Evaluation of Near-Real-Time GSMaP precipitation estimate over Yankul region, Sultanate of Oman

Global precipitation data is very important for climate change research, and real-time precipitation estimates derived from satellites are attractive for a wide range of applications, such as extreme precipitation monitoring and natural hazard warning. The precipitation in the rain gauge represents the point distribution of the precipitation, which would lead to inaccurate results when calculating the average precipitation in a region with statistical methods. This document verifies the satellite-based precipitation data GSMaP_NRT with the precipitation amounts measured in the Yankul region of Oman. The results showed that the GSMaP_NRT precipitation data have high precision and a good correlation with the precipitation observed in the region. Furthermore, the results of these analyzes indicate that the GSMaP_NRT data have lower values than the observed data and can be significantly underestimated with heavy rains.


Introduction
The amount of precipitation and its spatial distribution are important for assessing water resources and forecasting floods. In many developing countries such as Oman, where the topography is rugged and the availability of ground monitoring stations is limited and unevenly distributed, making water resource assessment and flood prediction difficult [1]. Global Satellite Mapping of Precipitation (GSMaP) [2] provides information on the occurrence, amount and distribution of precipitation in these data-poor regions. GSMaP data has a high resolution satellite product with finer temporal (hourly and daily) and spatial (0.1 o ) resolution that provides effective and inexpensive means of calculating area precipitation in poorly measured regions in order to provide terrestrial data for water resources management and current time to supplement flood forecasts.
As satellite data, infrared and microwave satellite products such as the Global Satellite Precipitation Mapping (GSMaP) as a combination of various satellite precipitation data could be used to derive large-scale precipitation estimates over a global area [3]. The GSMaP rain product is based on the use of four satellite microwave radiometers in combination with geoinfrared radiometer data to achieve 0.1-degree spatial resolution [4]. There are several types ________________ *Corresponding author: osamaragab241@gmail.com of GSMaP rain products. In this document, GSMaP_NRT (Near Real Time) was used because its data caused flooding in the Yankul region.
GSMaP_NRT is an hourly raster product with a spatial resolution of 0.1 o × 0.1 o latitude / longitude, which has been available worldwide since 2003. Verification of this global product has been carried out in Japan and several other regions, but not yet through Oman.
Although satellite data are estimates and are subject to varying degrees of error depending on the sensor used, the region being examined, and other factors, the estimates can be used because they have two advantages over precipitation measurement stations: (i) greater success rate and easy availability of information, which is mainly intended for use in flood protection and warning systems; (ii) Generation of information in the form of a large-scale precipitation field [5].
Several studies in different regions such as Africa, South America, Japan, Australia and the United States have evaluated the precision of satellite-based precipitation products [6], [7], [8]. They found that satellite data generally underestimated rainfall. Much less is known about the reliability of satellite-based precipitation in the study region.
However, more studies are needed to evaluate the quality and limits of the precipitation data from satellites, in order to differentiate and quantify their uncertainties for the proper use of these products in each study area [9], [10], [11].
The paper used various precipitation precision estimation indices to estimate the precision of the satellite-based precipitation product GSMaP_NRT over the Yankul region of Oman. The central research question addressed in this paper is: How does GSMaP_NRT work in the Yankul region of Oman compared to the observed measurement data.

Study area
The Yankul area in the Sultanate of Oman (

Data and methodology
The rainfall observed in this study comes from 4 stations in Yankul province, namely, WUQBAH (Gauge 1), MERI (Gauge 2), SAYAA (Gauge 3) and YANQUL (Gauge 4) as shown in Fig. 1 These rain gauges are listed in Table 1. Precipitation data is recorded daily from April 24, 2013 to April 30, 2013 in the event of a flood by the Ministry of Regional Municipal and Water Resources. Recorded precipitation data is collected daily at each rain gauge. GSMaP_NRT precipitation data with 0.1 o grid resolution have a fixed shape and each file records a 1-hour observation result. The same days are chosen as the estimation period. Furthermore, the same coordinates of the observed data are selected for recording from the GSMaP data. Since the stations cover less area than the YANKUL region, the IDW method in QGIS is used here to record the spatial distribution of daily precipitation in the study area. This method is used for observed data and GSMaP_NRT data to perform spatial comparison.
The GSMaP data should not only be compared with the data observed in the region, but also with the daily precipitation of each station to demonstrate the reliability of the GSMaP_NRT data from several aspects. According to the given station names and their latitude and longitude, it is necessary to find the neighboring grid from the GSMaP observation files and collect the grid data according to time in each file to calculate the daily precipitation from the GSMaP satellite. Qualitative and quantitative validations were carried out as follows. The qualitative method consists of measuring the correspondence between the value of the estimates and the observations. To quantify the correspondence value, the following five statistical indices were used [12], the relative bias (B), the mean error (E), the Nash Sutcliffe (CNS), the root mean square error (RMSE), the coefficient correlation (R). These indices are given by the following equations.
Where, N is the total number of the rain gauge data or GSMaP data; Gi is the satellite estimates and Pi is the rain gauge observation values. The other validation statistic is the quantitative method that evaluates the GSMaP precipitation detection capacity, a 2x2 contingency table of events yes / no with rain / without rain is also used as shown in Table 2, "Hit" (a) represents correctly estimated rain events, "false alarms" (b) represent when rain has been estimated but has not occurred, "missing" (c) represents when rain was not estimated but did occur, and "correct negatives" (d) represent correctly estimated rainless events. The POD (Probability of Detection), which measures the proportion of correctly diagnosed observed events, the False Alarm Ration (FAR), which indicates the proportion of diagnosed events that were found incorrect, the Heidke Skill Score (HSS), which shows the Precipitation Detection Accuracy measures satellite estimates relative to random number matches, and the CSI (Critical Success Index), which indicates the value of warnings, is calculated as follows to assess the precision of the GSMaP_NRT precipitation data [13], [14].   Figure 2 shows the relationship of daily precipitation between the observed data and GSMaP_NRT in the study area. In Fig. 2 it can be seen that the GSMaP_NRT precipitation data is sometimes a little smaller and sometimes a little higher than observed, but the total precipitation in the same precipitation event is approximately the same for both data.

Results and discussion
The correlation between the observed precipitation and the GSMaP_NRT precipitation data is shown in Fig. 3. From Fig. 3 it can be concluded that the two-time series generally have a linear relationship and that the correlation coefficient is 0.694.
The GSMaP_NRT validation statistics are shown in Table 3. Overall, the GSMaP_NRT precipitation was approximately equal to the precipitation from the gauge data: the total precipitation in the study event from the gauge data was 80 mm, while the total precipitation of GSMaP_NRT was 78.4602 mm. The GSMaP_NRT data in the study area strongly correspond to the data from the rain gauge (r = 0.694), with the bias value was -1.92%. Moreover, the variance (-0.22 mm/day and 7.74 mm/day, respectively). The greater difference between them, the greater the variance in the individual errors in the sample.
Also, the consistency of GSMaP_NRT to measure the amount of precipitation can be described by the CN index. The CN index of the study area was 0.46 (46%), which means that GSMaP_NRT has the consistency to measure precipitation of around 46%. The good correlation coefficient, however, shows the qualified correlativity of the observed precipitation data and GSMaP_NRT on the regional scale.
PODs of GSMaP_NRT are close to 100% and FAR is generally small (4.2%). The HSS statistics show that the GSMaP_NRT estimates have a relatively good ability to detect the occurrence of rain (86.8%). As a result, the CSI critical success rate was around 0.958, indicating that the current GSMaP near-real-time calibration algorithm does not require any further improvement in data quality in the study area.  From the comparison above, GSMaP_NRT rainfall data has a lower precision for single station than for the region. Therefore, GSMaP_NRT satellite data is more suited for rainfall observation of region scale.   Table 3. Comparison of average precipitation in the region

Conclusion
In this study, GSMaP_NRT was evaluated against soil precipitation data during 2013 rainy season in the Yankul region of the Sultante of Oman. In addition to the evaluation based on regional and daily scales, the ability of these satellite data to record precipitation on rainy days in the study area was also taken into account.
The results show that the GSMaP_NRT data has a good correlation with the data observed by the stations in the Yankul region, and the GSMaP_NRT precipitation data can be used to study hydrological simulation and water resources after systematic calibration. The GSMaP_NRT data is reliable and has a higher precision in the study area with a correlation coefficient of 0.69.
It is recommended to consider more flood events in this area that will improve the evaluation of GSMaP datasets.