Evaluation of GSMap (Global Satellite Mapping of Precipitation) with the reference to rain gauge observations in Banjarbaru City, Indonesia

Climate change as one of the essential impacts of global warming in the last few decades, has resulted many disasters in different ways, i.e., floods and landslides, as the effect of extreme rainfall events. At the end of February 2023, a severe flood event occurred in Banjarbaru City, resulted an inundation in many areas. A flood control system is recommended to face this problem. In this research, the flood control presented is a non-structural method by analyzing rainfall data based on field data and GSMaP satellite-based rainfall data. This research aims to determine the type of correlation coefficient relationship between BMKG data and GSMaP satellite-based rainfall data. The data collection includes primary data in the form of field surveys to identify rain events and secondary data obtained from GSMap satellite-based rainfall data and field rainfall data obtained from BMKG online. The Melchior design flood calculation method is used for calculating a flood-designed discharge. The results show that GSMaP data have significant values that correlate to rain gauge observations from BMKG stations in Banjarbaru City. An enough to high correlations values indicate that this GSMaP can be used as an alternative source of rainfall data for hydrological analysis, in particular to ungauged basin.


Introduction
Floods are inundations of dry land, such as agricultural fields, settlements, and city centers.Many factors cause flooding in downtown areas, including high rainfall, water discharge that exceeds drainage capacity, limited water catchment areas due to conversion into residential areas, people who throw garbage into rivers and cause the river flow to be blocked, illegal logging, drainage overflow, and many other factors [1].Based on the data from the South Kalimantan Provincial Government in 2023, at least 782 flood events have been recorded from 2018 to 2021.According to Kompas.com, three sub-districts in Banjarbaru were flooded: Cempaka, South Banjarbaru, and Landasan Ulin.The high intensity of rain caused the Kemuning River to overflow, causing losses to more than a thousand people.
Preventive measures must be taken to reduce the number of flood events.Several studies that have been conducted are expected to provide accurate data and analysis of the significance of these actions.The lack of observation stations, data series that are not long enough, and the uneven distribution of these stations are the biggest obstacles in analyzing rainfall data.Moreover, rainfall data has high spatial and temporal variability [2].With the development of remote sensing technology such as radar and satellites, it is possible to measure rainfall more accurately because this technology can reach places that conventional rainfall gauges cannot reach, and then rainfall can be monitored on a large scale [3] IOP Publishing doi:10.1088/1755-1315/1343/1/012005 2 One of the initiatives that uses observations from several sensors as input to determine rainfall rates is the Global Satellite Mapping of Precipitation (GSMaP), initiated in 2002 [4].GSMaP is a blended satellite-based precipitation product combining available passive microwave sensors and infrared radiometers [5].The advantage of GSMaP is that it has a spatial resolution that reaches 0.1ºx0.1ºor the equivalent of 11.06 x 11.06 km, so that rain data can be obtained in all regions of Indonesia which have high diversity.The use of GSMaP rainfall data for the Indonesian region is very profitable, seeing Indonesia's vast territory and the varied rainfall patterns in various parts of Indonesia.[8] Despite being extensively used in meteorology and hydrology research, GSMaP rainfall data are still has a little use in South Kalimantan Province.This data has been applied by Helda and Wijayanto (2022) in the Identification of Rain Events Using GSMaP Satellite Data in Banjarbaru for eight months data and one control point.Hence, this research is a follow-up study where the research is conducted with more data and control points in order to convey the vast rainfall variability, in space and time.
The results to be achieved are the availability of information on hydrological conditions, including the identification of extreme rainfall, rainfall period that causes flooding, discharge period that causes flooding, and the extent of the reliability of GSMaP satellite-based rainfall data in identifying rainfall events.

Study Area
The research location is in Banjarbaru, the provincial capital of South Kalimantan.It includes seven ( 7

Figure 1. Research Location
Daily rainfall data is downloaded via the BMKG website Online from two BMKG stations in Banjarbaru city for the Rainfall periods of January 2023 to June 2023 as well as GSMaP satellite data, covering the same time step.Visual observation data for the same period is used to ensure data validity.The timing of rainy periods is based on the possible frequency of rain events and the amount of rain.
GSMaP satellite rainfall data is downloaded via JAXA Global Rainfall Watch (GSMaP) (https://sharaku.eorc.jaxa.jp/GSMaP)with the same rainfall periods.They refer to the coordinates of two BMKG stations.As well as five (5) other coordinate points as control points.GSMaP data has high spatial (0.10 o ) and temporal (hours) resolution with FTP (File Transfer Protocol) format.An initial registration is required on the GSMaP website to download the data (Sub-Division of Weather Satellite Imagery Management, 2017).The visual observation period was carried out from January 1 st, 2023, to June 30th, 2023, along with the observed rainfall data.Figure 2 is a direct inspection at the location.In measuring the linear association between satellite and gauge measurements, the Correlation coefficient (CC) is used: Where CC is the correlation coefficient, Si denotes a satellite estimate, Smean represents the average of estimate values, Gi is a rain gauge measurement, Gmean represents the average of gauge measurements, and n is the number of data pairs.Correlation analysis is a statistical test that measures the relationship density between the two variables [8].The correlation coefficient gauges the strength of the relation between the two variables as shown in Table 1.2) to evaluate the capabilities of satellites in detect rain events, as follows: In this research, the thresholds used as the boundaries between rain and no rain condition is 0.5 mm/day.Furthermore, for category performance analysis, the calculations are conducted with some parameters which include Probability of Detection (POD), False Alarm Ratio (FAR), Critical Success Index (CSI), Heidkle Skill Score (HSS).POD will determine the level of accuracy (Hit rates) of satellite data, when the rain event identified by BMKG stations are correctly detected by the satellite.FAR indicates the detection of improper rain events when there is no rain recorded in the BMKG rainfall station, while satellite detected rain.CSI combines POD and FAR to obtain more comprehensive results.HSS eliminates detection which is true due to the presence of random coincidence conditions (random chance).

Categorical Performance Measures. Statistical analysis based on Performance category refers to the contingency table (Table
The calculation method of performance category analysis are as follows: 1. POD (Probability of Detection)  =  ( + ) (5)

HSS (Heidkle Skill Score)
Where a is the frequency of rain that occurs and detected, b is the frequency of rain that does not occur but detected, c is the frequency of rain that occurs but does not detect and d is the frequency of rain that does not occur and undetectable.

Continuous Performance Measures
Results of the calculation of statistical analysis of monthly rainfall data from BMKG stations with GSMaP can be seen in Table 3.It can be seen from Table 3 that GSMaP rainfall data has an enough to high correlation with the rainfall data from the BMKG stations (quite high CC), as referred to Table 1.MAE values and RMSE values indicate the accuracy of GSMaP satellite data or levels of an error occurred.The smaller the value, the better in accuracy.The best MAE and RMSE values are zero.It can be seen that GSMaP satellite data more accurately predict rain events at Syamsudin Noor BMKG station compared to BMKG station at Trikora.The NSE values show how well the predicted value is against observations, indicates the performance of GSMaP data.A negative NSE indicates that the data performance is still poor, so it must go through a calibration and validation process for further use.Positive values of rBias show general GSMaP overestimate BMKG station data, which is shown GSMaP data at Syamsudin Noor.On the contrary, GSMaP data at Trikora underestimate the BMKG station data.The next discussion is about the satellite category performance.Before performance analysis by category is done, the calculation of the number of values is first carried out by using performance category which are represented by a, b, c, and d. as shown in Table 4. Performance analysis results by category further can be seen in Table 5.The values of POD, FAR, CSI, FBI, and HSS are able to provide some assessments of GSMaP satellite data quality.POD and CSI values are expected to range from 0-1, The FAR ranges from 1-0, and the HSS is expected to be ~-1.Table 5 shows that GSMaP satellite data performance still needs to be corrected by correction factors from the data with more extended data series.

Correlation of Observation Data and Satellite Data
Data distribution between the BMKG climate station and GSMaP can be seen in Figures 1 (Trikora  Staklim) and 2 (Staklim Syamsudin Noor).Scatter plot graph correlates BMKG station rainfall data and GSMaP rainfall data.Data from BMKG stations and GSMaP data were processed into daily maximum rainfall data and plotted according to the calculation.G Trikora represents BMKG Trikora data, while S Trikora represents Satellite data for Trikora location/point.
Following Figure 3 shows some regression models of the rainfall data comparison of BMKG and GSMaP data for January to June 2023.From the existing regression models, satellite rainfall data are corrected using the regression model equation of highest R 2 values.The value of R 2 (R-squared) is also known as Coefficient of Determination.In a linear regression, R 2 plays important role in understanding how well the independent variables explain the dependent variable.In this case, the independent variable is the BMKG rainfall data and the dependent variable is GSMaP rainfall data.The R 2 values are also crucial to assess the goodness of fit of a model, i.e., the adequacy of a model.Basically, R 2 values are ranging from 0 to 1; a higher coefficient of determination is expected to be achieved in a linear regression to indicate a fairly good model (R 2 > 0.6).Therefore, an R 2 value approaching 1 denotes a very good model.
A recapitulation of the regression model of BMKG stations and GSMaP rainfall data can be seen in Table 6 on the next page.Tabel 7 represents the maximum daily rainfall data before and after correction using the best linear regression model.
) observation points, including two (2) BMKG stations in the city of Banjarbaru, namely Staklim Syamsudin Noor (3.442 O S and 114.754O E) and Staklim Banjarbaru (3.44265° S and 114.84°E).Direct rainfall observations in five different locations were carried out only by visual observation of rainfall occurrences without any equipments.The goal is to record the time of rainfall events and the intensity level of the rainfall.There are five control points (as shown by red triangles) and two BMKG stations (as shown by green circles), namely BMKG Trikora and BMKG Syamsudin Noor. Figure 1 shows the research location and the observation points.It covers an area of 233,370 Km 2 .

Figure 2 .
Figure 2. Direct Inspection at the location 2.2.Performance of Precipitation Estimation Analysis 2.2.1.Statistical Analysis of Continuous Performance Measurements.The satellite product and rain gauge time series were compared to assess their accuracies and relationships using numerous statistical indices as follows: a) Degree of Linear AssociationIn measuring the linear association between satellite and gauge measurements, the Correlation coefficient (CC) is used:

Table 2 .
Probability threshold values to compare GSMaP data with BMKG data

Table 1 .
[9]e of Correlation Coefficient Relationship[9]Some quantification of discrepancies between satellite estimates and gauge measurements are Root Mean Square Error (RMSE), percent of bias, and Nash Sutcliffe Efficiency (NSE) as follows:

Table 3 .
Calculation of Monthly Statistical Analysis

Table 4 .
The amount of data for each performance category class of BMKG Climate Station