Merged and Gridded GPM and Atmospheric River Data Product

The Global Precipitation Measurement (GPM) Mission Core Observatory satellite launched in 2014 as a joint mission between National Aeronautics and Space Administration (NASA) and JAXA. Global Precipitation Measurement (GPM) has, since that time, provided continuous, valuable dual‐frequency radar and passive microwave radiometer observations. Here, we introduce a gridded data set of collocated GPM Core Observatory observational products merged with a reanalysis‐derived Atmospheric river (AR) data set in the North Atlantic and North Pacific sectors. The three data sets that are merged and gridded are: (a) the NASA Goddard Profiling (GPROF) precipitation product, which uses GPM passive microwave radiometer observations to derive surface precipitation rates, (b) a water vapor data product derived from the GPM Core Observatory radiometer, provided by Remote Sensing Systems (RSS), and (c) the Mattingly et al. (2018, https://doi.org/10.1029/2018jd028714) AR data set that is specifically tuned to the high‐latitude regions. This novel merged data set spans from May 2014 to December 2022 with plans to update annually through 2026 at minimum. This gridded product combines RSS passive water vapor and precipitation estimates with coincident AR detection. This data product benefits the scientific community by providing (a) user‐friendly gridded satellite data compared to standard satellite data sets, while maintaining high temporal resolution, and (b) coincident satellite observations to assess the link between ARs and precipitation.


Introduction
The transport of water vapor through the atmosphere is tied to the intensity and phase of precipitation around the world, and more recently has been used in forecasting the severity of precipitation events (Baggett et al., 2017;Ralph et al., 2020).Atmospheric rivers (ARs) are long, narrow features of relatively high integrated water vapor transport (IVT; atmospheric water vapor flux through a vertical column) that evolve with atmospheric circulation (Gimeno et al., 2014;Guan & Waliser, 2015;Ralph et al., 2018;Zhu & Newell, 1998).Recent work has investigated the links between model-derived ARs and precipitation over land and coastal topography (e.g., Cann & Friedrich, 2020;Guan et al., 2010;Mateling et al., 2021;Prein & Heymsfield, 2020;Slinskey et al., 2020).Few studies of ARs include satellite data due, in part, to AR algorithms relying on the complete spatio-temporal coverage of gridded model data sets, in contrast with the often-inconsistent spatio-temporal data observed by satellites (e.g., Ralph et al., 2004).Recently, spaceborne radar retrievals from the Global Precipitation Measurement (GPM) Core Observatory have illustrated the use of satellite remote sensing observations to study AR impacts on precipitation at select locations (e.g., the western U.S. coast; Cannon et al., 2017Cannon et al., , 2020)), but these studies have generally focused on the mid-latitude regions.Here, we provide a merged data set of GPM Core Observatory precipitation and water vapor products with ARs in the North Atlantic (NA) and North Pacific (NP) sectors.The goal of this product is to encourage further exploration of high-latitude ARs and their impact on precipitation estimates from spaceborne microwave radiometer observations.AR data products, derived from model and reanalysis data, generally vary in criteria for IVT magnitude, anomalies, and geometry in identifying ARs (Shields et al., 2018).Recent studies show that the magnitude and size of ARs are correlated to the intensity, phase, and location of precipitation around the world (Guan & Waliser, 2015;Ma et al., 2020;Mattingly et al., 2016).More specifically, the incidence of ARs leads to higher precipitation rates (Cann & Friedrich, 2020;Guan et al., 2010;Lavers & Villarini, 2013), and increased likelihood of precipitation being liquid versus frozen (Goldenson et al., 2018;Guan et al., 2016;Mateling et al., 2021).In Mattingly et al. (2018), the authors created an algorithm that identifies ARs in relatively cold and dry atmospheric conditions (high latitudes).Researchers use this Mattingly et al. (2018) AR scheme (hereafter, "M18") to study AR impacts on Greenland Ice Sheet surface mass balance (M18; Mattingly et al., 2020Mattingly et al., , 2023) ) and Upper Great Lakes precipitation (Mateling et al., 2021).
The GPM Core Observatory (GPM-CO) is a low-earth orbiting satellite that is part of a joint mission by National Aeronautics and Space Administration NASA and JAXA (Japanese Aerospace and Exploration Agency) launched in March 2014 and has since provided valuable data from its onboard dual-frequency radar and passive microwave radiometer (Hou et al., 2014;Skofronick-Jackson et al., 2017, 2018).The GPM Microwave Imager (GMI) has 13 passive microwave channels (10-183 GHz) and obtains brightness temperatures at the top of the atmosphere (Draper et al., 2015).The Goddard Profiling Algorithm (GPROF) is the operational passive microwave algorithm for GMI and retrieves surface precipitation rate and phase (Kummerow et al., 2015;Randel et al., 2020).Also using GMI observations Remote Sensing Systems (RSS) created a daily integrated atmospheric water vapor product (as well as other data; Meissner et al., 2012).The incidence of M18 ARs, defined by IVT, and GMIretrieved water vapor can be used to diagnose or evaluate methods used to identify the location, shape, and size of ARs.Further, coincident GPROF precipitation estimates provide insight into how the presence of highlatitude ARs impact observed precipitation rates and phase.
Here we present a gridded and merged data set that includes the M18 ARs, precipitation estimates from GMI GPROF, and the GMI-derived atmospheric water vapor.This data set is intended to be a comprehensive, satellitebased tool for the broader scientific community with the aim of investigating ARs and high-latitude precipitation over the NA and Pacific regions.In Section 2, we detail the GMI data products and the AR scheme.Section 3 outlines the gridding and merging process for the combined data set.Section 4 illustrates examples from the data set during boreal summer (June-July-August, JJA) and winter (December-January-February, DJF).In Section 5, we demonstrate an application of the merged data set.

Data Sets Used in Merged Product
The three distinct data sets merged are the GPM GMI GPROF VERSION 07 (V7) precipitation product, the RSS GMI-derived atmospheric water vapor product, and high-latitude ARs derived by Mattingly et al. (2018) using the Modern-Era Retrospective analysis for Research and Applications Reanalysis, Version 2 (MERRA-2; Gelaro et al., 2017).We grid the spatio-temporally irregular GMI GPROF V7 to a common 0.25°× 0.25°grid for the merged product to match the already-gridded RSS water vapor product.For this initial version of the database, we opted to use the NA boundaries of: 45-70°N, 70°W-10°E, and the NP boundaries of: 45-70°N, 140°E-120°W.Both sectors are latitudinally limited by the GPM-CO orbital inclination, 65°N, but we extended the domain up to 70°N as GMI observes brightness temperatures at latitudes up to ∼69°N.The database covers the time period from May 2014 to December 2022, beginning shortly after the GPM-CO launch date in March 2014, and will be updated annually.Files are provided for each basin and each month in network Common Data Form (NetCDF-4) file format.See Table A1 in Appendix A for a full list of each netCDF-4 file's contents.
Two of the three data sets in this merged product are derived from the GPM-CO microwave radiometer (GMI).Though the GPROF algorithm is applied to several other spaceborne microwave radiometer observations, GMI is the best calibrated.This multi-channel dual-polarization radiometer has a swath width of 885 km, observing at higher latitudes (∼|69|°) than the limit of the GPM-CO orbit (|65|°).Additionally, the GPM-CO has a high returnrate at high latitudes (∼|60|°) due to orbital inclination.

GPROF Data
The GMI GPROF V07 precipitation product is provided by the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC; Kummerow, 2022).The GPROF algorithm is a Bayesian scheme applied to GMI observations to estimate precipitation rate (Kummerow et al., 2015;Randel et al., 2020).The GMI-only version of GPROF (GPM_2AGPROFGPMGMI) included in the merged product uses the observed brightness temperature vector from GMI to search a pre-existing, or a priori, database containing 1 year of matched GPM-CO observations with atmospheric information.This Bayesian scheme then retrieves a hydrometeor profile used to estimate precipitation rate.This database was created using the GPM Combined Algorithm (Grecu et al., 2016) which retrieves the optimal precipitation profile based upon the dual-frequency radar reflectivity values (from the GPM Dual-frequency Precipitation Radar, or DPR) and multi-frequency brightness temperatures.
The nearrealtime GPROF algorithm searches the a priori database using the observed GMI brightness temperature vector and is constrained using Japanese Global Analysis (GANAL) two-m temperature (T2M), total precipitable water (TPW), and surface type (Randel et al., 2020).GPROF determines precipitation phase using ancillary surface wet bulb temperature and pre-defined lookup tables (Sims & Liu, 2015) which allows separation of liquid (rain) and frozen precipitation rates.We include both total and frozen precipitation rates in this merged data set.GMI GPROF underperforms when determining light or frozen precipitation due to the lack of DPR reflectivities ≤12 dBZ within the a priori database (Skofronick-Jackson et al., 2019).For a full explanation of the GPROF V7 algorithm, see Randel et al. (2020) or the Algorithm Theoretical Basis Document (ATBD; Passive Microwave Algorithm Team Facility, 2022).Variables obtained and gridded from the GMI GPROF V7 product include surface precipitation rates (liquid plus frozen precipitation), frozen precipitation rates, T2M, and the surface type flag (land, ocean, mixed).The GPROF product footprint resolution is 18.1 km × 10.9 km and defined at the center of the GMI footprint.All precipitation rates presented are liquid water equivalent (LWE).

RSS Water Vapor Data
We additionally merge the atmospheric water vapor from GMI observations created by the Remote Sensing Systems (RSS) group.To derive this product, GMI passive microwave data (obtained through GES DISC) are back-processed and put through the RSS Version-8.2algorithm to derive brightness temperature.The RSS radiative transfer model then uses these brightness temperatures to retrieve the total column water vapor (Meissner et al., 2012;Wentz et al., 2015).We downloaded daily averages of water vapor from May 2014 to December 2022 at 0.25°× 0.25°spatial resolution from the RSS website (see data availability section).Provided are two files per day, for ascending and descending GPM-CO orbital tracks, with time steps given in fractional hours per day.The temporal resolution is either a daily average of water vapor or as the most recent value of water vapor, overwriting derived water vapor from a preceding overpass on the same day.The water vapor is not computed over land or near coastlines as these surface types can complicate the GMI observations and create large uncertainties (Meissner et al., 2012).More detailed information on the derivation of water vapor from GMI data can be found in the product's ATBD (Meissner et al., 2012).

Atmospheric Rivers
The final component of the merged data set is the inclusion of ARs as defined in M18.M18 uses 6-hourly MERRA-2 wind and specific humidity data between 1,000 and 200 hPa to compute integrated water vapor transport (IVT) in kg m 1 s 1 , which is then re-gridded to 0.5°× 0.5°.IVT is composed of the total water vapor summed throughout the column multiplied by the winds that are transporting it: where g is gravitational acceleration, q is specific humidity, V is vector wind, and dp is the difference between pressure levels.The RSS water vapor data directly relates to the moisture component of the IVT equation but is from GMI rather than the MERRA-2 reanalysis that is used to calculate IVT for AR identification.An AR is identified when IVT exceeds both a minimum absolute threshold of 150 kg m 1 s 1 and the 85th percentile of the grid-cell-specific 31-day centered climatology, with potential AR objects further filtered by geometric criteria: (a) Earth and Space Science 10.1029/2023EA003333 length of 1,500 km and (b) a length-to-width ratio of 1.5 (Mattingly, 2018).This AR data set was developed with the intention of identifying the size and location of ARs in the high-latitude regions where atmospheric moisture is relatively low, requiring a more sensitive detection threshold of IVT than many other AR algorithms (e.g., Rutz et al., 2019).Additionally, if the AR object center is poleward of |70|°, the algorithm does not require the IVT direction to be poleward like most other AR algorithms.This may slightly increase the number of GPM-AR intersections at the northern edge of the study domain, particularly during summer (JJA).

Methodology
Merging the GPM-CO data products, which have irregular, but high-resolution spatial and temporal coverage, with a spatially-gridded, 6-hourly temporal resolution M18 AR data set requires additional steps to best match the respective product information.We chose a 0.25°× 0.25°(latitude × longitude) spatial grid for the merged data set to match the already-gridded RSS water vapor product and to best maintain the high spatial resolution of the GMI GPROF product.See Table 1 for spatial and temporal resolution of these data products and the merged data set.The merged data are output into monthly files where each time step represents a GPM-CO overpass through the predefined (NA or Pacific) basin.The data structure is 3-dimensional (time step by latitude by longitude), with each monthly file containing ∼250 time steps.
To grid GMI GPROF data, we pre-define a latitude-by-longitude grid where each gridbox is defined by its lower left corner latitude and longitude.We apply the GMI GPROF product quality flags (L1CqualityFlag = 0, qualityFlag = 0, pixelStatus = 0) that are often applied over land due to radio frequency interference.For each individual GPM-CO overpass file, we aggregate all GMI GPROF precipitation observations that fall within each 0.25°× 0.25°gridbox.Each gridbox contains the conditional mean of precipitation rates ≥0.01 mm hr 1 LWE.We identify the start and end times of each overpass through the basin and find the median time; this is then assigned as the date and time for the gridded overpass.The number of GMI footprints that fall within each grid box while gridding the data is saved as "count" variables to provide sampling data to users.We create a gridded surface flag variable from the footprint-resolution GMI GPROF V7 surface flag by setting grid boxes with 100% ocean footprints as "0" (or ocean surface).Grid boxes that contain footprints flagged as 100% land are assigned a flag of "2" (or land surface).Any grid boxes that contain a combination of ocean and other surface types (which includes sea ice and coastlines) are flagged as "1" (or mixed surfaces).
The RSS water vapor product, derived from GMI brightness temperature observations, is already spatially gridded to 0.25°× 0.25°.Atmospheric water vapor data are daily averages for both the ascending and descending portions of GPM-CO's daily orbit separately.We match these data to individual swaths of GMI GPROF data by converting the given fractional hour information and identifying the associated overpass within GMI GPROF timesteps.
Finally, to merge the M18 AR data, we split the 0.5°× 0.5°grid boxes into 0.25°× 0.25°to match the RSS water vapor and now-gridded GMI GPROF V7 data.Next, we find the midpoint of each GPM-CO overpass through the basin (either NA or Pacific) and identify the nearest 6-hourly timestep defined in the M18 AR data set (00, 06, 12, and 18 UTC).By matching the spatio-temporal resolution between these data sets, we are able to analyze satellite precipitation products during AR events (or precipitation when no AR is present, or "No AR" events).We have eliminated AR data outside the GMI swaths to remain a GMI-based merged data set, though the M18 data set encompasses the Northern Hemisphere at latitudes ≥10°N.
Notably, for the remainder of this paper, we only show the merged data product over ocean surfaces (where the gridded surface flag is "0"), where ARs are more frequent and intense due to the availability of water (Guan & Waliser, 2015).Additionally, this remains consistent with the available observations from the RSS water vapor product, and avoids any biases introduced by ground clutter contamination or impacts from coastlines in the GMI observations.Though not included here, the merged product does contain the available gridded GMI GPROF and M18 AR data over land and "mixed" (coastline or sea ice) surfaces that fall within the specified basin.The lack of surface observations over the NA and Pacific Oceans complicates validating AR algorithms over-ocean and studying AR impacts on meteorology.The perspective of spaceborne instrumentation is thus a bridge between reanalysis and satellite observations to study high-latitude ARs and precipitation over ocean.See Table A1 in Appendix A for the outline of variable names within the merged data set that are used in the following figures.

Example Gridded Output
An example of the pre-and post-gridded RSS daily mean water vapor and GMI GPROF V7 precipitation rates is shown in Figure 1.This GPM-CO overpass, at ∼0700 UTC on 23 December 2016, preceded an extreme precipitation event.Figure 1a shows the RSS ascending daily mean atmospheric water vapor obtained from the RSS website at 0.25°× 0.25°spatial resolution.Figure 1c-show the GMI GPROF precipitation estimates obtained through the GES DISC at the original GMI footprint resolution.M18 ARs are designated by the contour at 0.5°s patial resolution.The post-gridded and merged water vapor (Figure 1b), GMI GPROF precipitation (Figure 1d), and M18 AR (contours) are the data available in the merged data set.As noted above, we mask data over coastlines and land surfaces for these figures, though the data are available in the merged data set.The merged data shown in Figures 1b and 1d can be recreated using the time step "2016-12-23 07:25:00" in the file name gridded_atlantic_201612.nc in the provided product.Notably, the merged data (Figures 1b and 1d) maintain similar features to the original resolution for all three data products (e.g., high water vapor and precipitation rates along the meridional length of the AR).The discontinuity in the RSS water vapor product (Figures 1a and 1b) west of Norway that is not present in the GMI GPROF precipitation (Figures 1c and 1d) can be attributed to the difference in the original data resolution between the RSS water vapor (daily means of ascending or descending water vapor) and GMI GPROF (individual swaths).
Figure 2 shows the counts of GMI footprints collected per 0.25°× 0.25°grid box for the North (a) Atlantic and (b) Pacific for January-December 2016 over ocean only.There is a sharp meridional gradient of footprints with maximum concentration at ∼61°N due to the satellite orbital inclination.This maximum is lower in the NP than the NA, due to the presence of sea ice at ∼60°N, and subsequent masking of observations over land/ice.Though we do not include data over land or sea ice in this paper, data for all surface types are available in the merged data set netCDFs. Figure 3 shows the composite frequency of the M18 ARs coincident with available GMI footprints in DJF (Figures 3a and 3b) and JJA (Figures 3c and 3d).In DJF (Figures 3a and 3b), M18 AR frequency is <10% in the western (northwestern) portion of the NP (Atlantic), but M18 AR > 10% along and to the east of the storm tracks (southwest to northeast; Blackmon, 1976;Hoskins & Hodges, 2019a, 2019b;Hoskins & Valdes, 1990).Literature has identified these eastward and poleward tracks as locations of AR frequency maxima (Guan & Waliser, 2015;Zhu & Newell, 1998).This indicates that the GPM-CO has high enough spatio-temporal coverage to reasonably represent seasonal frequency of ARs.In JJA (Figures 3c and 3d), M18 ARs occur more frequently (>10% of the time) in both the NA and NP basins.Unlike the zonal contrast of AR frequency in DJF, there is little variation in frequency across both basins in JJA.The highest frequencies (>15%) in JJA are in the southeast NP basin, south of the Kamchatka Peninsula, and just east of Baffin Island.Frequency is normalized by the total number of GMI footprints per season regardless of precipitation observations.This figure is created using the variable "AR_flag" and is normalized using "opass_counts" shown in Figure 2.

Earth and Space Science
10.1029/2023EA003333

ERA5
To show an application of the merged data set, we also utilize hourly, gridded (0.25°× 0.25°) T2M and total column water vapor (another term for TPW) data products from the ECMWF Reanalysis v5 (ERA5; Hersbach et al., 2020).The spatial resolution of ERA5 matches that of the merged data set (0.25°× 0.25°).To match hourly ERA5 data to the irregular timesteps within the gridded data set, we pair the nearest-time ERA5 timestep and location to the corresponding timestep and location in the merged data set (i.e., the median times of each overpass through the basin at specific grid points).T2M and TPW are two of the ancillary meteorological variables used to estimate precipitation rates in the GMI GPROF algorithm.We then use the gridded AR flag to examine the meteorological (T2M and TPW) differences based on coincident M18 ARs (see details in Section 5).

Examples of Merged Data
Figures 4 and 5 show the gridded mean RSS water vapor for DJF and JJA in the NA and NP, respectively.These composites are created using the variables "RSS_wv" and "AR_flag" in the merged data set.Noting the different colorbar scales, the mean water vapor is generally higher in JJA (Figures 4c,4d,5c,and 5d) than in DJF (Figures 4a,4b,5a,and 5b) for both basins.Additionally, there is higher mean water vapor when an M18 AR is present (Figures 4a,4c,5a,and 5c).This is expected given that existing AR detection schemes include percentile analyses of integrated water vapor (Ralph et al., 2018), however it has not been before illustrated with satellitebased observations of water vapor.Inversely, when no M18 AR is present, mean water vapor content is much lower as a function of the respective season (Figures 4b,4d,5b,and 5d).Regions of relatively high mean water vapor in the basins during ARs (Figures 4a, 4c, 5a, and 5c) coincide with high AR frequency (see Figure 3): along storm tracks and in the southeast basin during DJF (Figures 3a and 3c) and basin-wide in JJA (Figures 3b and 3d).This result is expected, and indicates that the RSS water vapor product, independent of the M18 AR data set, is capable of representing meteorological conditions during AR and no AR events.
We bin the gridded GMI GPROF V7 precipitation rates (LWE) in Figure 6 seasonally and by basin, with the blue (red) line representing M18 AR (no AR) coincidence.The dashed lines at 0.5 and 1.0 mm hr 1 LWE represent thresholds used for additional variables in the merged data set (see Section 6).There is little difference in the amount of precipitation or the frequency of AR-coincident precipitation between DJF (Figures 6a and 6c) and JJA (Figures 6b and 6d).Precipitation in the NA (Figures 6a and 6b) is more frequent than in the NP (Figures 6c and  6d).Instances of higher precipitation rates in the NP are most likely coincident with an AR (>0.5 mm hr 1 in DJF, >1.0 mm hr 1 in JJA; Figures 6c and 6d).In the NA, ARs are present for approximately half of precipitation occurrences (Figures 6a and 6b).In all cases, most occurrences of light precipitation are not coincident with AR.
Figures 7 and 8 show the DJF and JJA, gridded conditional mean (≥0.1 mm hr 1 LWE) GMI GPROF precipitation rates in the NA and NP, respectively, from the merged data set.Precipitation rates are generally higher when ARs are present (Figures 7a,7c,8a,and 8c).The mean precipitation rate when an AR is present in the NA is 0.72 (0.78) mm hr 1 in DJF (JJA) and in the NP, is 0.70 (0.81) mm hr 1 in DJF (JJA).Alternatively, precipitation rates are lower when no AR is present (Figures 7b, 7d, 8b, and 8d), with a basin-wide mean precipitation rate in the NA of 0.36 (0.51) mm hr 1 in DJF (JJA) and in the NP of 0.31 (0.49) mm hr 1 in DJF (JJA).JJA precipitation rates (Figures 7c, 7d, 8c, and 8d) are generally higher than DJF precipitation rates (Figures 7a, 7b, 8a, and 8b) for both coincident AR and.
In DJF, precipitation rates are highest in the southeast region of both basins, and low in the northwest region.For example, in the NA (Figures 7a and 7b), precipitation rates are lower (<0.5 mm hr 1 ) north and west of the NA storm track.There is a similar pattern in the NP (Figures 8a and 8b), where precipitation rates are lower (<0.25 mm hr 1 ) north and west of the Pacific storm track (Guan & Waliser, 2015;Zhu & Newell, 1998), with the exception of higher rates in the Sea of Okhotsk (Figures 8a and 8b) for both AR and non-AR scenarios.JJA precipitation rates do not show a similar pattern along the storm tracks for either basin.In both the Atlantic (Figures 7c and 7d) and Pacific basins (Figures 8c and 8d) during JJA, mean rates are highest (≥1.0 mm hr 1 ) at lower latitudes (45-50°N) and when an AR is present.Mean rates exceed 2 mm hr 1 in the southwest portion of both basins during JJA AR events (Figures 7c and 8c).
Tables 2 and 3 list the percentage of GMI observations in DJF and JJA, respectively, filtered by minimum gridded GMI GPROF precipitation threshold (≥0.1, 0.5, and 1.0 mm hr 1 ), presence of an AR, and latitudinal boundaries (5°latitudinal increments and 45-65°over ocean only).In both the NA and NP basins (45°-65°N), high precipitation rates are most often coincident with ARs, and low rates with no AR.In DJF, higher rates are more likely to be associated with ARs than during JJA.This is most apparent in the NP basin, where 72.5% of high precipitation rates (≥1.0 mm hr 1 ) in DJF occur during an AR event.The percentage of AR-coincident high precipitation rates also increases as a function of latitude (poleward) in both basins, in agreement with the findings of Nash et al. (2018).
The exception is that in JJA in the NP basin, there is a slight decrease from 55°-60°N to 60°-65°N.One factor that may account for this is that there are fewer GMI footprints in the NP due to the higher concentration of land surface grid boxes in the basin.Additionally, all JJA scenarios have more footprints than DJF as there are 2 months more data included in the 2014-2022 data set (January and February 2014 pre-date GPM-CO) as well as reduced sea ice coverage (and therefore more over-ocean footprints) in JJA.

Example Application of Merged Data Set
To demonstrate a scientific application of this merged data set, we examine the ERA5 T2M and total column water vapor (also known as TPW) data products during precipitating (threshold of "surface_precip" ≥ 0.1 mm hr 1 ) AR and No AR conditions ("AR_flag" = 1, and 0, respectively).The GPROF algorithm estimates precipitation rates in   conjunction with an a priori database that includes GANAL T2M and TPW, values both of which may be impacted by the presence of an AR.Histograms of T2M and TPW for the NA (Figure 9) and NP (Figure 10) can be directly compared to the ocean surface type histogram in Figure 8.2 of Randel et al. (2020) representing the environmental conditions within the GPROF a priori database.
The corresponding histograms between the two basins (Figures 9 and 10) further explore the seasonality and AR presence during these environmental conditions.The composite of all 4 panels within each figure would closely resemble Figure 8.2 in Randel et al. (2020), indicating that precipitating AR events are represented in the GPROF   9d, 10c, and 10d).This corresponds to the colder, drier conditions in DJF and relatively warm, more moist conditions in JJA.Most retrieved precipitation in DJF occurs when the atmospheric water vapor content is low (<20 mm), but at or above-freezing near-surface temperatures (T2M between 270 and 280 K).Retrieved precipitation in JJA occurs at a wide range of atmospheric moisture content (TPW between 10 and 65 mm) but most precipitation occurs at warm temperatures (T2M between 280 and 290 K).
The differences between T2M and TPW during precipitating AR versus No AR events are more subtle than the differences between seasons.As expected, the environment is often warmer and more moist during precipitation with ARs present (Figures 9a, 9c, 10a, and 10c) than when no AR is present (Figures 9b, 9d, 10b, and 10d).Precipitation is most frequent during No AR conditions and in DJF, can occur during very cold (T2M < 270 K) and dry (TPW <5 mm) conditions (Figures 9b and 10b).Further, Figures 9 and 10 demonstrate the use of T2M and TPW as a good choice of constraints for GPROF.This merged data set thus provides insight on the ability to examine retrieval performance and characteristics as a function of the environment, which is critical information for algorithm developers.

Summary and Conclusions
This paper presents a new data set of coincident gridded GPM Core Observatory data products and M18 (Mattingly et al., 2018) ARs in the NA and Pacific sectors (45°N-68°N) to provide the scientific community with a spaceborne perspective of AR effects on high-latitude precipitation.Matching an irregular grid of GPM-CO data in time and space to gridded, reanalysis-derived M18 ARs can be computationally challenging, and we therefore

Figure 1 .
Figure 1.An example of Remote Sensing Systems (RSS) water vapor (a), (b) and GPM Microwave Imager (GMI) GPROF precipitation estimates (c), (d) during one time step (an ascending overpass) on 23 December 2016.The matched M18 Atmospheric river (AR) is outlined in magenta (top row) and teal (bottom row).All data are shown before gridding (a), (c) and after gridding for the merged data set (b), (d).Before gridding, GMI GPROF precipitation data are at GMI footprint resolution.The original gridded resolution of the RSS water vapor and M18 AR data is 0.25°× 0.25°and 0.5°× 0.5°, respectively.The merged data set resolution for all data is 0.25°× 0.25°.In panels (b) and (d) are created using the variables "AR_flag," "surface_precip," and "RSS_wv" in the merged data set.

Figure 2 .
Figure 2. Cumulative number of GPM Microwave Imager footprints (ice-and land-free), January-December 2016 per 0.25°× 0.25°grid box for the North (a) Atlantic and (b) Pacific.This figure is created using the variable "opass_counts" in the merged data set.

Figure 3 .
Figure 3. Frequency of atmospheric rivers in (a), (c) DJF and (b), (d) JJA coincident with GPM Microwave Imager (GMI) footprints (May 2014-December 2022).Frequency is normalized by the total number of GMI footprints per season regardless of precipitation observations.This figure is created using the variable "AR_flag" and is normalized using "opass_counts" shown in Figure2.

Figure 4 .
Figure 4. Merged, gridded seasonal averages (May 2014-December 2022) of the daily mean atmospheric water vapor in (a), (b) DJF and (c), (d) JJA in the North Atlantic during Atmospheric river (AR) and No AR events.Water vapor data are originally from the Remote Sensing Systems and derived from GPM Microwave Imager data.This figure is created using the variables "AR_flag" and "RSS_wv."

Figure 5 .
Figure 5. Merged, gridded seasonal averages (May 2014-December 2022) of the daily mean atmospheric water vapor in (a), (b) DJF and (c), (d) JJA in the North Pacific during Atmospheric river (AR) and no AR events.Water vapor data are originally from the Remote Sensing Systems and derived from GPM Microwave Imager data.This figure is created using the variables "AR_flag" and "RSS_wv."

Figure 6 .
Figure 6.Histograms of gridded GPM Microwave Imager GPROF V7 precipitation rates in the (a) and (b) North Atlantic and (c), (d) North Pacific from May 2014 to December 2022.Precipitation rates are categorized as coincident with an M18 Atmospheric river (blue line) or not (red line) present.Vertical dashed lines are at 0.5 and 1.0 mm hr 1 liquid water equivalent for reference.This figure is created using the variables "AR_flag" and "surface_precip" in the merged data set.

Figure 7 .
Figure 7. Seasonal conditional mean (≥0.1 mm hr 1 liquid water equivalent) GPM Microwave Imager GPROF precipitation rates (May 2014-December 2022) in (a), (b) DJF and (c), (d) JJA during Atmospheric river (AR) and No AR events in the North Atlantic.This figure is created using the variables "AR_flag" and "surface_precip" in the merged data set.

Figure 8 .
Figure 8. Seasonal conditional mean (≥0.1 mm hr 1 liquid water equivalent) GPM Microwave Imager GPROF precipitation rates (May 2014-December 2022) in (a), (b) DJF and (c), (d) JJA during Atmospheric river (AR) and No AR events in the North Pacific.This figure is created using the variables "AR_flag" and "surface_precip" in the merged data set.

Figure 9 .
Figure 9. 2D Histograms of ERA5 T2M and total column water vapor (i.e., total precipitable water) in the North Atlantic when matched gridded GPM Microwave Imager GPROF precipitation rate ≥0.1 mm hr 1 liquid water equivalent (May 2014-December 2022).
Files are separated into individual basins (North Atlantic or North Pacific) and each file contains 1 month of data.For example, the file containing the merged data set for May 2014 in the North Atlantic is named, "atlantic_201405.nc."The dimensions for this file are: timestep = 288 (number of overpasses through the North Atlantic basin during May 2014), latitude = 100, and longitude = 320.Precipitation rates are expressed as mm hr 1 liquid water equivalent.The monthly files are compressed into year and basin: either the North Atlantic or the North Pacific (e.g., NA_2014) and zipped.The files have the basin name indicated and are by year and month (e.g., gridded_atlantic_YYYYMM.nc).Each netCDF file contains the variables.

Table 1
Resolution of the Three Original Data Sets and the Merged Data Set, Including References for the Data Sets

Table 2
Percentage of GPM Microwave Imager (GMI) Footprints in DJF (May 2014-December 2022) With GMI GPROF Precipitation Rates (Liquid Water Equivalent) Exceeding a Minimum Threshold (Per 5°Latitude Bands: 45°-50°N, 50°-55°N , etc.) and North Pacific (Top Portion) and North Atlantic (Bottom Portion) Basins South of 65°N (Over Ocean Only) Note.Percentages categorized by both the presence of an AR and the minimum precipitation rate.
Note.Percentages categorized by both the presence of an AR and the minimum precipitation rate.
MATELING ET AL. a priori database.There are more instances of precipitation at lower T2M and TPW values in DJF (Figures9a, 9b, 10a, and 10b) than in JJA, and more instances of precipitation at higher T2M and TPW values in JJA (Figures 9c,