Comparison of the application of HBV and HEC-HMS hydrology models for accessing climate change in the upper Citarum Watershed, Indonesia

Hydrologically, the upper Citarum watershed was critically damaged. This study propose to compare HBV and HEC-HMS model performance for estimating discharge in upper Citarum watershed. Besides, this study wants to know changes in discharge in the future as a result of climate changes. Discharge simulation was carried out with the HBV and HEC HMS models in the period 2006-2007 as calibration and 2008 as validation. Meanwhile, future discharge changes are calculated in 2045 based on climate model projections output from CORDEX namely MPI, CNRM, EcEarth, and CSIRO. Model performance is calculated based on the value of statistical bias, NSE, and correlation. The results showed that HEC-HMS model has bias, NSE, and corellation value of 6.33, 0.57, and 0.8. Whereas, the HBV values are 3.67, 0.61, and 0.76. Therefore, based on NSE and bias value, the HBV model performs slightly better than the HEC-HMS in upper Citarum watershed. There are an increase in the daily discharge by RCP4.5 scenario. In contrast, the daily discharge by RCP8.5 decreased in upper Citarum.


Introduction
Upstream Citarum watershed is part of Bandung Basin.Hydrologically, the area has suffered severe degradation.This condition is a result of the use of land and water resources that exceed their carrying capacity.There was a decrease in the conservation area in the Citarum River during 2009-2018 so that the rainfall tended to run off into water bodies rather than being retained or infiltrated [1].This condition could result in water and land resource disasters, for example floods.Therefore, estimating the amount of discharge as a response to rainfall that falls on the watershed must be considered in the management of the upstream Citarum watershed.
The problems and limitations have encouraged the application of various hydrological models to estimate flow discharge in a watershed.Various hydrological models with various advantages have been applied to deal with the problems and limitations.The HBV model (Hydrologiska Byråns Vattenbalansavedlning) was a hydrological model developed by the Swedish Meteorological and Hydrological Institute (SMHI) in Sweden since 1972 [2,3].The HBV model was quite simple, because it was only require rainfall data and potential evapotranspiration to produce model output in the form IOP Publishing doi:10.1088/1755-1315/1314/1/012072 2 of a flow hydrograph from a watershed [4].The performance of HBV model is not affected by seasonal changes and could be used to analyze the process of flooding due to various factors [5].Moreover, the HBV model offered a user-friendly and computationally efficient model.Furthermore, the use of the HBV model to adjust the impact of climate change on water resources has been carried out throughout the world [4].Apart from HBV, this research also used the HEC-HMS (Hydrologic Modeling System) model, a numerical hydrological model developed by Hydrologic Engineering Center (HEC) of the US Army Corps of Engineers.The HEC-HMS program was a computer program for calculating rainfall transformations and routing processes in a watershed system.HEC-HMS is widely used for many exploratory studies due to its simplicity and it is freely available as an open-source software [6].HEC-HMS has provided calibration facilities, distribution model simulations, and the ability to read GIS data [7,8].The HEC-HMS model packages various methods used in hydrological analysis.Several studies on the use of the HEC-HMS model show that the model was able to simulated low and medium flows accurately [9][10][11].Other studies represent that the HEC HMS model is quite reliable for simulating the magnitude and timing of peak flows, both during flood events and in long-term simulations [6,[12][13][14][15][16].The HEC-HMS model produced a plausible compatibility between observed and simulated discharges on daily and monthly time scales [9] Both HEC-HMS and HBV were adequate for simulating river flow in tropical catchments with high humidity and limited data [17].The HBV model simulation by Masitoh and Dasanto (2018) in the Upper Citarum watershed has quite good performance [18].Besides, the HEC-HMS model has good performance in predicting flood discharge in the upper Citarum watershed with a correlation value of no less than 0.81 [19,20].Hence, this research aims to compare the performance of the HBV and HEC HMS models for estimating flow discharge in the Upper Citarum watershed.Besides, this study wants to know changes in discharge in the future as a result of climate changes.

Overview of Upper Citarum Watershed
Geographically, the Upper Citarum watershed is located at 6°46' -7°15' S and 107°22'-107°57' E (Figure 1).The upper Citarum watershed has a humid and cool mountain climate with an average air temperature of 23.4°C.The average annual monthly rainfall is 198.8 mm with an average of 18 rainy days per month.Topographically, it is located at an altitude of 650 m to more than 2000 meters.The upstream Citarum watershed covers two (2) administrative areas, namely: Bandung City and Bandung Regency.Most of the Upper Citarum watershed area is mountainous, with a tropical climate.The main drainage pattern of the upstream Citarum watershed is the Citarum River with several main tributaries, namely Cikapundung, Cikeruh, Citarik, Cirasea, Cisangkuy, Ciparay, and Ciwideuy.The Upper Citarum River is located on Mount Wayang, Bandung Regency which flows into the Saguling Reservoir and empties into the North Coast of Java, precisely in Karawang Regency.The Upper Citarum watershed area is geologically divided into four (4), namely tertiary deposits, old volcanic products, young volcanic products and lake deposits.

Data and Methods
The model simulation used daily discharge data (m 3 /s) of the Nanjung waterlevel checking point from BBWS Citarum.This study also utilized rainfall and daily air temperature data from Bandung ITB Geophysical Meteorology Station from BMKG.Furthermore, BIG National Digital Elevation Model (DEMNAS) data is also used to calculate the area and boundaries of watersheds and river flows.Simulations using the HEC-HMS model also required maps of soil types in the upstream area of the Citarum watershed.The climate model output used in this research is the output from Regional Climate Model (RCM) CORDEX CORDEX SEA (Coordinated Regional Climate Downscaling Experiment Southeast Asia), namely MPI, CNRM, EcEarth, and CSIRO (2045).This research uses climate scenarios both RCP4.5 and RCP8.5 scenarios.

HBV models.
The HBV model has four main modules namely (1) snowmelt and its accumulation (only used in subtropical and high latitude regions), (2) soil moisture and effective precipitation, (3) evapotranspiration, and (4) runoff response) [21,22].Model HBV estimated effective precipitation based on soil water content (Soil Moisture) as follow the equation [22]: (1) Where P Eff is effective rainfall, SM is Soil Moisture, FC is Field Capacity, and  is model parameters (slope coefficient ) .
Potential evapotranspiration (PEm) was calculated based on the Thornthwaite method by entering the monthly average temperature (Tm) parameter.Furthermore, the daily potential evapotranspiration of the HBV model is calculated based on the equation [22]: Where PE is Daily adjusted potential evapotranspiration, C is Model parameters, Q is daily temperature, and Mr is average daily temperature per month Actual evapotranspiration (Ea) was determined from the actual daily evapotranspiration value by combining the soil water content and the Permanent Wilting Point value.
Determination of surface runoff, interflow and base flow in the HBV model uses the concept of a flow tank located at the outlet of the watershed.The HBV model flow system generally has two tanks, namely the first tank model is close to surface runoff and the second tank is used to simulate baseflow which is the contribution of groundwater [22,23].
Where S1 is water level in the first storage area (mm), S2 is water level in the second storage area (mm), L is threshold of the water level in the upper tank (mm), A is area of watershed (km 2 ).

HEC-HMS models.
The HEC-HMS model has three main components, namely: 1) Basin Model, namely the elements contained in a sub-watershed as well as the parameters in runoff; 2) Meteorologic model, which contains rainfall and evapotranspiration data; and 3) Control specifications, which are simulation time intervals to start or end data simulations [8].
The HEC-HMS modeling program provides several calculation methods, both runoff, unit hydrographs and baseflow.This research was used the user specified hyetograph for rainfall model method, the SCS Curve Number for runoff volume model, the Snyder for direct runoff model, and the lag channel for flow model.
The Soil Conservation Service (SCS) curve number (CN) calculation method assumed that the rainfall for producing runoff is a function of cumulative rain, land use, soil type and humidity [8,16].
The CN (curve number) value varies from 100 (for water s) to around 30 (for non-waterproof surfaces with high infiltration values).
The Snyder method is a synthetic unit hydrograph model developed by Snyder.The parameter in the Snyder method required for the process of forming the HEC-HMS flow hydrograph is the time lag (tlag) [8,24].
=   (  ) 0.3 (10) where tlag is time lag (hours), Ct is basin coefficient; L is length of the main stream from the outlet to the divide; Lc is length along the main stream from the outlet to a point nearest the watershed centroid; and C is a conversion constant (0.75 for SI) [8].
Where A is watershed area, C1 is constanta (2.75 for SI), and Cp is coefficient peak UH. quantitatively with statistical values.The statistical criteria used in this study include the Nash-Sutcliffe Efficiency (NSE) coefficient [21], bias [27], and correlation [28].
where Qmod and Qobs are simulated (model) and observed river discharges, and  ̅̅̅̅̅̅̅ is the mean of observed values.NSE value > 0.7 or 70% can be said that the model has a very good level of accuracy.If the NSE value ≤ 0.5, the simulation can be said to have a low level of accuracy, so the simulation must be repeated (Table 1).The calibration results for the annual period can be seen in Table 2.The NSE values obtained ranged from 0.11 to 0.61.High NSE values were obtained in 2006, both for the HBV and HEC-HMS models.

Model Calibration
The 2007 calibration produced the lowest NSE value.Apart from NSE, the bias value and model correlation with observations are also looked at.The bias of the HBV model tends to be smaller compared to the HEC-HMS model.In the HEC-HMS model, the smallest bias was obtained during the 2006 calibration period so that this calibration parameter is used in subsequent simulations or model validation.
Based on the calibration results, if we look at the 2007 period for both the HBV and HEC-HMS models, the small NSE value and large bias indicate problems in the calibration process.These problems include model uncertainty and equity [29].Uncertainty can arise due to random errors that can come from the input data (rainfall) used or from the measured discharge data.Apart from that, problems can also arise due to inaccurate determination of parameter values (especially in model optimization) and biases in the model structure and incomplete data.

HBV Model Simulation
The parameter values obtained from the HBV model calibration results are: FC, Beta, C, K0, L, K1, K2, Kp, and PWP respectively at 400, 6, 0.02, 0.05, 50, 0.08, 0.05, 0.1, and 100.Each parameter has a different sensitivity to the model.FC and C parameters are not sensitive whereas the values of Ko, K1, Kperc, and L are very sensitive.These parameters directly affect surface runoff and intermediate runoff, thereby affecting the overall discharge volume.Apart from that, the error of the entire model.are also influenced by these parameters.While the FC and C parameters do not affect the performance of the model, they do affect the model's error value linearly.
The HBV model simulation results in the Upper Citarum watershed both during the calibration period (Figure 2) and validation (Figure 3) have a pattern that is almost the same as the observation results.The NSE value of 0.61 (Table 2) also showed that the model is able to produce a discharge response even though it is not very precise.The simulation model still calculated daily flow discharges that are underestimated or overestimated at certain times.The annual discharge volume of the HBV model simulation tends to be ±31.58mm lower than the observed annual discharge.In addition, the model also could not describe the peak discharge and the date of the peak discharge.During the observation calibration period was recorded at 348 m 3

HEC-HMS Model Simulation
The upstream Citarum watershed is divided into seven sub-watersheds according to the main river flow pattern.The HEC-HMS model calibration obtains parameterized values according to the method used in each sub-DAS.Parameters obtained in this study include initial abstraction, curve number, initial discharge, recession constant, ratio to peak, peaking coefficient, and standard lag (Table 3).The magnitude of this coefficient will be different in each basin, according to land cover, soil type, and physical characteristics of the watershed.Initial abstraction is the amount of precipitation that must fall before surface excess occurs5.This value together with the Curve Number will affect the amount of effective precipitation where the curve number is as big as indicating that the rainfall that falls will become runoff as big as that curve number.
The HEC HMS model with an NSE value of 0.57 is also able to build flow discharge with fairly good accuracy.The model discharge has almost the same diversity as the observation data (Figure 4 and Figure 5).The simulation model tends to produce discharges that are underestimated during calibration, on the contrary, the model discharges are overestimated during the 2008 simulation.The overestimated value is caused by the rainfall that occurs, which is assumed to be overtopping to become a complete surface runoff.Limited data means that this model is not able to provide a response to the discharge that occurs in accordance with the original conditions.
As with the HBV model, the HEC-HMS model cannot describe the peak discharge yet.3 and Table 4 shows changes in peak discharge and daily flow discharge between the historical period (2005) and the future period (2045) using the HBV model.All models, except the Csiro RCP 4.5 scenario, estimated peak discharges in the future period that are greater than historical.The highest increase, for both the rcp45 and rcp 85 scenarios, was detected by the ecearth model, amounting to 1207.98 and 741.09, respectively.The CNRM RCP 4.5 model produces the smallest peak discharge increase of 5.75%.Overall, the average increase in discharge with the HBV model was 45 -49%.In contrast to peak discharge, daily flow discharge in the future experiences a decrease for the cnrm, csiro, and mpi models.However, both the peak discharge and flow discharge of the Ecearth model increased to 116.35 rcp45 and 99.75 rcp85.The decrease in daily discharge ranges from 4.45 to 17.9% rcp 45.The rcp85 scenario shows a greater decrease in flow discharge, namely between 13.97 to 46.59%.
Based on the Table 5, HEC-HMS model resulted that the peak discharge in the RCP4.5 scenario increase in most climate scenarios.The highest peak discharge increase was found in the eearth model at 835.20 m3/s.However, the peak discharge calculated by the cnrm model actually decreased by 5.30 and 61.8 for scenarios RCP 4.5 and RCP8.5 respectively.The average increase in peak discharge was 41.02%.
Similar with the HBV model, there were decrease in the daily flow rate with the HEC-HMS model.Only the ecearth model consistently experienced an increase in both peak discharge and daily discharge up to 67.99 in the r45 scenario.The mpi model provides the largest decrease in daily discharge of up to 45% with scenario 45.The smallest decrease in discharge was detected by the csiro model at 8.59 m3.The average decrease in discharge was 6.62%

Model Comparison Evaluation
The HBV and HEC HMS models have advantages and disadvantages in modeling river flow discharge from rainfall data.The development of the HEC-HMS model structure is relatively more complicated than the HBV, because it requires several models and appropriate parameters to construct the model structure.Therefore, the HEC-HMS model is able to estimate more detailed DAS response parameters Both HMS and HBV HEC models have not been able to accurately describe the observed peak discharge.Both of them predict the peak discharge at the end of the simulation year, while the observation data shows the peak discharge in April.The statistical criteria for simulation results between the HBV and HEC-HMS models do not show significant differences (Table 2).However, model calibration shows a better level of accuracy in the HBV model.
Utilization of the HBV model is better for locations that have limited data.Meanwhile the HEC HMS model will be more effective in locations with fairly complete watershed parameter data [30].A single hydrological model is still inadequate for accurately simulating runoff processes [11].

Conclusions
The HBV model has slightly better performance than the HEC HMS model in the Upper Citarum watershed.This model has proven to be more practical in estimating flow rates in watersheds with limited data.HEC HMS will be more effective in estimating flow discharge when more complete watershed parameter data is available.There are an increase in future peak discharge, but there are a decrease in daily discharge, both of scenario, RCP 4.5 and RCP8.5.It is necessary to analyze the effect of extreme events due to climate changes on river discharge.

Figure 1 .
Figure 1.Upper Citarum Watershed /s on 20 April 2006 while the peak discharge of the model occurred on 26 December 2006 amounting to 312.06 m 3 /s.

Figure 2 .Figure 3 .
Figure 2. Flow discharge results calibration (period 2006) HBV models The peak discharge observed in 2006 and 2008 was measured on 20 April 2006 and 8 April 2008 of 348 and 486.6 m3/s, respectively.While the model peak discharge occurred on December 26, 2006 and December 17, 2008 of 267.3 and 263.3 m3/s.The inability to describe the peak discharge also occurs in the use of HEC-HMS in the Liqvan watershed, Iran by Loyeh and Jamnani (2017) [10].

Table 3 .
Comparison of historical and future flow discharge using HBV modelHistorical period sub-catchment and river flow.Estimating the parameters of the HBV model randomly requires a long calibration time to obtain small error results.While the estimated parameters of the HEC-HMS model are relatively shorter due to the automatic optimization module in the model.

Table 2 .
Criteria Value Statistics Period Model Calibration and Validation

Table 4 .
Changes in peak discharge and daily discharge using HBV model

Table 5 .
Changes in peak discharge and daily discharge using HEC-HMS model