Application of Tropical Rainfall Measuring Mission (TRMM) Data for Flood Estimation in Lack Data Catchment

Providing real-time hydrological data for the design, development and management of water resources is a fundamental issue that is still being faced by hydrologists and practitioners. This data limitation is not only caused by the uneven distribution of hydrological measuring instruments, but also due to the inability of measuring instruments to present the data in real time. This paper aims to evaluate hydrological data, especially spatial-based rainfall data known as Tropical Rainfall Measuring Mission (TRMM) data as an alternative to rainfall data to predict flood hydrographs in a catchment with limited data. The study was initiated by selecting pairs of automatic rainfall data (ARG) and water level (AWLR) and combined with TRMM rainfall data for the same time series. These two types of rainfall data are then used as input for the HEC-HMS hydrological model involving a number of catchment parameters, such as: basin area, stream network, land use/land cover (LULC), soil characteristics and several other parameters. Three analytical methods are assigned to determine the shape of the hydrograph which include: Snyder UH (transformation method), SCS-CN (loss method), and recession (baseflow method). Discharge data converted from water level data is input at the catchment outlet as a reference for setting calibration parameters which are evaluated with the RMSE error indicator. The simulation results show that the predicted discharge deviation from TRMM rainfall data is higher than the predicted discharge from ARG rainfall data with RMSE 0.979 and 1.731 respectively. However, basically this TRMM data can be an alternative data by validating it with ARG rain data to be applied to lack of catchment data.


Introduction
Preparation of hydrological data is the most important part in the planning, design, management and engineering of water resources [1].The data relates to the determination of discharge which will be a reference for the utilization of water resources and the determination of the dimensions of the hydraulic structure [2].The limited data can lead to inaccurate decision-making in the management of water resources.Often the mistake in determining the magnitude has an impact on the risk of failure which is expressed by the safety factor: the safety factor is too high or too low [3][4].
There are many kinds of important hydrological data relating to the design and management of water resources, two of which are rainfall data and discharge data.Rainfall data is a substitute data if discharge data in a river is not available [5][6].This means that discharge data can be obtained by transforming rainfall data through hydrological modeling.However, often these two types of data are not available in an adequate range or even not available at all.Therefore, alternative fulfillment of data, especially rainfall data, is the recommended choice [7][8].
IOP Publishing doi:10.1088/1755-1315/1343/1/012003 2 Tropical Rainfall Measuring Mission (TRMM) rain data is one type of alternative data that has the opportunity to be applied.This NASA satellite-based spatial data is a record of potential rainfall in the atmosphere that covers all areas on the earth's surface.This data has been used for various purposes, especially for forecasting rainfall in an area.Several researchers have published the results of their studies both in conference proceedings and in journals.Adler et al. (2000) determine the distribution of tropical rainfall by consolidating TRMM rainfall data and observational rainfall.The combination of these two types of data is intended to increase the density of data distribution in the area studied [9].Detailed information is not available regarding the results of the study regarding trend compatibility between TRMM rainfall data and observational rainfall data.A similar study was conducted by Assyakur et al. (2013) who studied rainfall variability in Indonesia based on TRMM rainfall data.The monthly rainfall data recorded by the TRMM satellite for 13 years are checked for statistical parameters and then compared with the statistical parameters of observed rainfall data as a representation of rainfall variability [10].The results of the study show a trend of similar variability between the two types of data.The results of the analysis also imply that the spatial patterns of Indonesian rainfall are very clear, clearly affect by oceans, islands, monsoons, and topography.On the same topic, Kuswanto et al. (2019) clustered rainfall patterns in Indonesia using TRMM satellite rainfall data.A statistical test was performed to evaluate the significant bias of TRMM.The results of a study that classifies rainfall patterns in Indonesia into three main clusters: monsoonal, anti-monsoonal and semi-monsoonal show that TRMM rainfall data is a reasonable choice for clustering rainfall pattern [11].This also proves that the TRMM rainfall data is sufficient to be used as an alternative data for clustering rainfall patterns.
Studies related to flood prediction using input TRMM rainfall data have also been carried out by several researchers.Harris et al. (2017) evaluated the use of TRMM rainfall data for hydrograph-based flood forecasting [12].Hydrographs produced from several types of hydrological models were compared with observational hydrographs, which showed similarities even though some events indicated deviations.Furthermore, Collischonn et al. (2018) performed hydrographic modeling using TRMM input data in the Tapajós river basin, one of the major tributaries of the Amazon basin [13].The results of the study show that hydrographs generated from TRMM rainfall data indicate similarities with hydrographs transformed from observational rainfall data.
Further studies found in publications do not discuss much about the use of TRMM rainfall data for hydrographic modeling, especially in tropical regions such as Indonesia.As a region with a tropical climate, rainfall variability in Indonesia is very high as indicated by high fluctuations in rainfall intensity, rainfall distribution and rainfall duration.Often also the beginning of the rainy season can shift erratically.The high variability of rainfall in Indonesia is strongly influenced by various factors as stated by As-syakur et al. (2013), which is not only affected by monsoons but also by oceans, islands and topography [10 Therefore this study is important regarding the opportunity to use rain data in tropical regions such as Indonesia for flood prediction.

Study Area
The location of this research is in Sambo Catchment with an area of around 45.56 km 2 .This catchment is one of the sub-catchments of the Palu watershed which stretches from west to east and empties into the main channel of the Palu River (Figure 1).Geographically, the Sambo catchment is located at coordinates 119°47'39.91"E-119°52'36.45"Eand 1°5'13.25"S-1°8'45.43"Sand occupies an area on the west side of the Palu river.Even though the catchment area is indicated to be a very small proportion of the Palu watershed, the Sambo river affects the discharge of the Palu river both at peak and low flows.Based on a review of the morphometric aspects, this catchment has a moderate category of river network density with a wide shape at the upstream and gradually narrows downstream to the outlet (Figure 2a).The density trend of the river network is proportional to the topographic changes  In general, the catchment land cover is relatively well conserved.Medium density secondary forest dominates the upstream catchment area (Figure 3a).Downstream on the low topographic slope, most of the natural land cover has been converted to mixed plantations such as coconut, cacao and other crop varieties.When examined in more detail, in fact several open space areas can be seen both in the upstream, middle and downstream areas.These are actually the landslide areas that formed after the 2018 Palu Earthquake.The uniform characteristics of the land cover in this catchment are closely related to the type of surface soil.It appears that this catchment was formed by the dominant Red-Yellow Podzolic/Lithosol type soil and a small portion of the Brown Forest/Alluvial type (Figure 3b).Furthermore, this area is very vulnerable to flooding in connection with the increasing intensity of rainfall due to global climate change.After the 2018 Palu Earthquake, the frequency of flooding has increased, accompanied by mass sediment loss and flash floods.The decrease in slope stability both on the river bank and in the upstream catchment due to earthquake shocks makes the soil surface more prone to erosion and landslides.In the period 2015 -2022 floods on the Sambo River have inundated residential areas and agricultural areas several times (Figure 4).The floods on May 30, 2017 and April 2019 were the biggest floods which caused the cessation of local community activities, not only damaging public facilities but also relocating residents to evacuation areas.Floods with a depth of more than 1 meter carry various types of materials and are dominated by sediments that are eroded either on the surface of the watershed or on the cliffs and river channels by flowing water.

Data
Research materials used in the study include topographic maps, land cover, soil type data, TRMM rainfall data, hourly and daily rainfall data (ARG), hourly and daily discharge data (AWLR).Topographic maps, land cover and soil types can be obtained from BAPPEDA Sigi Regency while rainfall and discharge data can be obtained from Mercy Corps Indonesia.Meanwhile, TRMM rainfall data was obtained from https://giovanni.gsfc.nasa.gov.Other data are direct discharge measurement data carried out at the Sambo AWLR gauge for calibration purposes and soil sample testing data for soil type verification.
The equipment used in this study is a set of hydrological and hydrometric measuring tools installed at Sambo catchment by Mercy Corps Indonesia (Figure 5).

HEC HMS-Model
One of the semi-distribution-based rainfall-runoff transformation models is the HEC-HMS [14].This model was introduced by USACE-HEC to support hydrologists and practitioners in predicting discharges with rainfall input in a catchment system.The output of the system is highly dependent on the nature of the rainfall and the characteristics of the transformation medium which includes rainfall intensity, rainfall distribution, land use/land cover (LULC), topographic slope, stream network, soil texture, and other important factors.Most of the flow-determining parameters can be accommodated in the HEC-HMS model which is defined into the basin model and direct run-off model [15].
The HEC-HMS model can be applied to various characteristics of catchment systems, both simple catchments and high complexity catchments represented by the basin model.Basin models are not only described by the characteristics of natural catchment systems, but also by anthropogenic factors such as management of catchment systems with various hydraulic structures in them such as dams, weirs, sabo dams, flood dykes and other similar structures.Along with the development of computational technology, currently the HEC-HMS model is not only used for flow simulation but can also be applied to estimate surface erosion rates [16][17].
The three important methods integrated in this model package are expressed in the transformation method, loss method and baseflow method.The first method is aimed at predicting the volume of direct runoff by using several approaches, one of which is the Snyder Unit Hydrograph method which is expressed by: =   −   −  4 (2) is defined as time lag of catchment (hour),   is duration of rainfall (hour) ,   is time lag of desired UH,   is duration of desired UH and   is standard peak UH,  is catchment area,   is UH peaking coefficient, and  is conversion constant (2.75).
The second method which represent the total loss of effective rainfall can be predicted by SCS Curve Number (CN) formula as follow: = 0.2 (6) refers to excess rainfall (mm), and   is initial abstraction,  is cumulative rainfall  is potential maximum retention, and  = curve number: 0 (permeable surface) to 100 (water), The final method which illustrates the baseflow can be expressed by the recession equation which can be written as: define as baseflow at time t,  0 is initial baseflow and  is a recession coefficient.

Evaluation Criteria
Root Mean Square Error (RMSE) is a criterion evaluation method for measuring the reliability of a model in simulating a real system.This criterion indicator is widely applied in the field of hydrology because it can explain the accuracy of the model expressed by the number of errors or differences in deviations.This error indicator is formulated by [18][19][20][21]: refers to predicted discharge (m 3 /s),  0 is observed discharge (m 3 /s) and  is the number of discharge data.

Rainfall Intensity
Two types of rainfall data as previously reported: observed and TRMM data are used as input for the HEC-HMS Model which will be evaluated on the shape of the hydrograph.The data is daily rainfall which is transformed into rainfall intensity using the Mononobe approach based on depth and duration of rainfall.This transformation is performed due to the limitation of hourly rainfall in TRMM data.The duration of the rainfall to determine the intensity of rainfall is based on the duration of observed data (ARG) with a duration of 4 hours.The results of the observed rainfall intensity transformation tend to be similar to ARG rainfall intensity.Rainfall intensity of observed and TRMM data can be seen in Figure 6.

Parameters of the HEC-HMS
The HEC-HMS model has been assigned to determine the flood hydrograph based on two types of inputs namely observed rainfall intensity and TRMM rainfall intensity as illustrated in Figure 6.Due to various types of uncertainties in the model, calibration has been performed to determine optimal parameters.At least 8 parameters of the HEC-HMS model have been calibrated based on pairs of observed rainfall data and observed discharge data as shown in Table 1.The optimal parameters in Table 1 are then used to simulate observed and TRMM rainfall data in predicting hydrographs.As shown in Figure 7, the three types of hydrographs show a similar shape both on the rising side and on the recession side, even though the peaks of the hydrographs are indicated differently.The peak discharge hydrograph with observed rainfall input is lower than the peak discharge observed hydrograph.Meanwhile the hydrograph peak discharge with TRMM rainfall input is higher than the observation hydrograph peak discharge although the is not so significant.Even though the difference between the peak discharge hydrograph due to observed rainfall is relatively lower than the peak discharge hydrograph due to TRMM rainfall to the peak discharge of observed hydrograph, basically the correlation is higher with a coefficient of determination of 0.9487 and 0.9623 respectively (Figure 8).This shows that the performance of TRMM rainfall data is lower than the performance of observed rainfall data as hydrographic transformation input with RMSE indicators of 0.979 and 1.731 respectively (Table 2).This relates to the fact that not all potential atmospheric rainfall (TRMM) can become net rainfall falling on the surface of the catchment [8].This is in line with the results of a study by Harris et al. (2017) and Collischonn et al. (2018) who reported that TRMM rainfall data provides good results in predicting hydrographs [12][13].The results of the verification and validation also show the same tendency that the use of this data is relatively satisfactory even though in some cases deviations appear and it implies that this data should be validated first.The HEC-HMS Model has been assigned to transform the rainfall data into a hydrograph using three main components: Snyder as the transformation method, SCS-CN as the loss method and recession as the baseflow method.The estimated results of the hydrograph are then compared with observational discharge (AWLR) and simulated discharge due to observational rainfall, which has previously performed calibration.
The results of the study indicated that the accuracy of simulated discharge with TRMM data input was relatively lower than the simulated discharge with observational rainfall input with RMSE indicators of 0.979 and 1.731, respectively.This is related to the TRMM rainfall data which do not fully have the potential to become runoff in the catchment due to translation and evaporation in the atmosphere.In addition, this deviation is also caused by the sensitivity of the HEC-HMS model in simulating the rainfall-runoff transformation.However, this TRMM rainfall data should be validated first and this data can be an alternative to rainfall data for catchments with limited data.

Figure 1 .Figure 2 .
Figure 1.Location map of the study area influenced by surface elevation where the density tends to be low in flat topography.This is relatively unconfirmed by the topographical slope as shown in Figure2b.Slope distribution does not show consistency with stream density.Catchment topography is generally dominated by slopes of more than 15%.This characterizes the topographical shape of Palu's tributaries where the length of the main river is relatively small with a relatively high bed slope.Overall, this affects the flow characteristics, especially peak time and peak flow of a flood hydrograph where the hydrograph shape is generally skew to the left with sharp peaks.

Figure 3 .
Figure 3. Land cover (a) and soil type (b) of Sambo catchment a). Farming area b).Settlement area Figure 4. Inundation in the lower part of the catchment (a) (b)

5 .
These two tools function to measure hourly and daily rainfall and water levels.These two gauges are important instruments for monitoring water level fluctuations in rivers related to the Early Warning System (EWS) after the 2018 Palu Earthquake.The Sigi Regency Government has launched a disaster mitigation program in collaboration with Mercy Corps Indonesia in several disaster-prone areas, one of which is by placing a tool potential disaster monitors such as rain monitors and water level monitors.Data from these two instruments can be accessed in real time by the wider community.a).Rainfall station b).Water level station Figure Hydrological station in the catchment

Figure 6 .
Figure 6.Rainfall intensity of observed and TRMM Data transformed using Mononobe approach Optimal parameters as a result of HEC-HMS model calibration have been used to predict the hydrograph.Two types of rainfall data have been inputted into the Time Series Data Manager as shown in Figure6which depicts the depth of rain over a four-hour duration.Hourly observation discharge data has also been set at the Sambo catchment outlet on the same component as the rainfall data entry.This data is applied as a hydrograph suitability controller as a prediction result of the HEC-HMS model based on observed rainfall input data and TRMM rainfall, as shown in Figure7.

Table 1 .
Parameters of the HEC-HMS Model in the study area

Table 2 .
Discharge deviation measured by RMSE indicator Utilization of Tropical Rainfall Measuring Mission (TRMM) data has been applied to predict flood hydrographs in one of the lack data catchments in Central Sulawesi Province, Indonesia, namely Sambo.