2021 Volume 99 Issue 4 Pages 1003-1022
This paper focuses on the uncertainty of summer precipitation estimations produced by Global Satellite Mapping of Precipitation (GSMaP) over Mongolia, a region with complex terrain and sparse weather observation networks. We first compared the average summer precipitation over Mongolian territory as reported by several precipitation products. Although the interannual variability of the product was comparable, the amount of recorded precipitation differed among various products. The rain gauge-based analysis reported the lowest amount of precipitation, whereas the satellite-based GSMaP_MVK (Moving Vector algorithm with Kalman filter) reported the highest amount. Our results represent the first estimate of the characteristic differences among various precipitation-monitoring products, including Global Precipitation Measurement (GPM)-based products, as they relate to climatic and hydrometeorological assessments in Mongolia. We then performed a detailed comparison using a case study, in which a heavy rainfall event was captured by the GPM mission's core observatory near Ulaanbaatar in July 2016. In this case, gauged and ungauged GSMaP estimates of the precipitation over the mountain area significantly differed between algorithm versions 6 and 7. An intercomparison of atmospheric numerical modeling, the GPM core observatory, and rain gauge observation revealed that the rain gauge calibration of GSMaP effectively moderates the large error of the ungauged GSMaP data. The source of the significant ungauged GSMaP error is likely to be the rain rate estimates in algorithm version 7. However, the GSMaP gauge-calibrated estimates of the precipitation over mountainous areas may be affected by a potential underestimation of gauge analysis due to the missing localized precipitation occurring in the large gaps of the routine observation network. We expect that these findings will be helpful for developers aiming to further improve the GSMaP algorithm.
Mongolia, which is situated in northeastern Eurasia, is characterized by a meridional gradient climatic zone determined by precipitation and temperature. The annual precipitation varies from less than 100 mm in the desert and arid/semiarid regions in the southern part of Mongolia to more than 350 mm in the boreal forest region in the north. Summer precipitation in June, July, and August accounts for 50–60 % of the annual precipitation. Following the United Nations Framework Convention on Climate Change, the Third National Communication of Mongolia (Ministry of Environment and Tourism 2018) reported that summer precipitation occasionally induces natural disasters triggered by short-range events of heavy rain and flash floods. Such disasters account for 22.4 % of the total disasters in the territory of Mongolia. Over the past decade, the occurrence of extreme events related to atmospheric phenomena has doubled. To properly evaluate the hydrological and socioecological impacts, accurate measurement and monitoring of precipitation over a sufficient period and geographic area are important. However, the sparseness of the meteorological observation network in Mongolia limits the country's ability to satisfy the growing societal demand for reliable meteorological information. The government currently operates 70 manned meteorological stations and 60 automatic stations (Battur 2010). The complexity of mountains that occupy most of Mongolia (Fig. 1a) and divide the meridional climatic regimes makes it difficult to create a dense observation network. Such a sparse meteorological network could very well miss precipitation that occurs in remote areas far removed from the gauged stations in the short term and potentially hide the total amount of terrestrial water across the national territory.
Satellite measurement is one of the most powerful tools for monitoring global precipitation, especially in areas such as Mongolia, which lack sufficient surface observational networks. In February 2014, the Global Precipitation Measurement (GPM) mission (Hou et al. 2014), a successor to the Tropical Rainfall Measuring Mission (TRMM), was launched as the mission's core satellite observatory. The GPM core observatory carries both a dual-frequency precipitation radar (DPR) and a multichannel GPM microwave imager (GMI) developed by the Japan Aerospace Exploration Agency (JAXA) and the National Aeronautics and Space Administration (NASA), respectively. The DPR operates in two radar bands (13.6 GHz for the KuPR and 35.5 GHz for the KaPR) designed to directly capture the vertical structure of the precipitation. Satellite-based precipitation estimates are known to have a substantial degree of uncertainty, especially at higher latitudes (Tian and Peters-Lidard 2010). Even though the swath width of the GPM core observatory is rather narrow, it has greatly expanded coverage to include mid-/high latitudes (65°N/S) as compared with the more limited coverage of TRMM (35°N/S), with its non-sun-synchronous orbit (Skofronick-Jackson et al. 2017). Thus, it is anticipated that measurements by the GPM core observatory will help reveal the Earth's water distribution and energy cycle in mid-/highlatitude land areas.
In addition to the GPM, the development of precipitation products based on multi-satellite combinations is ongoing. For example, JAXA's Global Satellite Mapping of Precipitation (GSMaP; Kubota et al. 2007) provides gridded precipitation rates in high temporal (1-h intervals) and spatial (0.1° in longitude and latitude within 60° N/S) resolutions using a combination of microwave and infrared radiometers on multiple satellites and the Moving Vector algorithm with Kalman filter (Ushio et al. 2009; hereafter called GSMaP_MVK). GSMaP_MVK can also be calibrated by rain gauge observation (Mega et al. 2018) using CPC gauge-based analyses of global daily precipitation with a 0.5° grid (Xie et al. 2007; Chen et al. 2008). This GSMaP product is called the GSMaP_Gauge.
Numerous studies based on statistical indices on various spatiotemporal scales have evaluated the systematic error and bias of GSMaP in the central and east Asian areas surrounding the Mongolian territory by comparing the performance of GSMaP with the gridded analysis originating from rain gauge observatories (e.g., Tan et al. 2018; Ning et al. 2017; Deng et al. 2018; Guo et al. 2017). Generally speaking, the rain gauge-calibrated GSMaP_Gauge performs better than the GSMaP_MVK in these regions. Calibration by the gridded analysis of rain gauge observations effectively reduces errors and biases related to several factors, such as season, rain type, rain intensity, land surface condition, and topography. However, the benefit of rain gauge calibration depends on the density of the gauge network or the source of the dataset. In some cases, such calibration can cause overcorrection or generate invalid results, especially in mountainous regions (e.g., Derin et al. 2016, 2019; Deng et al. 2018; Rozante et al. 2018; Takido et al. 2016; Yuan et al. 2019). Moreover, multi-satellite combination products such as GSMaP depend on the sensors on each satellite (microwave imager, sounder, and infrared radiometer). Chen et al. (2019) found that the precipitation estimates of GSMaP using GMI measurement, which works as a reference for partner satellites in the GPM constellation, tend to have relatively large biases for spring and summer rainfall over the Chinese mainland. The authors pointed out that the unsatisfactory detection capability of GSMaP's GMI algorithm, which resulted in a relatively large hit and miss biases, needs to be improved further.
Although anterior statistical analysis has demonstrated the general error features of GSMaP in the region by using a large number of grids and time series, it remains unclear when and under what circumstances the uncertainty associated with these satellite-based precipitation estimates tends to be most serious. To bridge the gap, an accumulation of case-based analyses of specific satellite precipitation errors in conjunction with national-scale climatological analyses would be of great help in the identification of the error structure and improvement of the algorithm used for satellite precipitation estimation. Notably, a general or severe error in specific grids due to the algorithm might not always result in a significant mistake in climatic or national-scale precipitation estimates. Averaging over a longer period and in a broader area may cancel out the error components in the individual grids. Thus, the combination of a case study focused on a small region and climatic analysis averaged over a national-scale area should provide users with more helpful information with regard to the performance of GSMaP.
The present paper reports a case study involving a heavy rainfall event captured by the GPM core observatory in the vicinity of Ulaanbaatar (UB), Mongolia, in July 2016. Here, GSMaP_MVK and GSMaP_Gauge exhibited a significant difference in their precipitation measurements. To evaluate the measurement reliability, we compared the precipitation structure described by GSMaP, atmospheric numerical modeling, and the GPM core observatory's active and passive measurement (KuPR and GMI) with area rain gauge observations. Of course, this single case study does not establish the general features of satellite retrieval and gauge calibration errors. However, we believe that our results represent a useful step toward improving satellite estimation of daily-scale rainfall in the GPM era. Moreover, to the best of our knowledge, there have been few validation studies of GSMaP focused on high-latitude Eurasia (Guo et al. 2017). The performance of GSMaP and gauge calibration is still unclear in Mongolia, with its complex mountainous terrain and sparse meteorological network. In this study, we first compared the national-scale summer precipitation estimates of gauged and ungauged GSMaP with other precipitation products (rain gauge-based analysis, atmospheric reanalysis, and a combined satellite and rain gauge product). More specifically, we performed a simple comparison of the interannual variability and average amount of precipitation over Mongolia for various products. This enabled us to make an educated first guess of the characteristic differences among the products in their climatic and hydrological assessments. However, it should be noted that a detailed statistical analysis to validate GSMaP for the Mongolian territory is beyond the scope of this study.
In conducting our analysis, we used the standard products of version 6 and 7 of the GSMaP algorithm (Kubota et al. 2020). The GSMaP microwave radiometer algorithm detects the precipitation rate based on the measured brightness temperature and a look-up table. Over the land region, the retrieval algorithm depends only on the scattering signature from ice crystals over the spectrum of higher frequencies (e.g., 89 GHz), as the high and variable emissivity over land inhibits the use of an emission signature over the spectrum of lower frequencies (Aonashi et al. 2009). The look-up table is created by a radiative transfer model using both atmospheric gridded analysis and a precipitation database built by the TRMM precipitation radar (version 6) and GPM_KuPR (version 7). Moreover, the GSMaP algorithm after version 6 has installed an orographic/nonorographic rainfall classification scheme. This scheme is expected to reduce underestimation due to shallow precipitation systems over the mountainous regions and improve precipitation estimates over the TRMM observation area (Shige et al. 2013; Taniguchi et al. 2013; Yamamoto and Shige 2015). The threshold for orographic rainfall detection is based on the magnitude of both moisture flux convergence and the forced vertical motion (w) derived from the horizontal wind speed in the atmospheric gridded analysis. When the threshold is satisfied, the look-up table switches to an orographic table. Therefore, the estimated precipitation rate will differ from the native microwave radiometer estimate. However, version 6 of the algorithm overestimates orographic rainfall as both wind speed references at the surface and the threshold of w are fixed (Yamamoto et al. 2017). Accordingly, version 7 of the algorithm modifies the definition of horizontal wind speed and uses the upstream and low-level atmosphere, adopting a fractional threshold of w dependent on the wind speed magnitude (Yamamoto et al. 2017). Subsequently, the gauge calibration adjusts the hourly precipitation rate following optimal theory to fit the daily value of the CPC gauge-based analyses (Mega et al. 2018). As a result, the precipitation estimates by GSMaP with or without the gauging process will differ, even though the same satellite measurement is referenced. In this paper, we call the products of version 7 as GSMaP_MVK and GSMaP_Gauge, whereas those of version 6 are referred to as GSMaP_MVKv6 and GSMaP_Gaugev6. Daily precipitation was calculated from the hourly product. For the intercomparisons of summer precipitation (June, July, and August), we used the monthly product from 2014 to 2017, since GSMaP includes measurements by the GPM observatories after March 2014. Moreover, we referenced the reanalysis version of GSMaP prior to 2014 compiled using version 6 of the algorithm without GPM observatories, calling the products GSMaP_RNL and GSMaP_Gauge_RNL.
2.2 GPM observationIn the case study, we focused on the precipitation event around UB in Mongolia on 09 July 2016. The GPM core observatory (orbit number 013432) passed near UB at approximately 23:57 UTC and captured heavy precipitation. We used the GPM_KuPR, GPM_KaPR, and DPR products compiled with a level 2 algorithm version 05A (Iguchi et al. 2018; Seto et al. 2021). The GPM_KuPR product consists of 150 vertical layers with 125-m intervals; the spatial resolution is 5 km, covering a 245-km-wide swath scanned by 49 beams. The matched scan of GPM_KaPR covers a 125 km-wide swath scanned by 25 beams inside the swath of GPM_KuPR. The normal DPR scan data were used in this study. Conversely, the horizontal swath width of GMI is approximately 904 km, which is significantly broader than that of GPM_KuPR. The retrieval from GMI is derived by the Goddard Profiling algorithm (Kummerow et al. 2001), hereafter referred to as GMI/GPROF.
2.3 Other precipitation datasets, rain gauge observation, and numerical model configurationFor the rain gauge-based dataset, we used the Asian Precipitation – Highly Resolved Observational Data Integration Towards Evaluation of Extreme Events (APHRODITE-1, v1101) data from 1979 to 2007 and APHRODITE-2 (v1901) data from 1998 to 2015 (Yatagai et al. 2012, 2020). For the combination of satellite and rain gauge observations, data from the Global Precipitation Climatology Project (GPCP) version 2.3 for 1979 to 2017 (Adler et al. 2003) and version 3.1 for 1984 to 2017 (GPCPV31, Huffman et al. 2020) were utilized. The horizontal grid spacings of both are 2.5° and 0.5° longitude and latitude grids, respectively. GPCPV31 also provides the precipitation estimate without gauge calibration (hereafter referred to as GPCPV31_SatOnly). Thus, it can compare with that of GSMaP_MVK. Moreover, the information about the relative weight of the rain gauge on each grid point is also provided to quantify the gauge analysis's percent weighting merged to the satellite-only estimate.
The estimation of precipitation by atmospheric reanalysis used the ERA5 data generated by the European Centre for Medium-Range Weather Forecasts from 1979 to 2017, with 0.25° grids (Hersbach et al. 2020). Moreover, we used ERA5 to detect the synoptic atmospheric condition for the case study. All products are converted to monthly precipitation; the various time series of summer precipitation averaged over the Mongolian territory were compared.
As a reference for the satellite measurements and GSMaP estimates, we utilized the rain gauge observation records from the Bagabayan station, installed by the Information and Research Institute of Meteorology, Hydrology and Environment in Mongolia. The Bagabayan rain gauge was installed on a mountainside located approximately 35 km north of the Chinggis Khaan International Airport (Fig. 1b). It observed precipitation in 10-min intervals over the entire summer season of 2016. Because Bagabayan was outside the routine observation network, we treated the Bagabayan observation data as independent from the satellite measurements and the data calibrated by rain gauge observation.
a) Map of the topography over East Asia and the location of grid cells with installed rain gauge observation systems in APHRODITE (red cell) and CPC gauge analysis (blue cell). The grid cells of APHRODITE, which had more than one rain gauge in July 2015 and are situated below 55°N, are presented. Those with CPC gauge analysis systems also had more than one rain gauge on 9 July 2017; b) same as a) but for the domain of atmospheric numerical modeling enclosed by a black dashed box in a). Only grid cells of CPC data are presented.
As a second reference, we employed the regional atmospheric modeling approach used by the Weather Research and Forecasting model version 4.0 (WRF; Skamarock et al. 2019). The regional atmospheric model provided us with the capacity to examine precipitation that occurs in gaps of the surface observation networks, such as in mountainous regions. Thus, WRF could help establish a reliable distribution of accumulated precipitation. The model domain, which consists of 230 × 196 grids with 5-km resolution, covers the northern part of Mongolia centered around UB (Fig. 1a). The vertical resolution has 51 layers, up to 50 hPa. Model integration started at 12:00 UTC on 09 July 2016, which preceded by approximately 12 h the time at which the GPM core satellite passed above UB. The initial and lateral boundary conditions used ERA5, which operates every hour on 37 pressure levels. As for the microphysics scheme in the model, the Thompson scheme that includes ice, snow, and graupel processes (Thompson et al. 2008) was adopted to calculate the cloud microphysics; the Rapid Radiative Transfer Model for GCMs (Iacono et al. 2008) was used for short- and longwave radiation; the Unifi ed Noah Land Surface model (Tewari et al. 2004) served as the land surface model; Mellor–Yamada–Nakanishi–Niino (MYNN) level 2.5 (Nakanishi and Niino 2009) was used for both the atmospheric boundary scheme and surface layer scheme. The sub-grid cumulus scheme was not adopted. Since WRF outputs the precipitation level every 10 min, we were able to evaluate the total precipitation in 10-min intervals.
We fi rst examined the fundamental characteristics of the summer precipitation represented by GSMaP, focusing on the eastern part of Eurasia (Fig. 1a). The distribution of the rain gauge networks is dense at low-latitude areas and sparse at high-latitude areas (Fig. 1a). The gauge grid points (APHRODITE and CPC) in the Mongolian territory have relatively poor coverage compared with East Asia and India. In the enlarged map around UB (Fig. 1b), gauge stations are mainly located at relatively low elevations, such as at valleys and basins in the complex terrain.
Figures 2a and 2b present the accumulation of summer precipitation represented by GSMaP_Gauge and GSMaP_MVK averaged from 2014 to 2017. The distribution of summer precipitation indicated by GSMaP_Gauge tends to be quite smooth, contrary to the GSMaP_MVK distribution, which exhibits a jagged pattern owing to some considerable precipitation in small regions. The GSMaP_Gauge minus GSMaP_MVK precipitation difference tends to be negative at higher latitudes and positive at lower latitudes (Fig. 2c), indicating that calibration by rain gauge observation decreases the precipitation estimated by satellite-only means at higher latitudes. The region exhibiting the largest decrease corresponds to the northern side of the mountainous region in northern Mongolia (Figs. 1a, 2c). Figures 2d and 2e present the standard deviation of daily precipitation in GSMaP_Gauge and GSMaP_MVK, which was calculated using days with more than 1 mm of daily precipitation. The variability of daily precipitation for GSMaP_MVK is higher than that for GSMaP_Gauge, which reflects the more complicated distribution of precipitation characterized by extreme values in narrow regions. In the northern part of Mongolia, GSMaP_Gauge damps the high variability estimated by GSMaP_MVK.
a) Total summer precipitation (mm) between June and August averaged from 2014 to 2017 by GSMaP_Gauge; b) same as a) but for GSMaP_MVK; c) the anomaly in the summer precipitation of GSMaP_Gauge vs GSMaP_MVK; d) standard deviation of daily precipitation (mm day−1) calculated by GSMaP_Gauge; e) Same as d) but for GSMaP_MVK; f), g), and h) same as a), b), and c) but for GPCPV31 and GPCPV31_SatOnly; i) the relative weight of gauge analysis merged to the satellite-only estimation in GPCPV31. The weight is averaged in June, July, and August from 2014 to 2017.
The differences observed between GSMaP_Gauge and GSMaP_MVK were also identified between GPCPV31 and GPCPV31_SatOnly, averaged summer precipitation during the same period of GSMaP (Figs. 2f–h). The horizontal pattern of the anomaly which is positive at low latitudes and negative at high latitudes is consistent with that of GSMaP, although the magnitude is small (Figs. 2c, h). The northern part of the Mongolian territory was also characterized by negative anomalies, which indicates that the satellite-only precipitation estimates are likely to be large (Figs. 2b, g), but the gauge calibration moderates them (Figs. 2a, f). This region could be regarded as a typical region depending on the quality of rain gauge dataset because the weight of gauge analysis is higher than the surrounding region, which is more than 60 % weighting to the satellite-only estimates (Fig. 2i).
Overall, the calibration by rain gauge observation reduces the mean and variability of precipitation determined by satellite measurement at higher latitudes. If a product based on sparse rain gauge observations tends toward underestimation, using such observations to calibrate satellite measurements may lead to misrepresentation of the actual precipitation and loss of spatial variability in some areas.
b. Interannual variability and amount of precipitationNext, we compared the interannual variability and average amount of summer precipitation over Mongolia indicated by various precipitation products (Fig. 3). The variability of each product tended to exhibit similar fluctuations. The correlation coefficients of ERA5, GPCPV31, and GPCP against APHRODITE between 1984 and 2015 were 0.88, 0.89, and 0.98, respectively (the APHRODITE time series was coupled with v1 and v2 from 1998 onward to construct the long-term series). The correlation coefficient of GSMaP_Gauge_RNL between 2000 and 2013 was 0.95, whereas the satellite-only product for GSMaP_RNL had a slightly lower coefficient of 0.72. GSMaP_Gauge also demonstrated the same interannual fluctuations as ERA5 and GPCP, although the observation period here is very short. Algorithm versions 6 and 7 of GSMaP yielded almost identical results to those of GSMaP_MVKs and GSMaP_Gauges. Thus, the difference in algorithms had little bearing on the outcomes—producing neither a large improvement nor a significant decline in the quality of national-scale precipitation estimates over Mongolia.
Interannual variability of summer precipitation (mm) averaged over Mongolia from 1979 to 2017. The Mongolian region is highlighted by the black box presented in Fig. 2c. It extends from 88°E to 120°E and 42°N to 52°N.
Based on these comparisons, the implication is that the differences among products pose no serious problem for a climatic assessment of Mongolia focused on variability on the national scale. However, for estimating water resources, care should be taken in the selection of the precipitation dataset, as the amount significantly varies, depending on the source of the products. APHRODITE, based on rain gauge observations, estimates lower precipitation than the others, whereas GSMaP_MVKs and GSMaP_RNL, based only on satellite measurements, show a considerably higher amount of precipitation. ERA5, GPCP, and GSMaP_Gauges estimate similar amounts, especially over the recent decade, even though their data sources are different; ERA5 is the forecast of an atmospheric model, whereas the others are a combination of satellite and rain gauge observations. The estimated amounts fall between those of APHRODITE and satellite-based GSMaP products. Interestingly, the modern GPCPV31 is nearly comparable with APHRODITE. The amount of precipitation over Mongolia in GPCPV31 fits well the gauge analysis than that of the previous GPCP, although the correlation coefficient slightly falls.
At first glance, it appears that rain gauge observation corrects the overestimation of satellite measurement but that the other products still overshoot APHRODITE, except for GPCPV31. Although APHRODITE and other rain gauge-based datasets are thought to approximate the truth, insufficient surface observation networks such as those in the high latitudes of Eurasia with complex terrain could cause substantial uncertainty due to the broad gaps between observation points.
3.2 Heavy precipitation event around Ulaanbataar a. Synoptic-scale atmospheric conditionTo further examine the differences among the GSMaP products and satellite radar measurements, we focused on a single heavy precipitation event in the vicinity of UB. On 9 July 2016, the GPM observatory captured heavy precipitation in the UB area. Figure 4a presents a composite map showing the surface precipitation rate estimated by GPM_KuPR and GMI/GPROF (The variables used here are precipRateE-Surface in GPM_KuPR and surfacePrecipitation in GMI/GPROF, respectively). The GPM observatory passed over UB between 23:57 UTC and 23:59 UTC, which corresponded to 07:57 and 07:59 Local Time of UB. A synoptic-scale atmospheric condition characterized by a geopotential height of 850 hPa indicated a cyclonic system that is located in the northeastern part of Mongolia (Fig. 4b). The precipitation occurred on a large horizontal gradient of potential temperature identified by northwestern cold air and eastern warm air (Figs. 4a, b). The vertically integrated moisture flux and precipitable water demonstrated that there were two streams of moisture transport (Fig. 4c): one was an abundant southwesterly moisture flux passing over a desert region in southern Mongolia, and the other was a northwesterly moisture flux that came from the boreal forest region. In this case, these features indicated that a large-scale atmospheric frontal system related to the synoptic disturbance was the dominant driver of the precipitation. The GPM core observatory passed just along the convergent zone of moisture fluxes.
a) Composite map of surface precipitation rate captured by GMI/GPROF and GPM_KuPR at 23:00 UTC on 9 July 2016 (orbit number 013432, mm h−1, shade). GPM_KuPR is over-written within the swath of GMI. Thick solid and long dashed lines indicate the swath coverages of GMI and GPM_KuPR, respectively. Thin dashed lines exhibit the swath time of GMI labeled by boxed digits, indicating a minute in 23:00 UTC; b) synoptic atmospheric condition for the geopotential height (m, contour) and potential temperature (K, shade) at 850 hPa at 00:00 UTC on 10 July 2016 derived by ERA5; c) Same as b) but for the precipitable water (kg m−2, shade) and vertically integrated water vapor flux (kg m−1 s−1, vector). The black star indicates the location of Ulaanbaatar.
Figures 5a and 5b present the precipitation pattern captured by GPM_KuPR, magnified around UB and Bagabayan, respectively. GPM_KuPR estimated a precipitation rate exceeding 50 mm h−1 at the surface of the mountains northeast of UB, not far from the Bagabayan rain gauge station (the topography is presented in Fig. 6a). Figures 5c and 5d present the vertical structures of the precipitation system depicting the precipRate in the GPM_KuPR product. Two lines have been drawn through Bagabayan's location. Line A–B corresponds to the direction in which the precipitation system was traveling, whereas line C–D runs parallel to the frontal system. In Figs. 5b and 5c, a precipitation rate exceeding 50 mm h−1 occurred in a narrow region on the windward side of the mountains. The vertical structure along with the frontal system (Fig. 5d) demonstrates a precipitation rate of approximately 1 mm h−1 at a height of nearly 8 km aloft of the lower-level heavy rainfall area northeast of Bagabayan. The GPM_KuPR algorithm involves a rain-type classification module that classifies rain into three main types (stratiform, convective, and other), as referenced by the vertical profile and horizontal distribution of the radar reflectivity factor, which are called V and H methods (Awaka et al. 2016). While a few pixels were classified as convective rain in this case, large parts of the precipitation were classified as stratiform rain (Figs. 5c, d).
a) Horizontal pattern of the surface precipitation rate (mm h−1) captured by GPM_KuPR magnified around Ulaanbaatar. The out-of-a-swath-width area of GPM_KuPR is shaded in gray. The contour lines indicate the topography identified by 1500 m. The black circle and black star indicate the locations of Bagabayan and Chinggis Khaan International Airport, respectively; b) same as a) but magnified around Bagabayan. Black thick lines correspond to the lines of A–B and C–D in Fig. 5a. Black circles indicate the locations of 1) windward, 2) hillside, and 3) near the mountain peak in Fig. 5c. Blue stars indicate the locations of α and β in Fig. 5d, respectively; c) vertical structure of the precipitation rate along a line A–B depicted in Fig. 5a. The black solid and black dotted lines indicate zero degree height and the clutter-free bottom, respectively (the variables used here are heightZeroDeg and binClutterFreeBottom in GPM_KuPR). The colored grid cells below the upper x-axis indicate the type of precipitation classified by the GPM_KuPR algorithm. The gray-shaded region at the bottom of the figure indicates the missing precipitation rate, which roughly corresponds to the shape of topography; d) same as c) but for line C–D in Fig. 5a.
a) Same as Fig. 5b but for the topography. The red thick lines indicate the width for an average of vertical profiles; b) averaged vertical profiles about Zm detected by GPM_KuPR and GPM_KaPR. Subscripted numbers indicate the profile location corresponding to the numbers in Figs. 5b and 6a. Horizontal lines indicate the zero degree height for each location. The averaged surface precipitation rates at each location estimated by GPM_KuPR and GPM_KaPR are summarized in the bottom left table; c) same as b) but for Ze; d) same as b) but for Dm using the DPR algorithm.
To elucidate the difference in the precipitation along the mountain slope, we show the average vertical profiles of the measured and attenuation-corrected radar reflectivity factors (Zm and Ze, called zFacotorMeasured and zFactorCorrected in the product, respectively) by both GPM_KuPR and KaPR (Fig. 6). Additionally, Fig. 6 presents the mean diameter of the precipitation parcel (Dm, which is stored in paramDSD in the product) as estimated by the DPR algorithm drop size distribution module (Iguchi et al. 2018; Seto et al. 2021). The vertical gradients of Zm and Ze from beneath the bright band to the surface provide additional information with regard to rain type. Kobayashi et al. (2018) demonstrated that a downward increase in Ze, which results from raindrop growth related to warmtype clouds, is frequently observed under the stratiform rain pixel over the North Pacific Ocean and East Asia. Porcacchia et al. (2019) also sought to detect the signatures of the reflectivity factor in DPR dominated in a collision-coalescence process focused on the contiguous United States. Following Porcacchia et al. (2019), the Zm of KuPR in the liquid layer tends to increase toward the surface when the collision–coalescence process dominates. The profiles in Fig. 6 were chosen at three locations: 1) the windward side, 2) the hillside, and 3) near the mountain peak presented in Figs. 5c and 6a. They have an average width of 55 km (11 beams) along the track of the GPM satellite, such as between α and β in Figs. 5d and 6a. The average procedure is only conducted if all 11 beams store no-missing data in each vertical level. The mean topographic heights at each location were roughly 1380, 1700, and 1930 m (the variable used here is the elevation in GPM_KuPR). The surface precipitation rate estimated by GPM_KuPR and KaPR indicated that the hillside of the mountain experienced a relatively intense precipitation compared with the others, reaching approximately 52.9 mm h−1 and 12.7 mm h−1, respectively (Fig. 6b).
The average vertical profi les of KuPR demonstrated that the melting layer, which was characterized by the peak of Zm and Ze, was located beneath the level of 0°C height (Figs. 6b, c). Such Zm and Ze shapes are similar to a typical stratiform rain situation. The height of the melting layer gradually increased from the windward side to the mountain peak, indicating that a cloud base lies along the mountain slope. This feature suggests that the low-level atmosphere was forced to rise by the orography and formed a cloud. Beneath the melting layer, the Zm of KaPR decreased with decreasing height, which means that KaPR suffered from large attenuation by rain particles at lower levels, especially on the hillsides and near the peaks of the mountain range (Zm2 and Zm3). Whereas the vertical gradients of the Zm of KuPR were nearly zero beneath the melting layer (Fig. 6b), those of the Ze of KuPR slightly increased toward the surface, especially on the hillsides (Ze2 in Fig. 6c).
The vertical distribution of Dm also significantly differed among locations (Fig. 6d); the large diameter for the entire level was estimated at the hillside (Dm2). Thus, the raindrop size might contribute to the intensifi cation of the precipitation rate. Dm2 and Dm3 also slightly increased toward the surface, which is refl ected in the profi les of Ze, especially in the liquid layer below the 0°C height. Such features lead us to expect that the growth of raindrops by the collision–coalescence process might have occurred in this case. The downward increase in the Zm of KuPR, which, according to Porcacchia et al. (2019), is one feature of a dominant collision–coalescence process, was not observed in our case (Fig. 6b).
c. Intercomparison of several precipitation estimatesThis section examines the reliability of precipitation estimates by several products compared with the observations. The time series of surface precipitation recorded by the Bagabayan station is presented in Fig. 7.
a) Time series of observed and estimated surface precipitation rate (mm h−1) at Bagabayan. Observation is calculated in 10-min intervals (OBS). GPM_KuPR and GMI/GPROF are a snapshot at 23:57 UTC and 23:58 UTC, respectively. The GSMaP products have 1-h precipitation rates. Line markers of GSMaP are drawn at the center of time coverage by 1-h precipitation rate of GSMaP (e.g., 1-h precipitation rate between 23:00 UTC and 00:00 UTC is shown by the marker located at 23:30 UTC). OBS_1h is the 1-h average of observed precipitation, which can compare to the precipitation rate of GSMaP. The pink vertical lines indicate that GSMaP installed micro wave radiometer, which does not correspond to the microwave radiometer's overpass time; b) same as a) but for WRF. WRF_1h is the 1-h average of precipitation estimated in 10-min intervals. The 1-h average window delays 30 min against that of Obs_1h in Fig. 7a (e.g., 1-h precipitation marked at 23:50 UTC is averaged between 23:30 UTC and 00:30 UTC, which corresponds to Obs_1h averaged between 23:00 UTC and 00:00 UTC).
In terms of sub-hourly precipitation, the rain gauge observed surface rainfall of approximately 9.6 mm h−1 between 23:50 UTC and 00:00 UTC on 10 July 2016 when the GPM core satellite observed over the study area. The GMI/GPROF and GPM_KuPR grid points nearest to Bagabayan estimated surface precipitation rates of 7.8 mm h−1 and 19.8 mm h−1, respectively. While the GMI/GPROF value was close to the in situ observation, the GPM_KuPR estimate was markedly larger. Contrary to GPM_KuPR, the horizontal pattern of the precipitation rate produced by GMI/GPROF mostly exhibited weak precipitation of less than 10 mm h−1 (Fig. 8a). This pattern followed that of the polarization-corrected brightness temperature at 89 GHz observed by GMI (GMI_PCT89, Fig. 8h), as precipitation retrieval uses the high-frequency band of GMI over the land region. The definition of PCT89 here follows Cecil and Chronis (2018). PCT89 exhibits a relatively lower temperature in the area northeast of Bagabayan. Thus, GMI measured high ice scattering at a higher level, especially around Bagabayan, and GMI/GPROF used PCT89 to estimate the surface precipitation rate. Conversely, precipitation retrieval by space-borne radar systems, such as GPM_KuPR, is regarded as superior to that achieved by microwave radiometers, such as GMI, particularly over land. The surface precipitation rate of GPM_KuPR is extrapolated from the upper-level precipitation rate above the clutter-free bottom (dotted lines in Figs. 5c, d). The heavy precipitation system detected by GPM_KuPR is somewhat reasonable for the area near Bagabayan, as the rain gauge recorded intense rainfall of about 20 mm h−1 before the passing of the GPM core observatory (Fig. 7a). In addition, WRF succeeded in evaluating a similar tendency and intensity of precipitation amounts to observed one (Fig. 7b), slightly delayed relative to the actual state (the peak of precipitation was delayed for about 30 min). The horizontal distribution by WRF also suggested intense precipitation around UB similar to that of GPM_KuPR (Fig. 8b). Thus, we expected that GMI/GPROF would potentially underestimate the snapshot precipitation rate, especially around the hillside of the mountain. A lower-atmospheric process, such as collision–coalescence, may cause a difference in precipitation intensity estimates between GMI/GPROF and GPM_KuPR.
Horizontal patterns of the surface precipitation rate (mm h−1) estimated by a) GMI/GPROF, b) WRF, c) GSMaP_MVK, d) GSMaP_MVKv6, e) WRF_1h, f) GSMaP_Gauge, and g) GSMaP_Gaugev6; h) same as a) but for PCT89 estimated by GMI (K). The range of the drawings is the same as in Fig. 5a. GMI/GPROF is a snapshot at the same time as in Fig. 4a. WRF indicates the precipitation rate calculated between 00:20 UTC and 00:30 UTC, and WRF_1h indicates the hourly precipitation estimated between 23:30 UTC and 00:30 UTC. GSMaP products show the 1-h precipitation rate between 23:30 UTC and 00:30 UTC. The solid and dotted lines exhibit the topography identified by 1500 m and the swath coverage of GPM_KuPR, respectively. The black circle and black star indicate the locations of Bagabayan and Chinggis Khaan International Airport, respectively.
In terms of hourly precipitation, the gauge observation recorded 12.2 mm between 23:00 UTC and 00:00 UTC (Fig. 7a). If we evaluate hourly precipitation by a snapshot precipitation intensity, the GMI/GPROF (7.8 mm h−1) explains about 64 % of the observed amount of hourly precipitation, whereas the observation (9.4 mm h−1) explains 77 % of oneself. In Fig. 7b, the WRF hourly precipitation (WRF_1h) well replicated that of observation as estimate at 13.0 mm; the window of 1-h precipitation was between 23:30 UTC and 00:30 UTC to fit the actual precipitation tendency. The WRF's snapshot precipitation rate displayed at 00:20 UTC, which was regarded as the corresponding time to the GMI/GPROF, was 5.5 mm h−1, which explains 42 % of hourly precipitation in oneself. Therefore, the snapshot precipitation intensity of GMI/GPROF also slightly underestimates if it is regarded as the amount of hourly precipitation in this case. However, the horizontal pattern of GMI/GPROF can be qualitatively representative of that hourly state of WRF, at least around Bagabayan (Figs. 8a, e).
Contrary to GMI/GPROF, GSMaP_MVK estimated a more extreme value (58.2 mm h−1) than the other GSMaP products, even though a GMI was installed (Fig. 7a). The horizontal pattern of GSMaP_MVK indicated that the extreme amounts of precipitation—over 50 mm h−1—mostly occurred on the surface of the mountain slope around UB (Fig. 8c). Conversely, the GSMaP_MVKv6 estimates were closer to the GMI/GPROF values (Fig. 8d). It means that the difference in the algorithms caused this extreme estimate of precipitation. It is possible that version 7 of the algorithm succeeded in detecting orographic rainfall in this case but considerably overestimated its amount. After the extreme precipitation period, GSMaP_MVK again indicated greater precipitation than did GSMaP_MVKv6 between 01:00 UTC and 02:00 UTC when the microwave measurement was performed (Fig. 7a).
Comparisons with observations revealed that GSMaP_MVKv6 and GSMaP_Gauge produced a relatively reliable precipitation rate around Bagabayan (Fig. 7a), at least in this case. Especially for GSMaP_Gauge, data from the routine observation site situated somewhat close to Bagabayan effectively moderated the extreme precipitation rate of GSMaP_MVK. Conversely, GSMaP_Gaugev6 indicated the lowest precipitation rate due to the over-calibration of GSMaP_MVKv6, whose estimates were near the observed values. Based on this narrow comparison, GSMaP_Gauge could be regarded as a better product than the other GSMaP products. However, the calibration by rain gauge observation significantly damped the horizontal pattern of precipitation (Figs. 8f, g). Over the mountains, the GSMaP_MVK estimates are mostly similar to the horizontal pattern of GMI_PCT89, whereas GSMaP_Gauges presents a much smoother picture, especially GSMaP_Gaugev6, which demonstrates a modest pattern. These results imply a loss of structure detail with rain gauge calibration.
d. Difference between gauged and ungauged GSMaP in 24-h accumulationFinally, we examined the accumulated precipitation. Figure 9a presents the 24-h accumulated precipitation from 18:00 UTC 09 July, as evaluated by WRF, which is regarded as replication of the precipitation system. The largest accumulated precipitation was observed in the northern part of the mountains rather than around UB. Thus, the heavy rainfall system around UB captured by GPM_KuPR (Fig. 5) was merely episodic. Figures 9b and 9e present the accumulated precipitation estimated by GSMaP_MVK and GSMaP_MVKv6. The precipitation patterns were nearly the same, except around UB, where GSMaP_MVK produced overestimates. Both products also indicated a relatively large amount of rainfall in the northern part of the mountains, similar to WRF in the spatial distribution. Although the estimates of the occurrence of extreme rainfall around UB appear to be unreliable, the large rainfall estimate in the northern part of the mountains could be considered consistent as it is similar to the WRF output. Conversely, the calibration by rain gauge observatories encompassing the mountains reduced the precipitation not only around UB but also in broad regions over the mountain (Figs. 9c, f). This may indicate that the sparse gauge observation network is unable to detect local precipitation in their gaps, especially over the mountain. As a result, the horizontal patterns of accumulated precipitation are much smoother than those of GSMaP_MVKs. In fact, the reduction of precipitation exceeded 15 mm in the northern part of the mountains (Figs. 9d, g), which corresponds to roughly half the GSMaP_MVK estimate. These results indicate that GSMaP_Gauge may be affected by overcorrection in the gaps between rain gauge observatories, particularly over the mountain regions, where the sparse observatories potentially miss localized precipitation.
24-h accumulated precipitation (mm) estimated by a) WRF and b) GSMaP_MVK, c) GSMaP_Gauge, and d) the anomaly of GSMaP_Gauge vs GSMaP_MVK; e), f) and g) are the same as b), c) and d) but for version 6 of the algorithm. The range of drawings is the same as in Fig. 1b. The contour line indicates the topography identified by 1500 m. The dotted lines in a), b), c), e), and f) show the swath coverage of GPM_KuPR. The gray-shaded boxes in c), d), f), and g) are grid cells that have more than one rain gauge station in the CPC gauge analysis. The black star indicates the location of Chinggis Khaan International Airport.
An intercomparison of the GPM_KuPR, GMI/GPROF, and GSMaP products for the case described in this study suggests that the precipitation estimates of GSMaP products are influenced by uncertainties caused by both differences in the algorithms as well as the use of rain gauge calibration, especially over remote mountainous regions. The former uncertainty may be linked to estimated rain rates over a delineated orographic area, even where the latest algorithm succeeds in detecting orographic rainfall. The latter is likely the result of gaps in ground-based observatories. These uncertainties can frequently occur and contribute to the estimates of the total precipitation in a specific region, as they are strongly related to the geographic location.
To demonstrate other cases, we compared the hourly precipitation observed at Bagabayan and the estimates of the various GSMaP products over the entire 2016 summer season. Figure 10 presents scatter diagrams of the results of the observations and GSMaP products. The selected cases satisfied two requirements: 1) GSMaP installed microwave radiometer measurement and 2) observation or GSMaP products recorded over 0.1 mm h−1. The total precipitation for all cases observed at the Bagabayan station was 168 mm, as compared with the GSMaP_MVK and GSMaP_MVKv6 estimates of 349.7 mm and 229.5 mm, respectively. The overestimation by GSMaP_MVK was largely due to two extreme cases where roughly 40 mm h−1 and 60 mm h−1 were reported (Fig. 10a); the latter was the case described in this study. In both cases, GSMaP used GMI measurement, but it is unclear whether such overestimation solely depended on GMI because the sampling number is very small (only one summer season and one grid point). At least in this comparison, the precipitation rate evaluated by version 7 of the algorithm tended to be considerably overestimated than those of version 6 of the algorithm. Although differences in the algorithm results might be negligible for national-scale precipitation (Fig. 3), such extreme precipitation rates could well lead to a signifi cant error in the estimates of the total rainfall in a small area. Rain gauge calibration moderated these errors, at least for Bagabayan, but produced a slight over-calibration (Fig. 10b).
a) Scatter diagram of the hourly precipitation observed at Bagabayan and estimated by GSMaP_MVKs between June and August 2016. The red star and blue circle indicate versions 6 and 7, respectively. The filled circle indicates that the installed microwave radiometer in GSMaP is GMI. The amount of accumulated precipitation (mm) by Obs, MVK, and MVKv6 are demonstrated inside the diagram. The scatter diagram magnified by the axis range of 9 mm h−1 is embedded at the top right in the diagram; b) same as a) but for GSMaP_Gauges.
To evaluate these errors for other locations, we investigated each grid of GSMaP around UB, focusing on the 2014 and 2017 summer seasons. Figure 11a presents the frequency with which GSMaP_MVK with microwave measurements was recorded over 1 mm h−1 in the hourly products. A relatively large number of precipitation events were estimated at the mountain northeast of Bagabayan. The root-mean-squared error (RMSE) for GSMaP_MVK and GSMaP_MVKv6 in this northeastern mountain area, which was calculated using all events for each grid, was large (Fig. 11b). This indicates a signifi cant difference between the products in this area. Conversely, while a large number of precipitation events was also recorded at the mountain located at approximately 51°N and 102°E (Fig. 11a), the RMSE was smaller than at the northeastern mountain (Fig. 11b). Thus, the mountain northeast of Bagabayan was regarded as one of the sensitive areas, likely due to the differences in the algorithm.
a) Map of the frequency with which the hourly product of GSMaP_MVK estimated a precipitation rate of more than 1 mm h−1 during the summer seasons between 2014 and 2017. The gray-shaded boxes are grid cells with more than one rain gauge station in the CPC gauge analysis. The contour line indicates the topography identifi ed by 1500 m; b) same as a) but for RMSE for hourly precipitation rate between GSMaP_MVK and GSMaP_MVKv6; c) same as a) but for the ratio of the average precipitation rate of GSMaP_MVK to the average rate of GSMaP_ MVKv6; d) same as c) but for GSMaP_Gauge and GSMaP_MVK.
The ratio of the mean precipitation rate of GSMaP_MVK to the mean rate of GSMaP_MVKv6 demonstrated that the upgrade of the algorithm from version 6 to version 7 generally increased the estimated precipitation rate at high latitudes and in mountain areas (Fig. 11c). Based on the results reported in this paper, the increased precipitation rate estimated by version 7 of the algorithm did not always mean an improvement in the estimate, especially at higher latitudes, where precipitation is less than that at lower latitudes. Rain gauge calibration decreased the mean precipitation rate of GSMaP_MVK by more than half everywhere (Fig. 11d). More specifically, while the decrease in the rate near the grids of the observatory tended to be suppressed by roughly 50 %, the gaps between observatories were influenced by the slightly strong attenuation. The gauge calibration algorithm in GSMaP considers the reliability of the CPC data by counting the number of rain gauges within the 7 × 7 grid area around each grid; subsequently, it applies a weight to the CPC rainfall rate (Mega et al. 2018). However, this approach would likely evaluate nearly all of the grids in Fig. 11 as reliable. Thus, a rainfall rate stored in non-gauge grids can be introduced to the calibration even if the CPC data involves underestimates. The strong attenuation in the southern desert region, where there is less precipitation, is perhaps reliable, but over the mountain and boreal forest regions at high latitudes, it could result in over-calibration. In such a high-latitude region, GPCPV31 adopts the relatively high weight of the gauge analysis to merge satellite-only precipitation estimates (Fig. 2i). Thus, the GPCPV31 may also be affected by the calibration error due to the gauge analysis that potentially misses the precipitation in the gap of observatories.
The present study provided two analyses. The first compared the performance of gauged and ungauged GSMaP (GSMaP_Gauge and GSMaP_MVK) with several other precipitation products that estimate summer precipitation in eastern Eurasia and the Mongolian territory to judge their climatic assessment capability. The second analysis focused on a case study of extreme precipitation events observed in July 2016 in the vicinity of UB in Mongolia. It was found that the precipitation estimates of the GSMaP products were influenced by the uncertainties associated with the different algorithm versions and by the use of rain gauge calibration.
For national-scale precipitation, while the impact of algorithm differences is generally negligible, the use of rain gauge calibration can effectively correct the tendency of satellite-only GSMaP products to overestimate precipitation. Based on the results from several datasets, GSMaP_Gauge estimates were found to be comparable with other precipitation estimates. However, gauge-corrected satellite precipitation and atmospheric reanalysis, excepted for GPCPV31, tended to record higher amounts of precipitation than the rain gauge-based APHRODITE, although the variability of the measurements was nearly the same.
Several implications emerge from the case study. We found that GSMaP_Gauge may overcorrect over a mountain region encompassed by the surface observation network. Moreover, it appeared that the gridded analysis of rain gauge observation potentially underestimates the actual precipitation over mountainous areas, especially in the gaps of the observation network. Thus, as Mongolia has a complex terrain and a sparse observation network, the reported amounts of precipitation should be treated with some degree of skepticism, even in cases where the analysis is based on rain gauge observations. The uncertainty of rain gauge-based analysis resulting from sparse observational networks is likely to limit the accuracy of satellite-based precipitation estimates.
When the focus shifts to smaller regions, the uncertainty due to the different algorithm versions significantly increases. Version 7 of the algorithm in GSMaP_MVK enables to obtain better estimates for cases over the TRMM observation areas owing to the improvement of the orographic/nonorographic scheme relative to version 6 (Shige et al. 2013; Taniguchi et al. 2013; Yamamoto and Shige 2015; Yamamoto et al. 2017). Moreover, it succeed in replicating the signature of intense rainfall along the mountain slope captured by the precipitation radar of GPM. The evaluated precipitation amount over the delineated orographic area was, however, considerably amplified. Such extreme values can produce significant errors in the estimate of the total amount of precipitation in a specific region where rainfall is relatively low, such as at higher latitudes. GSMaP_Gauge is able to correct such unreliable extreme rainfall estimates. However, the tradeoff is a loss of the horizontal distribution of precipitation due to the large gaps between observatories, especially over a mountain. In brief, GSMaP users should carefully select the GSMaP products and their algorithm versions to fit their particular scientific scope.
The algorithm versions of GPM DPR should also be carefully selected. This study used the product compiled by the algorithm version 05A, but that of algorithm version 06A has been released during our analyses. Iguchi et al. (2018) observed that the DPR and KuPR rain estimates in version 06 agree better with ground validation data over the United States, and the rain classification algorithm improved. Comparing between 05A and 06A for the case under this study, we found that the surface precipitation rate over Bagabayan estimated by KuPR of 06A exhibited better correspondence with those of the rain gauge and GMI/GPROF than that of 05A (9.0 mm h−1 in 06A and 19.8 mm h−1 in 05A). The horizontal and vertical structures characterized by the extreme precipitation rate in Fig. 5 are to be modest in 06A (Fig. S1), which looks more reliable than 05A.
The results of the present study are based on simple comparisons and a single case study. This means that they do not establish in statistical terms the general features of the error component of GSMaP for the Mongolian territory. An intensive observation campaign aimed at filling the gaps in the observation network would serve to improve the GSMaP algorithm, gauge calibration, and performance of the gaugebased analysis in Mongolia. We hope that the results reported here will be helpful for the developers of the GSMaP algorithm.
With regard to precipitation monitoring, the availability of the standard product of GSMaP, including GSMaP_Gauge, remains limited by the 3-day latency in the product release. Today, satellite-based precipitation products are expected to provide real-time monitoring for disaster prevention and/or hydrological assessment. To address the societal demand for more immediate information, JAXA has newly provided a near-real-time version of GSMaP and a bias-adjusted variation (GSMaP_NRT and GSMaP_Gauge_NRT, respectively), with the latency period shortened to 4 h. The error calibration of GSMaP_Gauge_NRT uses parameters based on a GSMaP_Gauge historical database covering the previous 30 days. A very recent preliminary study reported that, while GSMaP_Gauge_NRT effectively reduces the bias of GSMaP_NRT, its ability to accurately detect precipitation over the Chinese mainland did not significantly improve (Lu and Yong 2020). It is quite possible that the historical parameter for bias correction is affected by the quality of the original rain gauge analysis. To better respond to societal demands, future validation studies should focus on the accuracy or performance of such near-real-time versions as well as the standard products.
Figure S1: Same as Fig. 5 but for GPM_KuPR compiled by the algorithm version 06A.
This study was funded in part by JAXA PMM 8th RA (PI No. 208) and JAXA EO-RA2 (PI No. ER2GPF 009), ALRC, Tottori University (No. 30F2003), JSPS KAKENHI Grant Number 19H05668 and 19H00556, and the Arctic Challenge for Sustainability II (ArCS II), Program Grant Number JPMXD1420318865. GPCP Precipitation data and CPC Global Unified Precipitation data provided by the NOAA/OAR/ESRL PSL, Boulder, Colorado, USA, from their Web site at https://psl.noaa.gov/. GPCP version 3.1 is obtained from https://disc.gsfc.nasa.gov/datasets/GPCPMON_3.1. APHRODITE is provided from http://aphrodite.st.hirosaki-u.ac.jp/. The products of GSMaP and GPM were provided by JAXA. The authors wish to thank the editor and the two anonymous reviewers for their valuable comments.
