et al. 2016; Davis et al. 2009). Lower thresholds lead to larger object areas and more contiguous properties, while higher thresholds result in isolating the central convective parts of storms (Bytheway and Kummerow 2015). The selected threshold in this study is 1 mm h−1, applied on the datasets convolved by a 3 pixel × 3 pixel window. To exclude the issues in calculating the storm object properties, a threshold of 10 pixels (1000 km2) has been considered in tracking the storm objects.
In addition to MTD, further analysis was performed to deal with objects that split or merge during the storm lifetime, a known problem for storm identification and tracking algorithms (Zan et al. 2019). In our approach, either splitting or merging leads to the creation of a new sequence of object(s), which are connected to the previous sequence of the object(s). Thus, all the information regarding all aspects of objects contributing to this storm can be accessed using graph theory (see Fig. 1). Figure 2 illustrates an example of detected storm objects during Hurricane Matthew using MTD.
Graph of tracked contributing storm objects with split/merge events during the lifetime of one storm (see section 3c). Note that each blob represents a storm object at each time step.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Hurricane Matthew precipitation snapshots in (a) IMERG and (b) MRMS at 2000 UTC 6 Oct, 0430 UTC 8 Oct, and 1300 UTC 9 Oct 2016 with shape-match factors of 0.27, 0.04, and 0.4, respectively (see section 3c for more information).
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
b. Statistics
In Eqs. (4)–(6), “hit” denotes the number of rain pixels that have been detected in both datasets, “false” is rain pixels detected only in the IMERG product, and “miss” is rain pixels detected only in MRMS.
In this study, the Theil–Sen regression method was employed as an alternative to the parametric least squares regression line to illustrate the linear trend of a variable against another quantity. This method is a nonparametric method that is resistant to outliers. In this method the median of the slopes determined by all the possible pairs of the two variables is considered as the best-fitted line. A confidence interval may also be considered to quantify the uncertainty of the calculated slopes.
c. Object characteristics definition
We define an “storm object” as a group of connected pixels the value of which are higher than the selected threshold (here 1 mm h−1) in the convolved precipitation dataset at each time step. The storm object characteristics of interest in this study include 1) area, 2) translation speed, 3) object maximum intensity, 4) object average intensity, 5) object volume precipitation, 6) aspect ratio, and 7) orientation, all of which are defined in Table 1.
The defined storm object characteristics in comparing IMERG and MRMS.
We define a “storm” as a set of objects that are connected via split/merge events. The studied storm characteristics (defined in Table 2) in this research include 1) maximum track length, 2) storm-contributing objects, 3) storm split/merge number, and 4) storm lifetime.
The defined storm characteristics in comparing IMERG and MRMS.
In Fig. 1, this score has been calculated for three different snapshots of Hurricane Matthew. The calculated shape-match factors show that the shape of the IMERG hurricane object at 0430 UTC 8 October 2016 has the best agreement with MRMS hurricane objects among the other counterparts. Although both datasets at 1300 UTC 9 October 2016 have a more similar 2D shape of the hurricane object, the calculated shape-match factor is worse than the previous snapshot since this parameter compares the 3D shape of the storm objects (with intensity as z axis) in both datasets.
4. Study area
The time frame of this research covers the overlap period of the two datasets from the end of 2014 up to mid-2019, which also includes seven hurricanes (see Table 3) that occurred over the United States. Hurricanes are capable of producing extremely heavy precipitation, leading to costly damage, floods and storm surge. Thus, representing the hurricanes and their properties in datasets is a matter of great importance, and this collection of hurricanes allows us to compare hurricane rainfall characteristics from satellite and ground-based radar points of view. In this study, an object-based comparison has been carried out to compare the two products both over the whole selected period and separately during hurricane events.
Details of the selected hurricanes (Berg 2017; Beven and Berg 2018; Beven et al. 2019; Blake and Zelinsky 2018; Cangialosi et al. 2018; Stewart 2017; Stewart and Berg 2019).
Since the beam height increases with distance from the radar, precipitation is often undetected or underestimated at long distances like over ocean and Canada (Hunter 1996; Kitchen and Jackson 1993; Scofield and Kuligowski 2003; Smith et al. 1996). Over Canada and the U.S.–Canadian border the prevalent type of precipitation is snowfall, which adversely affects the radar data quality (Nelson et al. 2016; Smalley et al. 2014). Figure 3 indicates the total precipitation that occurred over the whole period. It shows that the agreement between MRMS and IMERG is best over the land regions limited to the United States borders. However, radar coverage over the western United States is poor due to the complex topography (Nelson et al. 2016; Prein et al. 2017; Smalley et al. 2014; Zick and Matyas 2016). Therefore, the study area is restricted to the eastern half of the continental United States (CONUS) where rainfall dominates and the radar coverage is large enough to capture the hurricanes’ tracks over the MRMS coverage. Note that for better representation of the hurricane tracks, the data over both land and ocean are included in Figs. 2 and 3 and Figs. S1 and S8 in the online supplemental material, but not in the analyses.
Annual precipitation over the eastern United States observed in (a) IMERG and (b) MRMS. (c) The bias precipitation with MRMS as the reference, and (d) bias PDFs over the United States and out of U.S. borders.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
5. Results
a. Object-based comparison of IMERG and MRMS
This section presents a comparison of the datasets based on all storm objects identified during the study period. A total of 1 399 917 storm objects are identified from MRMS and 1 143 189 from IMERG. All of these storm objects contributed to the statistics given below for the characteristics of area, volume precipitation, averaged intensity, maximum intensity, aspect ratio, and orientation. Due to issues such as the splitting/merging of storm objects however, translation speed and direction were obtained for only around 60% of these storm objects. Also, the detected storm objects that are connected via split/merging (see Fig. 1) produce 124 660 (113 537) storms in MRMS (IMERG) during the study period.
Figure 4 shows an overview of the storm characteristic distributions for all storm objects that occurred during the study period in either product. Overall, IMERG and MRMS capture similar storm characteristics. Most of the contributing storm objects have small areas, volume precipitation, and averaged intensity, with a maximum intensity around 2 mm h−1. Most of the detected storm objects are moving eastward with a translation speed around 20 km h−1, with the major storm axis oriented around 75° with an aspect ratio close to 0.4 in both datasets. However, there are differences that are statistically significant at the 0.01 level based on the Kolmogorov–Smirnov test. For instance, Fig. 4a shows that MRMS storm objects’ translation speeds are concentrated around a smaller value than the IMERG counterparts, meaning that some storm objects are moving slightly faster in IMERG. Similar results are obtained from Figs. 4b–d, which generally show that more MRMS storm objects have a smaller area and volume precipitation and lower averaged intensity. However, maximum intensities in MRMS storm objects are more intense in comparison with IMERG (Fig. 4e). IMERG observes a few more storm objects with an eastward direction and 75° orientation angle compared to MRMS (see Figs. 4g,h). However, Fig. 4f indicates that the MRMS storm objects are slightly more linear (lower aspect ratio) than the IMERG counterparts.
PDFs of storm object characteristics for the period (2015–19) defined in section 3c. All PDF differences are significant at the 0.01 level based on the Kolmogorov–Smirnov test. Orange depicts MRMS, and blue depicts IMERG records.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Based on Figs. 4 and 4i–l, both datasets indicate that most storms have few contributing objects (little splitting/merging), relatively short lifetimes, and maximum track length. However, some storms have hundreds of contributing objects with many split/merge events, can last for more than a week and can travel over 1000 km. It is worth noting that MRMS has more storms with high numbers of contributing objects and split/merge events and long lifetimes than IMERG. Although MRMS storms can last longer than IMERG counterparts, IMERG observes storms with a longer maximum track length. Note that although IMERG represents the raining objects with higher volume of precipitation compared to MRMS, the mean annual precipitation over most places within the U.S. borders are higher in MRMS (see Fig. 3d). This occurs because MRMS storms have more contributing objects that last longer, which increases the total precipitation over a specific point in comparison with IMERG storms. It should be mentioned that some plots are shown in logarithmic scale to better show the differences between the products. Therefore, in some cases like object maximum precipitation, we had to cut the density values less than 0.001, which excludes the objects with maximum precipitation that occurs less than 0.1% of the time (with max precipitation more than 100 mm h−1).
Figures 5a–c compares the object detectability performance based on POD, FAR, and CSI, and Fig. 5d shows the shape similarities between the datasets using the shape-match factor (explained in section 3c) with MRMS as the reference. The distribution concentration is mostly around 0.6 in Fig. 5a, which means, most frequently about 60% of all raining MRMS pixels are typically also detected as raining by IMERG. However, the FAR diagram indicates that more than 50% of pixels seen as raining by IMERG are not identified as such in MRMS, leading to the concentration of FAR distribution over the values greater than 0.5 (Fig. 5b). The concentration of relatively high values of FAR and POD results in having small common areas in comparison with the whole region, this leads to relatively small CSI values that are mostly concentrated around 0.35 (Fig. 5c). Also, Fig. 5d indicates that the shapes of the detected storm objects at each time step are not very well matched, as indicated by values greater than one.
PDFs of object detectability performance in (a) probability of detection (perfect value = 1), (b) false alarm ratio (perfect value = 0), and (c) critical success index (perfect value = 1) with MRMS as the reference dataset. (d) The shape similarities of the detected objects at both datasets and the best performance is indicated by 0. Every single value of the mentioned parameters is calculated for the both object-masked datasets over the whole selected domain; therefore, the distribution population for each parameter is equivalent to the number of time steps across the entire study period (more than 70 000 steps).
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
b. Hurricane objects
The same analysis has been carried out over seven hurricanes (Table 1) that occurred in the study period to compare both products on large-scale storms. To track a hurricane, the hurricane object ID at the first time step has been extracted visually from the datasets. Then, all connected objects due to splitting/merging are included in the hurricane storm. Since we are tracking the hurricane object far into the continent, the hurricane could be transformed to a tropical or an extratropical cyclone that might interact with other storm types. In this study the word “hurricane” includes the entire lifetime of this type of storm. Figure S1 illustrates the hurricane tracks, which are the location of the hurricane objects centroids at each time step. The agreement between the hurricane tracks demonstrates the performance similarity in detecting and tracking the hurricanes in the two datasets.
The time series of four storm characteristics, namely, hurricane area (first column), volume precipitation (second column), averaged intensity (third column), and maximum intensity (fourth column) are shown for each hurricane in Fig. 6. Both products agree well on tracking the hurricanes’ characteristics in time. However, IMERG still observes the hurricanes with a larger area and volume precipitation, but with lower maximum and averaged intensity than MRMS. These results agree with the details in Fig. S2, which illustrates the distribution of hurricane characteristics’ differences as the bias value is positive in the area and volume precipitation and negative in averaged and maximum intensity.
Time series of the area, object volume precipitation, object averaged precipitation, and object maximum precipitation in the selected hurricanes. Colors as in Fig. 4.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Figures 7a and 7b, which are called composite precipitation maps, are derived by averaging all hurricanes precipitation snapshots centered at the centroid of the hurricane precipitation objects. In other words, the composite maps represent the average shape of all selected hurricanes during their lifetimes. Both products are indicating that hurricanes, on average, have an oval shape with maximum intensity at the center and a major axis rotated by 45°. By taking MRMS as the reference, the bias map in Fig. 7c shows that MRMS represents the hurricanes with higher maximum intensity at the center. However, IMERG observes more intense precipitation over the regions around the hurricane center. The radial intensity plots in Fig. 7d show the average variation of rainfall intensity in both products as the distance from the hurricane center. Here we see that hurricane rainfall intensities are more concentrated near the storm center in MRMS.
All hurricane composite intensity maps observed by (a) MRMS and (b) IMERG. (c) Composite intensity bias map is the difference of (a) and (b) by taking the MRMS as the reference. (d) The variation of the averaged radial intensity with the distance from the center of all hurricane composite maps for the MRMS/IMERG (orange/blue) line.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Figure 8 depicts the matched hurricane characteristics’ differences in IMERG and MRMS against the hurricane characteristics that are averaged from both datasets at each time step, with MRMS as the reference. The hurricane area differences between datasets grow when the hurricanes are greater in area, volume precipitation, or have higher maximum intensity (see Figs. 8a–c). The positive tau value indicates that IMERG observes hurricanes with a larger area than MRMS counterparts. These results agree with Figs. S3f and S3j, which indicates that the averaged intensity and volume precipitation increases as the area increases, but with a steeper change in IMERG compared to MRMS. In Figs. 8d and 8e, MRMS tends to observe the hurricanes with higher averaged precipitation intensity when the averaged and maximum precipitation intensity increase. Figure 8f shows that slow-moving hurricanes have higher averaged intensity in MRMS. However, the differences decrease as the hurricanes’ speed increases, which is in agreement with Fig. S3a. Based in Figs. 8g–j, by increasing the hurricane area, volume precipitation, maximum and averaged intensity, MRMS show hurricanes with much higher maximum intensity in comparison with IMERG as the mentioned hurricane characteristics have a steeper relationship with maximum intensity in MRMS.
Scatterplots and probable relations between hurricane characteristics and hurricane characteristics’ differences. The shaded areas are best-fitted lines based on the Theil–Sen method with a confidence interval of 95%. Tau in the figures is Kendall’s tau (see section 3b), and it is significant at the level of 0.01 based on the Kolmogorov–Smirnov test. All the values over the x axis are the average of matched objects’ characteristics from both datasets.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
c. Object-based characteristics for hurricane versus non-hurricane
To investigate if the performance of both products is dependent on the storm structure, we compared the products over two storm categories: 1) hurricanes and 2) non-hurricanes. The non-hurricanes are the storms that occurred during the study period, excluding hurricane days to eliminate the effect of hurricanes. However, there is still significant overlap between the non-hurricane and all datasets. So, both datasets have similar distributions. Figure 9 shows the distribution of hurricane/non-hurricane averaged intensity, maximum intensity, and volume precipitation in the two datasets. At both categories, IMERG observes storm objects with larger precipitation volumes (Figs. 9a,d), but with a lower maximum intensity (Figs. 9b,e). However, POI values indicate that the difference in hurricane volume precipitation and averaged intensity between the two products is more significant than that of non-hurricanes. Besides, MRMS observes the non-hurricanes with lower averaged intensity (Fig. 9c), but Fig. 9f shows the opposite for the hurricanes. However, POI still indicates a greater difference in hurricane averaged intensity between the two datasets. Note that considering all cases, biases in different seasons and for specific storm types (e.g., frontal systems, mesoscale convective systems) can cancel out. This may be another reason why the two datasets differ more during the hurricane events.
Hurricane and non-hurricane (a),(d) volume precipitation; (b),(e) max intensity; and (c),(f) averaged intensity PDFs. All PDFs are significantly different at the 0.01 level using the Kolmogorov–Smirnov test.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Figure 10 compares hurricanes and non-hurricanes in object detectability and shape similarities between MRMS and IMERG. Figure 10a illustrates that the hurricane POD values are much higher than of non-hurricanes (mostly more than 0.8), indicating a significant improvement in detecting MRMS pixels by IMERG during hurricane events. The same results are extracted from FAR and CSI histograms. For instance, in Fig. 10b, the non-hurricane FAR distribution is mostly concentrated around values higher than 0.5. However, in hurricanes, FAR values are mainly focused around a much improved 0.4. The CSI improvement in hurricanes in Fig. 10c also indicates fewer missed/false alarms by IMERG during hurricane events. The MRMS and IMERG hurricane shapes are better matched than those of non-hurricane counterparts, as Fig. 10d represents a much better value (less than 1) compared to non-hurricanes.
(a) Probability of detection, (b) false alarm ratio, (c) critical success index, and (d) shape-match factor distributions at hurricane and non-hurricane storm objects categories. MRMS is the reference data field, and the perfect values are 1, 0, 1, and 0, respectively. The blue/orange color is related to non-hurricane/hurricane objects.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
d. Location-based hurricane versus non-hurricane characteristics
In addition to the previous results, which were derived using an object-based approach, some additional outputs from a location-based (or station) perspective are presented here. In this approach, an individual storm over a pixel is defined as the consecutive positive values in time. So, each storm over a pixel starts and ends with zero rain rate, and the length of the selected array is the storm duration from that pixel point of view. Note that all data are masked by the detected objects. Figure 11 compares the pixel-based storm duration distributions between the two products for storms lasting more than 10 h. Using the IMERG dataset during both hurricane and non-hurricane events, storms often last longer over a pixel. These results are consistent with hurricane and non-hurricane area distributions in the object-based approach.
Storm duration distributions from a pixel point of view during the (a) hurricane events and (b) non-hurricane events.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
Figure 12 depicts the normalized life cycle of storms lasting more than 10 h from the pixel point of view. The shaded area in this figure represents the uncertainty band (5%–95%) at each bin. This figure shows that non-hurricane storms reach their maximum intensity early in the storms’ life; however, it takes time for hurricanes to reach their maximum intensity in both datasets. MRMS hurricanes are more intense, on average, from a location point of view in comparison with hurricanes in IMERG (Fig. 12a). However, there exists the opposite in non-hurricane events (Fig. 12b). These results are consistent with hurricane and non-hurricane averaged intensity differences in the object-based approach (see Figs. 9c,f).
Averaged temporal rainfall as a function of normalized storm life cycle in 20 bins from a pixel point of view for storms lasting more than 10 h observed in MRMS/IMERG (red/blue colors). The shaded area is the (5%–95%) uncertainties at each bin.
Citation: Journal of Hydrometeorology 22, 1; 10.1175/JHM-D-20-0187.1
6. Discussion
Based on the results of analyses above, IMERG agrees reasonably well with MRMS in representing the object-based storm characteristics (Fig. 4b), and it can be used as a reliable source where ground radar data are not available. Note that this study is just focused on the eastern United States. So, the reliability of these results should be tested in other regions as well. Despite the observed similarities between the two datasets, there are some discrepancies in storm characteristics distributions that are statistically significant. For instance, IMERG mostly represents storm areas to be larger than does MRMS, which is also reported by Cui et al. (2020) in tracking MCSs over the United States. The top-down viewing angle of satellite sensors detects precipitation high in the atmosphere, a large proportion of which could evaporate before reaching the ground. Therefore, satellite precipitation products can overestimate the low–moderate precipitation intensity at the ground (Chen et al. 2020) since the low–moderate precipitation intensity is more frequent at higher altitudes. This might be more critical for IR sensors, providing cloud top temperature measurements that may show the storms having unrealistically larger areas due to false detection of nonprecipitable cold clouds (Kuligowski et al. 2016). Lower estimates of storm area by MRMS compared to satellite might be due to ground radars viewing a cross section of the storm well below the cloud tops. This viewing angle difference also plays a role in MRMS storms appearing more linear (Fig. 4f).
Storms having larger detected areas in IMERG than MRMS is consistent with Fig. 5 showing relatively good POD (around 0.6) and poor FAR values (often more than 0.5). Although storms appear larger in IMERG, MRMS represents the storms as having higher peak intensities (Fig. 4e), which may be caused by the aggregation of rain particles within the storms’ cores at lower altitudes leading to the formation of heavier droplets, falling faster, compared to the precipitation at high levels within the cloud (Xu and Zipser 2015). In addition, the higher spatiotemporal resolution of MRMS compared to IMERG also contributes to a better representation of the heavy precipitation rate in the storm cores. Since storms appear to have smaller areas with higher peak intensities in MRMS, they appear spatially more concentrated in comparison with their IMERG counterparts. Thus, employing MRMS for hydrological purposes will increase the level of simulated flood risk, especially from short-duration storms that are likely to be accompanied by more extreme events leading to flash floods in the future (Westra et al. 2014). Composite maps in Fig. 7 also indicate that hurricanes have more concentrated shapes in MRMS compared to IMERG, as the MRMS radial intensity in Fig. 7d is more concentrated around the storms’ centers. Figure 4c indicates that IMERG tends to represent the storms with higher volume of precipitation in comparison with MRMS, which may result in simulating higher risk for riverine floods in long-duration storms. This is in agreement with the findings of Tan et al. (2017) in which a positive normalized bias was observed by taking MRMS as the reference.
In addition, the observation of storms with smaller areas by MRMS could result in capturing storms with disconnected patterns (see the detected hurricane object at T = 2000 UTC 6 October 2016 in Fig. 2), which increases the number of storm-contributing objects and split/merging events in a storm lifetime compared to their IMERG counterparts (see Figs. 4i,j). Storm lifetimes are longer in MRMS compared to IMERG. One probable reason might be related to the higher sensitivity of ground radars compared to satellites in capturing very small-scale storms that might be missed by satellites with coarser resolution. Another thing might be the difficulty of satellite IR and PMW sensors to detect and estimate light rain and snowfall over land as shown by Behrangi et al. (2014). This is especially the case for warm rainfall as IR tends to assign precipitation rate to colder brightness temperatures and MW is more sensitive to scattering signal of ice over land. So precipitation from systems with little ice aloft might be missed.
Although there are strong similarities in resolving storm characteristics between the datasets, the similarity of both products was poor in capturing the storms’ shapes as shown by the high shape-match factor and low CSI values in Fig. 5. The POI values in Fig. 9 show that the rainfall volume and intensity differences between the two products grow in hurricanes in comparison with non-hurricane storms. However, the shape parameters like POD, FAR, CSI, and shape-match factor in Fig. 10 indicate that the hurricanes’ shape similarities are much better than those of non-hurricanes.
Figure 8 illustrates that some differences in IMERG and MRMS are related to the storm characteristics. For instance, MRMS mostly shows hurricanes as having higher averaged rain intensity than IMERG (Fig. 9f). However, the differences decrease (Fig. 8f) when the hurricanes have high translation speeds. According to Figs. 8g and 8i, hurricanes with larger area and volume of precipitation have a larger bias in maximum intensity. Also, hurricanes with higher peak intensity have a larger bias in storm areas. These relationships are derived from the relations between the hurricane characteristics shown in Fig. S3 in the supplemental material section, which shows similar relationships but with differing slopes.
An effort has been made to compare some pixel-based storm characteristics with the object-based ones to investigate the similarities and discrepancies between these two approaches. Figure 11 indicates that the pixel-based storm durations in MRMS are often shorter than storms in IMERG; however, from an object-based perspective, MRMS storms last longer. Pixel-based storm duration is related to the time it takes for a single storm object to pass over a pixel; therefore, the larger the storm object and the slower it moves, the longer precipitation falls over the pixel. However, in an object-based approach, storms are sets of storm-contributing objects that are splitting and merging. Thus, from a pixel-based perspective, MRMS storms have shorter duration but are occurring more frequently in comparison with the IMERG counterparts. (Freitas et al. 2020) have found similar results over Brazil where IMERG significantly overestimates the pixel-based storm duration compared to ground station records.
The normalized storm life cycles from a pixel point of view in Fig. 12 indicate that the majority of precipitation occurred at the beginning of the storms and the timing of peak intensity occurs at the early stages of the storm evolution. This result is in agreement with the findings of other studies over MCSs in the United States (Cui et al. 2020; O and Kirstetter 2018; Prein et al. 2017) and severe storms over Australia (Wasko and Sharma 2015). However, it takes longer for hurricanes to reach their mature stages, and most precipitation occurs close to the dissipation part of hurricane life cycles. In addition, MRMS observes higher precipitation rate in hurricanes compared to non-hurricane storms, which is consistent with the object-averaged intensity discrepancy seen in Figs. 9c and 9f.
This difference might be that MRMS better captures the hurricane’s core due to its higher resolution compared to IMERG. This discrepancy might alternatively be related to the higher optical saturation effect in capturing the hurricanes’ cores by both satellites and ground radars, which is also reported in other studies (Chen et al. 2020; Li et al. 2020; Omranian et al. 2018). MRMS is a merged product of different ground radars monitoring the hurricanes from multiple locations and different viewing angles; however, satellite sensors contributing in IMERG are observing the hurricanes only from above leading to different signal saturation properties of both products. There is also a possibility that the effect of droplet aggregation leading to heavier precipitation at lower altitudes is overcoming the evaporation effect in hurricanes, leading MRMS to show higher rainfall intensity over a larger part of the hurricanes compared to non-hurricanes. However, hurricanes are still smaller in MRMS due to the evaporation effect in outer marginal parts of the storms. These results are in agreement with a recent NASA report that compares IMERG and MRMS during hurricane Harvey over a selected domain in Texas, where the domain average precipitation rate indicate that IMERG underestimated precipitation in the storm core but overestimated it in the outer rainbands (Huffman et al. 2020).
7. Conclusions
This study was aimed at providing a more-advanced investigation of two high-resolution precipitation datasets: 1) Integrated Multisatellite Retrievals for Global Precipitation Measurement (IMERG) final product and 2) gauge-corrected ground radar-based Multi-Radar Multi-Sensor (MRMS) quantitative precipitation estimations. The datasets are compared in terms of their ability to represent storm characteristics over the United States using an object-based approach. A wide range of storms with different sizes from the small-scale storms to those with larger scale, in particular hurricanes, is studied during a 4-yr period (2015–19). The differences between the two datasets in capturing the storm properties were also investigated exclusively over seven landfalling hurricanes that occurred during the study period, and the results were compared with non-hurricane events. Finally, specific pixel-based storm characteristics were compared with object-based ones to evaluate the agreement of storm characteristics from the two perspectives.
Object-based storm characteristics as seen by IMERG resemble those as seen by MRMS; however, small discrepancies exist that are statistically significant. Storm objects appear spatially more concentrated in MRMS, with a smaller area and higher peak intensity compared to IMERG-detected storms. Thus, in rainfall runoff models, using this dataset may increase the estimated flood risk, particularly in short-duration storms. However, IMERG represents the storm as having a higher total volume of precipitation, so its application in hydrological simulations could result in a higher estimated risk of riverine flooding.
Several reasons could be contributing to the observed differences between MRMS and IMERG. These can include differences in sensors’ specifications and retrieval methods, temporal and spatial resolution differences, and the methods used to merge individual sensors and determine quality retrievals. Differences can also be partly related to different viewing angles of sensors contributing to the merged products. The vertical top-down viewing angle of the satellite sensors results in strong consideration of the storm-top precipitation in IMERG. Thus, the effect of light-precipitation evaporation at lower altitudes would result in smaller storm objects’ areas in MRMS. Despite smaller storm objects areas, MRMS storms consist of more contributing objects scattered throughout the storm lifetime. The misrepresentation of nonprecipitating cold clouds as rainfall in the IR algorithm is another reason for this discrepancy. In addition, the effect of rainfall aggregation at lower altitudes causes MRMS to represent storms with higher peak intensity. The higher resolution of MRMS also contributes to better capturing the heavy precipitation compared to IMERG.
The PDF overlap index (POI) values indicate that the similarity of the datasets deteriorates in terms of precipitation magnitude during hurricane events. However, the similarity of the storms’ shapes significantly improves during these events due to their strong signature throughout the atmospheric column.
The selected datasets have a better agreement in fast hurricanes in terms of averaged precipitation, as the object-averaged intensity is smaller in fast-moving hurricanes, and the differences between the datasets reduce when hurricane translation speed increases. Since MRMS represents storms with smaller areas and higher peak intensities, as the size of the hurricanes increases, the differences between the datasets also increase in terms of peak intensity and area.
Comparing the pixel-based and object-based approaches indicate that sometimes monitoring storms from different perspectives may lead to contradictory results. For instance, the pixel-based storm durations are often longer in IMERG compared to MRMS counterparts as the storm object areas are larger in this product, and it takes longer for the storm object to pass over a pixel. However, from an object-based perspective, MRMS storms last longer compared to IMERG counterparts since storms are defined as sets of storm objects merging/splitting during the storm lifetime than IMERG. However, in terms of precipitation magnitude, both approaches showed similar results. For instance, the normalized pixel-based life cycle of the storms indicates that MRMS represents a higher average intensity during the hurricane’s lifetime. However, in non-hurricanes this dataset shows the opposite, which is in agreement with the object-averaged intensity from an object-based perspective. The possible reasons for this result include 1) the higher resolution of MRMS leading to better representation of heavy precipitation, 2) the effect of saturation, which is less important in the MRMS product in capturing the hurricane core (due to multiple sensors using multiple viewing angles), and/or 3) water droplet aggregation at lower altitudes, which dominates over the effect of evaporation for a large proportion of the hurricanes. However, in non-hurricane events, which mostly consist of small- to moderate-sized storms, the evaporation effect is important over a large proportion of the storms and the effect of saturation is less important for capturing the storm core.
Despite some differences, the overall similarity in storm characteristics between MRMS and IMERG provides confidence that IMERG can be used to quantify storm characteristics, and their changes, in areas without ground-based radars.
Acknowledgments
This work was supported by the Australian Research Council as part of the Center of Excellence for Climate Extremes (CE170100023). A.B. was partly supported by NASA Weather and Atmospheric Dynamics (NNH19ZDA001N-ATDM) award.
Data availability statement
Data sharing is not applicable to this article as no new data were created or analyzed in this study. IMERG products were obtained from the Goddard Earth Sciences Data and Information Services Center (GES DISC) (https://disc.gsfc.nasa.gov/). MRMS dataset is also available to the public from the Iowa Environmental Mesonet (IEM) (https://mtarchive.geol.iastate.edu/).
REFERENCES
Ayat, H., M. Reza Kavianpour, S. Moazami, Y. Hong, and E. Ghaemi, 2018: Calibration of weather radar using region probability matching method (RPMM). Theor. Appl. Climatol., 134, 165–176, https://doi.org/10.1007/s00704-017-2266-7.
Beck, H. E., and Coauthors, 2019: Daily evaluation of 26 precipitation datasets using Stage-IV gauge-radar data for the CONUS. Hydrol. Earth Syst. Sci., 23, 207–224, https://doi.org/10.5194/hess-23-207-2019.
Behrangi, A., K. Hsu, B. Imam, S. Sorooshian, G. J. Huffman, and R. J. Kuligowski, 2009a: PERSIANN-MSA: A precipitation estimation method from satellite-based multispectral analysis. J. Hydrometeor., 10, 1414–1429, https://doi.org/10.1175/2009JHM1139.1.
Behrangi, A., K. Hsu, B. Imam, S. Sorooshian, and R. J. Kuligowski, 2009b: Evaluating the utility of multispectral information in delineating the areal extent of precipitation. J. Hydrometeor., 10, 684–700, https://doi.org/10.1175/2009JHM1077.1.
Behrangi, A., Y. Tian, B. H. Lambrigtsen, and G. L. Stephens, 2014: What does CloudSat reveal about global land precipitation detection by other spaceborne sensors? Water Resour. Res., 50, 4893–4905, https://doi.org/10.1002/2013WR014566.
Behrangi, A., K. J. Bormann, and T. H. Painter, 2018: Using the airborne snow observatory to assess remotely sensed snowfall products in the California Sierra Nevada. Water Resour. Res., 54, 7331–7346, https://doi.org/10.1029/2018WR023108.
Berg, R., 2017: Hurricane Hermine. National Hurricane Center, 63 pp., https://www.nhc.noaa.gov/data/tcr/AL092016_Hermine.pdf.
Beven, J. L., II, and R. Berg, 2018: Hurricane Nate. National Hurricane Center, 415 pp., https://www.nhc.noaa.gov/data/tcr/AL162017_Nate.pdf.
Beven, J. L., II, R. Berg, and A. Hagen, 2019: Hurricane Michael. National Hurricane Center, 86 pp., https://www.nhc.noaa.gov/data/tcr/AL142018_Michael.pdf.
Blake, E. S., and D. A. Zelinsky, 2018: Hurricane Harvey. National Hurricane Center, 77 pp., https://www.nhc.noaa.gov/data/tcr/AL092017_Harvey.pdf.
Bullock, R., B. Brown, and T. Fowler, 2016: Method for object-based diagnostic evaluation. NCAR Tech. Note NCAR/TN-532+STR, National Center for Atmospheric Research, 84 pp., https://doi.org/10.5065/D61V5CBS.
Bytheway, J. L., and C. D. Kummerow, 2015: Toward an object-based assessment of high-resolution forecasts of long-lived convective precipitation in the central U.S. J. Adv. Model. Earth Syst., 7, 1248–1264, https://doi.org/10.1002/2015MS000497.
Cangialosi, J. P., A. S. Latto, and R. Berg, 2018: Hurricane Irma. National Hurricane Center, 111 pp., https://www.nhc.noaa.gov/data/tcr/AL112017_Irma.pdf.
Chen, M., S. Nabih, N. S. Brauer, S. Gao, J. J. Gourley, Z. Hong, R. L. Kolar, and Y. Hong, 2020: Can remote sensing technologies capture the extreme precipitation event and its cascading hydrological response? A case study of Hurricane Harvey using EF5 modeling framework. Remote Sens., 12, 445, https://doi.org/10.3390/rs12030445.
Clark, A. J., R. G. Bullock, T. L. Jensen, M. Xue, and F. Kong, 2014: Application of object-based time-domain diagnostics for tracking precipitation systems in convection-allowing models. Wea. Forecasting, 29, 517–542 https://doi.org/10.1175/WAF-D-13-00098.1.
Cui, W., X. Dong, B. Xi, Z. Feng, and J. Fan, 2020: Can the GPM IMERG final product accurately represent MCSs’ precipitation characteristics over the central and eastern United States? J. Hydrometeor., 21, 39–57, https://doi.org/10.1175/JHM-D-19-0123.1.
Davis, C. A., B. G. Brown, R. Bullock, and J. Halley-Gotway, 2009: The Method for Object-Based Diagnostic Evaluation (MODE) applied to numerical forecasts from the 2005 NSSL/SPC spring program. Wea. Forecasting, 24, 1252–1267, https://doi.org/10.1175/2009WAF2222241.1.
Derin, Y., and K. K. Yilmaz, 2014: Evaluation of multiple satellite-based precipitation products over complex topography. J. Hydrometeor., 15, 1498–1516, https://doi.org/10.1175/JHM-D-13-0191.1.
Derin, Y., E. Anagnostou, M. N. Anagnostou, J. Kalogiros, D. Casella, A. C. Marra, G. Panegrossi, and P. Sano, 2018: Passive microwave rainfall error analysis using high-resolution X-band dual-polarization radar observations in complex terrain. IEEE Trans. Geosci. Remote Sens., 56, 2565–2586, https://doi.org/10.1109/TGRS.2017.2763622.
Derin, Y., and Coauthors, 2019: Evaluation of GPM-era global satellite precipitation products over multiple complex terrain regions. Remote Sens., 11, 2936, https://doi.org/10.3390/rs11242936.
Dowdy, A. J., and Coauthors, 2019: Review of Australian east coast low pressure systems and associated extremes. Climate Dyn., 53, 4887–4910, https://doi.org/10.1007/s00382-019-04836-8.
ElSaadani, M., W. F. Krajewski, and D. L. Zimmerman, 2018: River network based characterization of errors in remotely sensed rainfall products in hydrological applications. Remote Sens. Lett., 9, 743–752, https://doi.org/10.1080/2150704X.2018.1475768.
Freitas, E. da S., and Coauthors, 2020: The performance of the IMERG satellite-based product in identifying sub-daily rainfall events and their properties. J. Hydrol., 589, 125128, https://doi.org/10.1016/j.jhydrol.2020.125128.
Furl, C., D. Ghebreyesus, and H. Sharif, 2018: Assessment of the performance of satellite-based precipitation products for flood events across diverse spatial scales using GSSHA modeling system. Geosciences, 8, 191, https://doi.org/10.3390/geosciences8060191.
Gaona, M. F. R., A. Overeem, H. Leijnse, and R. Uijlenhoet, 2016: First-year evaluation of GPM rainfall over the Netherlands: IMERG day 1 final run (V03D). J. Hydrometeor., 17, 2799–2814, https://doi.org/10.1175/JHM-D-16-0087.1.
Gebregiorgis, A. S., P.-E. Kirstetter, Y. E. Hong, N. J. Carr, J. J. Gourley, W. Petersen, and Y. Zheng, 2017: Understanding overland multisensor satellite precipitation error in TMPA-RT products. J. Hydrometeor., 18, 285–306, https://doi.org/10.1175/JHM-D-15-0207.1.
Gebregiorgis, A. S., P. Kirstetter, Y. E. Hong, J. J. Gourley, G. J. Huffman, W. A. Petersen, X. Xue, and M. R. Schwaller, 2018: To what extent is the day 1 GPM IMERG satellite precipitation estimate improved as compared to TRMM TMPA-RT? J. Geophys. Res. Atmos., 123, 1694–1707, https://doi.org/10.1002/2017JD027606.
Ghaemi, E., M. Kavianpour, S. Moazami, Y. Hong, and H. Ayat, 2017: Uncertainty analysis of radar rainfall estimates over two different climates in Iran. Int. J. Remote Sens., 38, 5106–5126, https://doi.org/10.1080/01431161.2017.1335909.
Guo, H., S. Chen, A. Bao, A. Behrangi, Y. Hong, F. Ndayisaba, J. Hu, and P. M. Stepanian, 2016: Early assessment of integrated multi-satellite retrievals for global precipitation measurement over China. Atmos. Res., 176–177, 121–133, https://doi.org/10.1016/j.atmosres.2016.02.020.
Hong, Y., K.-L. Hsu, S. Sorooshian, and X. Gao, 2004: Precipitation estimation from remotely sensed imagery using an artificial neural network cloud classification system. J. Appl. Meteor., 43, 1834–1853, https://doi.org/10.1175/JAM2173.1.
Hsu, K.-L., X. Gao, S. Sorooshian, and H. V. Gupta, 1996: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176–1190, https://doi.org/10.1175/1520-0450(1997)036<1176:PEFRSI>2.0.CO;2.
Hu, M.-K., 1962: Visual pattern recognition by moment invariants. IRE Trans. Inf. Theory, 49, 179–187.
Huffman, G. J., and Coauthors, 2019: NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG). Algorithm Theoretical Basis Doc., version 06, 32 pp., https://pmm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V06.pdf.
Huffman, G. J., and Coauthors, 2020: Early results for version 06 IMERG. NASA, 20 pp., https://ntrs.nasa.gov/citations/20190029175.
Hunter, S. M., 1996: WSR-88D radar rainfall estimation: capabilities, limitations and potential improvements. Natl. Wea. Dig., 20, 26–381.
Khan, S., V. Maggioni, and P.-E. Kirstetter, 2018: Investigating the potential of using satellite-based precipitation radars as reference for evaluating multisatellite merged products. J. Geophys. Res. Atmos., 123, 8646–8660, https://doi.org/10.1029/2018JD028584.
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 P. M. Jackson, 1993: Weather radar performance at long range-simulated and observed. J. Appl. Meteor., 32, 975–985, https://doi.org/10.1175/1520-0450(1993)032<0975:WRPALR>2.0.CO;2.
Krajewski, W. F., and J. A. Smith, 2002: Radar hydrology: Rainfall estimation. Adv. Water Resour., 25, 1387–1394, https://doi.org/10.1016/S0309-1708(02)00062-3.
Kuligowski, R. J., Y. Li, Y. Hao, and Y. Zhang, 2016: Improvements to the GOES-R rainfall rate algorithm. J. Hydrometeor., 17, 1693–1704, https://doi.org/10.1175/JHM-D-15-0186.1.
Langston, C., J. Zhang, and K. Howard, 2007: Four-dimensional dynamic radar mosaic. J. Atmos. Oceanic Technol., 24, 776–790, https://doi.org/10.1175/JTECH2001.1.
Lengfeld, K., P.-E. Kirstetter, H. J. Fowler, J. Yu, A. Becker, Z. Flamig, and J. J. Gourley, 2020: Use of radar data for characterizing extreme precipitation at fine scales and short durations. Environ. Res. Lett., 15, 085003, https://doi.org/10.1088/1748-9326/ab98b4.
Li, J., K.-L. Hsu, A. AghaKouchak, and S. Sorooshian, 2015: An object-based approach for verification of precipitation estimation. Int. J. Remote Sens., 36, 513–529, https://doi.org/10.1080/01431161.2014.999170.
Li, J., K.-L. Hsu, A. AghaKouchak, and S. Sorooshian, 2016: Object-based assessment of satellite precipitation products. Remote Sens., 8, 547, https://doi.org/10.3390/rs8070547.
Li, J., C. Wasko, F. Johnson, J. P. Evans, and A. Sharma, 2018: Can regional climate modeling capture the observed changes in spatial organization of extreme storms at higher temperatures? Geophys. Res. Lett., 45, 4475–4484, https://doi.org/10.1029/2018GL077716.
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.
Li, Z., D. Yang, and Y. Hong, 2013: Multi-scale evaluation of high-resolution multi-sensor blended global precipitation products over the Yangtze River. J. Hydrol., 500, 157–169, https://doi.org/10.1016/j.jhydrol.2013.07.023.
Li, Z., M. Chen, S. Gao, Z. Hong, G. Tang, Y. Wen, J. J. Gourley, and Y. Hong, 2020: Cross-examination of similarity, difference and deficiency of gauge, radar and satellite precipitation measuring uncertainties for extreme events using conventional metrics and multiplicative triple collocation. Remote Sens., 12, 1258, https://doi.org/10.3390/rs12081258.
Moazami, S., S. Golian, M. R. Kavianpour, and Y. Hong, 2013: Comparison of PERSIANN and V7 TRMM Multi-satellite Precipitation Analysis (TMPA) products with rain gauge data over Iran. Int. J. Remote Sens., 34, 8156–8171, https://doi.org/10.1080/01431161.2013.833360.
Moazami, S., S. Golian, M. R. Kavianpour, and Y. Hong, 2014: Uncertainty analysis of bias from satellite rainfall estimates using copula method. Atmos. Res., 137, 145–166, https://doi.org/10.1016/j.atmosres.2013.08.016.
Moazami, S., S. Golian, Y. Hong, C. Sheng, and M. R. Kavianpour, 2016: Comprehensive evaluation of four high-resolution satellite precipitation products under diverse climate conditions in Iran. Hydrol. Sci. J., 61, 420–440, https://doi.org/10.1080/02626667.2014.987675.
Mohammed, S. A., M. A. Hamouda, M. T. Mahmoud, and M. M. Mohamed, 2020: Performance of GPM-IMERG precipitation products under diverse topographical features and multiple-intensity rainfall in an arid region. Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2019-547.
Nelson, B. R., O. P. Prat, D.-J. Seo, and E. Habib, 2016: Assessment and implications of NCEP Stage IV quantitative precipitation estimates for product intercomparisons. Wea. Forecasting, 31, 371–394, https://doi.org/10.1175/WAF-D-14-00112.1.
O, S., and P. Kirstetter, 2018: Evaluation of diurnal variation of GPM IMERG-derived summer precipitation over the contiguous US using MRMS data. Quart. J. Roy. Meteor. Soc., 144, 270–281, https://doi.org/10.1002/qj.3218.
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.
Omranian, E., H. Sharif, and A. Tavakoly, 2018: How well can Global Precipitation Measurement (GPM) capture hurricanes? Case study: Hurricane Harvey. Remote Sens., 10, 1150, https://doi.org/10.3390/rs10071150.
OpenCV, 2020: Structural analysis and shape descriptors. OpenCV, https://docs.opencv.org/3.4/d3/dc0/group__imgproc__shape.html#gaf2b97a230b51856d09a2d934b78c015f.
Panahi, M., and A. Behrangi, 2020: Comparative analysis of snowfall accumulation and gauge undercatch correction factors from diverse data sets: In situ, satellite, and reanalysis. Asia-Pac. J. Atmos. Sci., 56, 615–628, https://doi.org/10.1007/s13143-019-00161-6.
Pepler, A. S., A. Di Luca, and J. P. Evans, 2018: Independently assessing the representation of midlatitude cyclones in high-resolution reanalyses using satellite observed winds. Int. J. Climatol., 38, 1314–1327, https://doi.org/10.1002/joc.5245.
Petković, V., and C. D. Kummerow, 2017: Understanding the sources of satellite passive microwave rainfall retrieval systematic errors over land. J. Appl. Meteor. Climatol., 56, 597–614, https://doi.org/10.1175/JAMC-D-16-0174.1.
Porcú, F., M. Borga, and F. Prodi, 1999: Rainfall estimation by combining radar and infrared satellite data for nowcasting purposes. Meteor. Appl., 6, 289–300, https://doi.org/10.1017/S1350482799001243.
Prat, O. P., and B. R. Nelson, 2015: Evaluation of precipitation estimates over CONUS derived from satellite, radar, and rain gauge data sets at daily to annual scales (2002–2012). Hydrol. Earth Syst. Sci., 19, 2037–2056, https://doi.org/10.5194/hess-19-2037-2015.
Prein, A. F., and A. Gobiet, 2017: Impacts of uncertainties in European gridded precipitation observations on regional climate analysis. Int. J. Climatol., 37, 305–327, https://doi.org/10.1002/joc.4706.
Prein, A. F., and Coauthors, 2013: Added value of convection permitting seasonal simulations. Climate Dyn., 41, 2655–2677, https://doi.org/10.1007/s00382-013-1744-6.
Prein, A. F., and Coauthors, 2017: Simulating North American mesoscale convective systems with a convection-permitting climate model. Climate Dyn., 55, 95–110, https://doi.org/10.1007/s00382-017-3993-2.
Rios Gaona, M. F., A. Overeem, A. M. Brasjen, J. F. Meirink, H. Leijnse, and R. Uijlenhoet, 2017: Evaluation of rainfall products derived from satellites and microwave links for the Netherlands. IEEE Trans. Geosci. Remote Sens., 55, 6849–6859, https://doi.org/10.1109/TGRS.2017.2735439.
Roca, R., L. V. Alexander, G. Potter, M. Bador, R. Jucá, S. Contractor, M. G. Bosilovich, and S. Cloché, 2019: FROGS: a daily 1◦ × 1◦ gridded precipitation database of rain gauge, satellite and reanalysis products. Earth Syst. Sci. Data, 11, 1017–1035, https://doi.org/10.5194/ESSD-11-1017-2019.
Sadeghi, L., B. Saghafian, and S. Moazami, 2019: Evaluation of IMERG and MRMS remotely sensed snowfall products. Int. J. Remote Sens., 40, 4175–4192, https://doi.org/10.1080/01431161.2018.1562259.
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., W. F. Krajewski, A. Kruger, P. Domaszczynski, J. A. Smith, and M. Steiner, 2011: Radar-rainfall estimation algorithms of Hydro-NEXRAD. J. Hydroinf., 13, 277–291, https://doi.org/10.2166/hydro.2010.003.
Seo, B.-C., W. F. Krajewski, and J. A. Smith, 2014: Four-dimensional reflectivity data comparison between two ground-based radars: Methodology and statistical analysis. Hydrol. Sci. J., 59, 1320–1334, https://doi.org/10.1080/02626667.2013.839872.
Sharifi, E., R. Steinacker, and B. Saghafian, 2016: Assessment of GPM-IMERG and other precipitation products against gauge data under different topographic and climatic conditions in Iran: Preliminary results. Remote Sens., 8, 135, https://doi.org/10.3390/rs8020135.
Shige, S., S. Kida, H. Ashiwake, T. Kubota, and K. Aonashi, 2013: Improvement of TMI rain retrievals in mountainous areas. J. Appl. Meteor. Climatol., 52, 242–254, https://doi.org/10.1175/JAMC-D-12-074.1.
Smalley, M., T. L’Ecuyer, M. Lebsock, and J. Haynes, 2014: A comparison of precipitation occurrence from the NCEP stage IV QPE product and the CloudSat cloud profiling radar. J. Hydrometeor., 15, 444–458, https://doi.org/10.1175/JHM-D-13-048.1.
Smith, J. A., D. J. Seo, M. L. Baeck, and M. D. Hudlow, 1996: An intercomparison study of NEXRAD precipitation estimates. Water Resour. Res., 32, 2035–2045, https://doi.org/10.1029/96WR00270.
Stewart, S. R., 2017: Hurricane Matthew. National Hurricane Center, 96 pp., https://www.nhc.noaa.gov/data/tcr/AL142016_Matthew.pdf.
Stewart, S. R., and R. Berg, 2019: Hurricane Florence. National Hurricane Center, 98 pp., https://www.nhc.noaa.gov/data/tcr/AL062018_Florence.pdf.
Tan, J., W. A. Petersen, P.-E. Kirstetter, and Y. Tian, 2017: Performance of IMERG as a function of spatiotemporal scale. J. Hydrometeor., 18, 307–319, https://doi.org/10.1175/JHM-D-16-0174.1.
Tang, L., Y. Tian, and X. Lin, 2014: Validation of precipitation retrievals over land from satellite-based passive microwave sensors. J. Geophys. Res. Atmospheres, 119, 4546–4567, https://doi.org/10.1002/2013JD020933.
Tang, S., R. Li, J. He, H. Wang, X. Fan, and S. Yao, 2020: Comparative evaluation of the GPM IMERG early, late, and final hourly precipitation products using the CMPA data over Sichuan Basin of China. Water, 12, 554, https://doi.org/10.3390/w12020554.
Wasko, C., and A. Sharma, 2015: Steeper temporal distribution of rain intensity at higher temperatures within Australian storms. Nat. Geosci., 8, 527–529, https://doi.org/10.1038/ngeo2456.
Wen, S., 2016: Sensitivity of model verification results to object identification parameters using method for object-based diagnostic evaluation–Time domain. NCAR/UCAR, 26 pp., https://n2t.net/ark:/85065/d7xw4mh5.
Wen, Y., A. Behrangi, H. Chen, and B. Lambrigtsen, 2018: How well were the early 2017 California Atmospheric River precipitation events captured by satellite products and ground-based radars? Quart. J. Roy. Meteor. Soc., 144, 344–359, https://doi.org/10.1002/qj.3253.
Westra, S., and Coauthors, 2014: Future changes to the intensity and frequency of short-duration extreme rainfall. Rev. Geophys., 52, 522–555, https://doi.org/10.1002/2014RG000464.
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.
Xu, W., and E. J. Zipser, 2015: Convective intensity, vertical precipitation structures, and microphysics of two contrasting convective regimes during the 2008 TiMREX. J. Geophys. Res. Atmos., 120, 4000–4016, https://doi.org/10.1002/2014JD022927.
Young, C. B., A. A. Bradley, W. F. Krajewski, A. Kruger, and M. L. Morrissey, 2000: Evaluating NEXRAD multisensor precipitation estimates for operational hydrologic forecasting. J. Hydrometeor, 1, 241–254, https://doi.org/10.1175/1525-7541(2000)001<0241:ENMPEF>2.0.CO;2.
Zan, B., Y. Yu, J. Li, G. Zhao, T. Zhang, and J. Ge, 2019: Solving the storm split-merge problem—A combined storm identification, tracking algorithm. Atmos. Res., 218, 335–346, https://doi.org/10.1016/j.atmosres.2018.12.007.
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, A., L. Xiao, C. Min, S. Chen, M. Kulie, C. Huang, and Z. Liang, 2019: Evaluation of latest GPM-Era high-resolution satellite precipitation products during the May 2017 Guangdong extreme rainfall event. Atmos. Res., 216, 76–85, https://doi.org/10.1016/j.atmosres.2018.09.018.
Zhang, J., L.-F. Lin, and R. L. Bras, 2018: Evaluation of the quality of precipitation products: A case study using WRF and IMERG data over the central United States. J. Hydrometeor., 19, 2007–2020, https://doi.org/10.1175/JHM-D-18-0153.1.
Zhang, X., E. Anagnostou, and C. Schwartz, 2018: NWP-based adjustment of IMERG precipitation for flood-inducing complex terrain storms: Evaluation over CONUS. Remote Sens., 10, 642, https://doi.org/10.3390/rs10040642.
Zick, S. E., and C. J. Matyas, 2016: A shape metric methodology for studying the evolving geometries of synoptic-scale precipitation patterns in tropical cyclones. Ann. Amer. Assoc. Geogr., 106, 1217–1235, https://doi.org/10.1080/24694452.2016.1206460.
