HEC-HMS based rainfall-runoff model for Punpun river basin

Water is the most important natural resource on the earth. Human life will not survive without water (Abdullah 2013) and thus it is also called the gift of God. Water moves along the natural slope of ground surface through concentrated small rivulets and flows down in defined streams and thus stream flow is generated (Cunha et al. 2016; Efthimiou 2018). The estimation of stream flow or runoff from a catchment is necessary for the purposes of assessing the flood peaks, water availability for municipal needs, design of storage facilities for multiple purposes, planning of irrigation operations for agricultural or other industrial purposes, wildlife protection, estimating future dependable water supplies for power generation etc. Further, rainfall is the basic input of the hydrological cycle that can be measured easily and economically, while runoff is a dependent variable that needs to be estimated for the corresponding rainfall. The measurement of rainfall has been practiced for a long time and it is easily available as compared to runoff (Scharffenberg 2013; Tan et al. 2017; Tassew et al. 2019). Precipitation characteristics, catchment shape and size, soil types, land use and topography are the main factors affecting runoff from a catchment (Jean-Baptiste et al. 2011). Land use land cover (LULC) and soil type have been proved to be most sensitive factors in hydrological behavior as compared to catchment shape and size that affect the generation of stream flow (Chinnasamy & Agoramoorthy 2015; Hassan & Jin 2016; Jaiswal et al. 2020). It was observed that the hydrological response of a catchment is most sensitive to the percentage of land-use and distribution of soil type (Houborg et al. 2012; Ismael et al. 2017; Opie et al. 2020). Study of the effects of catchment behavior on floods, prediction of catchment response to changes in the input conditions, modeling of river behavior, real-time flood forecasting, adjusting and evaluation of water resource management and effects of river's activity such as erosion and sedimentation can be made with the help of geographic information systems (GIS). GIS can be defined as ‘a computer system capable of capturing, storing, analyzing and displaying geographically referenced information; that is, data identified according to location’ (Nenwiini & Kabanda 2013; USGS 2016; Koltsida & Kallioras 2019). The data for rainfall-runoff (R-R) modeling ranges from a digital elevation model (DEM) to derive physical basin characteristics, soil characteristics (for parameters about infiltration loss and soil moisture holding capability) and land use classes derived from satellite images etc. (Banerjee et al. 2020; Kumar & Bhattacharjya 2021). With the advent of GIS, these data can be easily processed and analyzed to estimate hydrological modeling parameters (Prasad & Narayanan 2016; Seekao & Pharino 2016; Kumar & Himanshu 2017). ArcGIS is a more advanced tool and provides a larger user base with better interface (Khatami & Khazaei 2014). ArcGIS developed by Environmental Systems Research Institute (ESRI) is the versatile and comprehensive GIS package used by the water resources group of researchers, academicians and engineers. Several modules of ArcGIS are specially developed for these purposes. Hydrologic Engineering Center (HEC) develops the extension tool viz. hydro arc tool and HEC Geo-HMS to prepare the HEC-HMS (Hydrologic Engineering Center-Hydrologic Modeling System) database. The ArcGIS tool is used to perform terrain analysis using DEM to delineate the basin and river-stream network. HEC Geo-HMS computes the various basin characteristics such as basin area, stream length, slope etc., which are necessary for developing the conceptual rainfall-runoff model in HEC-HMS (Ghani et al. 2012; Halwatura & Najim 2013; Khoshtinat et al. 2019; Roy & Saha 2019). HEC-HMS is software that simulates the rainfall-runoff processes in river basin systems. It can be used to solve various problems. The HEC-HMS model can be used to conduct studies on water availability, urban drainage, flow forecasting, simulating flood event, future urbanization impact, flood damage reduction, wetland hydrology, reservoir spillway design, flood plain regulation, and system operation (Ashish et al. 2012; Gautam et al. 2013; Oleyiblo & Li 2010; Shrestha et al. 2014; Sok & Oeurng 2016; Pichuka et al. 2017; Kazezyilmaz-Alhan et al. 2021).

The rainfall-runoff model is broadly classified into empirical, conceptual and physical models. The empirical model, also called a data-driven model, requires fewer parameters, which can be calibrated and measured from field properties or fixed through an empirical procedure (Ntoanidis et al. 2013; Gebre 2015; Gull & Shah 2020). Conceptual models are mostly applied in small homogeneous areas by spatial lumping and replacement of various components of the hydrological cycle by conceptual storage elements (Song et al. 2011; Khaddor et al. 2017; Pichuka et al. 2017; Winarta et al. 2019). The physical model is based on the understanding of hydrological processes and relates to catchment parameters. It is a distributed model and incorporates spatial and temporal variability of all the parameters. It is the most complicated model and requires a large amount of distributed data of topography, LULC, soil distribution etc. The differences in the model's characteristics affect the applicability, accuracy and reliability of the runoff response from extreme rainfall or prediction of runoff in catchments. Catchment morphology is one of the most important natural environmental factors that significantly affect hydrological processes. Researches on HEC-HMS showed its ability to simulate and forecast stream flow based on various datasets and catchment types, especially for the small catchments. Though the HEC-HMS model has been used globally, very few studies have been made in Indian region catchments.

The Punpun river basin has extensive agriculture (Gumindoga et al. 2016; Derdour et al. 2018; Joshi et al. 2019; Tassew et al. 2019). It carries a huge discharge during monsoon and inundates the flood plains and no discharge during lean period. It has also been severely impaired by land degradation. Patna city is situated near the left bank of the Punpun river. In the past, it has been observed that floods of the Punpun river inundate the flood plain and nearby area including part of Patna city, which causes huge loss to the people and property in the region. This monsoon discharge may be conserved and utilized during lean periods. Therefore, to implement soil and water conservation and flood protection measures, it is foremost important to compute the runoff from the basin. Keeping this in mind, this study has been conducted to estimate the runoff from the Punpun river basin for a given input of rainfall. The main objective of this study is to develop a rainfall-runoff model of the Punpun river basin using HEC-HMS.

Figure 1 presents a map of the study area, the Punpun river basin. It originates in the hills of Palamu district of Jharkhand state at an elevation of 300 meters at a latitude of 24° 11′ N and longitude 84° 9′ E. After flowing 25 km, it joins river Ganga at Fatuha, downstream of Patna. It carries high discharge during monsoons, creates floods and inundate the adjoining area but has no flow during lean periods. Both problems, high discharge and no flow, lead to water management issues in the basin (Panda & Wahr 2016). This river basin receives heavy rainfall during the months August to October. Generally, the climate in and around the basin is humid, with moderate to high wind velocity. The average annual rainfall varies from 99 cm near the confluence of the Ganga (Patna district) to 134 cm in the upper reaches (Palamu district). The Punpun basin receives 80–87% of its annual rainfall during the monsoon. Precipitation is uniform and does not vary frequently in the lower part of the catchment. A Central Water Commission gauging site is located at Sripalpur at a Latitude of 25° 30′ 06″ N and longitude of 85° 06′ 08″ E in the Patna district. The river outlet is considered to be at Sripalpur and accordingly the basin is delineated as shown in Figure 1. The study area of the Punpun river basin is 7055 km². The geology of the area is diverse from granite, gneiss and charnokites in the hills to recent alluvium in the plains. Infiltration losses vary with location due to differences in soil and slope characteristics.

ADEM of 30-meter pixel size was taken from the website of the United States Geological Survey (https://earthexplorer.usgs.gov/). Figure 2 shows the DEM of the study area. Daily rainfall data at four rain gauge stations (Sherghati, Inderpuri, Arwal and Patna) were collected for the period 2005–2017 from the Indian Meteorological Department (IMD) Pune. The monthly average rainfall during monsoon at rain gauge stations Sherghati, Inderpuri, Arwal and Patna were 420, 481, 398 and 480 mm, respectively. The average annual rainfall is 950 mm and the maximum rainfall generally occurs in month of September.

The mean daily rainfall over the basin has been computed using the Thiessen polygons method from the years 2005 to 2017. The daily discharge data at the Sripalpur gauge site from the years 2005 to 2017 were collected from the Central Water Commission, Patna. The soil map of the Punpun river basin was collected from the World Soil Database, with 1 km resolution (https://www.fao.org/soils-portal/). In the uppermost part of the basin is mostly forest cover, while the lower part is suitable for agriculture purposes. The Landsat image used for classification of study area in different land use cover was collected from the website of United States Geological Survey (https://earthexplorer.usgs.gov/).

Figure 3 presents the flow chart of different processes carried out using ArcGIS. Role of elevation is most important in the hydrological studies of any catchment for which a DEM is required. DEM is used to obtain geographical, morphological and topographical properties of the basin (Coulibaly et al. 2005; Young & Liu 2015). Basin physical parameters such as slopes, centroid, longest flow paths, information associated with elevation of the basin, river flow direction, stream orders of the river and drainage pattern have been obtained. Based on the data, slope and topographic elevation maps with contours for the basin have been obtained.

HEC-HMS is a semi=distributed model, in which data can be heterogeneous within basin but it is homogenous within sub-basins (Beven 2012). The basin has been delineated by generating 3 sub-basins based on physical properties of study area, i.e. LULC and soil distribution using ArcGIS for a particular river portion. These sub-basins (SB1, SB2 and SB3) are shown in Figure 4.

The supervised classification has been performed to prepare the LULC map of the basin. Hydrologic soil groups (HSGs) are commonly classified into 4 groups: A, B, C and D based on rainfall, runoff and infiltration (Rodell et al. 2009; Roy et al. 2013; Derdour et al. 2018; Efthimiou 2018; Frappart & Ramillien 2018). Group A indicates high infiltration rate and low runoff capacity. Group B represents silt loam having moderate infiltration rates and moderately coarse texture. Group C represents clay loam, which has slow infiltration rates and moderately fine textures. Generally, clay loam soil is good for every type of plants' growth. Group C soils have moderate water transmission rate, while Group D soils have low infiltration and high runoff. The soil map of the Punpun river basin was obtained with 1 km of resolution from the World Soil Database. The dominant soil in the Punpun river basin is DOMSOIL and HSG is determined based on FAO.

The hydrological characteristics of the basin have been computed using the physical properties of all the sub-basins for simulation. The SCS-CN loss method has been used to compute rainfall losses. This method was chosen due to the availability of data for the region and the lack of complexity in nature (Song et al. 2011; Hoseini et al. 2017; Yu et al. 2018; Tassew et al. 2019). The CN of each sub-basin is required to estimate infiltration using the CN method. The detailed procedure of CN creation and operation of the HMS model is shown in Figure 5.

SCS-UH method was used to calculate the outflow at the specified outlet. This is an innovative method that needs lag time and percentage impervious area of the basin (Khaddor et al. 2017; Barman & Bhattacharjya 2020; Tassew et al. 2019). The basin lag time was taken as 0.6 times the concentration time (Subramanya 2013). Table 2 presents the percentage impervious layer, potential soil storage, composite CN, initial abstraction, lag time and time of concentration computed for each sub-basin of the Punpun river. The SCS method has been given in many text books, so it is not repeated here.

HEC-HMS uses various methods to compute the stream flow at the basin's outlet, including the Muskingum method, which is simple and does not require multiple inputs (Song et al. 2011; Tassew et al. 2019). The Muskingum method is used to predict stream flow at the outlet. K and x are two important input parameters in the Muskingum method. The value of K is computed by dividing the length of reach by the average flow velocity and the value of x lies between 0.0 and 0.5. The channel has minimum attenuation if x is equal to 0.5 and maximum attenuation if x is 0.0. The calibration parameters were set to a value of 0.2 as an initial value, which was then modified during the calibration.

Historical data of different periods were used for calibration and validation of HEC-HMS model. Daily rainfall and runoff data from the years 2005 to 2012 were used for calibration and from the years 2013 to 2017 were used for validation. LULC and soil characteristics of the sub-basin were assumed as unchanged during the calibration and validation period. In HEC-HMS, Nelder Mead and univariate gradient are two algorithms to optimize the objective function. In this study, peak weighted RMS error was minimized using the Nelder-Mead algorithm to get the optimized model parameters for the best outcomes of simulated model. Optimized parameters were used to validate the model. Sensitivity analysis was also performed to recognize the expected parameters and for required accuracy. The RMSE and Nash-Sutcliffe Efficiency were used to evaluate the performance of HEC HMS model.

NSEs, E, vary from −∞ to 1. Its value is considered good if it ranges between 0.65 to 0.75 and very good between 0.75 to 1. An efficiency of 1 (E = 1) shows a perfect match of computed values with the observed data.

The LULC map was prepared from Landsat image using ArcGIS. Supervised classification has been carried out for Agriculture land, Barren land, Built-up area, Forest, Water body and Wetland, which are shown in Figure 6. The prepared LULC map indicates that the most of the study area consist agriculture and forest lands, which means the runoff coefficient is in the range from moderate to high. The urban built-up area is very small compared to agricultural land. An HSG map was prepared from world soil data base and shown in Figure 7. It shows that 74% area has HSG C and 26% area has HSG B. It means a major portion of study area has HSG C with moderate to high runoff potential zone. Table 1 presents the percentage area of the soil groups.

Based on the LULC and soil maps, the CN map has been generated and presented in Figure 8. It shows that CN values are high in the major portion and thus, the runoff potential of Punpun river basin is moderate to high. The average CN value is computed for each sub-basin and presented in Table 2.

Three models have been developed using HEC-HMS software to simulate daily, monthly and monsoonal flow in Punpun river basin. These models have been calibrated using rainfall and observed flow data from the year 2005 to 2012 and validated with the data from 2013 to 2017. The objective function in HEC-HMS is peak-weighted RMSE during calibration. Table 3 presents the optimal value of various parameters for HEC-HMS.

The daily flow model is developed using daily rainfall and daily observed stream flow data at Sripalpur gauging site on Punpun river. Figures 9 and 10 presents the results of calibration and validation of the daily model, respectively. The simulated stream flow matches well with observed flow data during calibration and validation except for few peaks. During calibration, the simulated stream flow was slightly lower than the observed values during the year 2005–2009, but after that it was slightly higher than the observed values. In validation, the simulated stream flow was slightly lower than the observed in the last two years (2016 and 2017). The differences in the simulated and observed data may be due to LULC map, which was prepared by Landsat data of year 2009 and it was used for the whole study period (2005–2017). In post-monsoon seasons, Peak flows were overestimated by 19% during calibration and underestimated by 23.1% during validation. Overall, the pattern of simulated stream flow was in good agreement with observed.

To check the accuracy of the daily model, a linear regression was fitted between the simulated and observed stream flows for calibration and validation and shown in Figure 11. The coefficients of determination were 0.88 and 0.79 during calibration and validation, respectively.

The monthly model has been developed using the monthly data of rainfall and stream flow. Figures 12 and 13 present the comparison of simulated and observed monthly stream flow during calibration and validation of the monthly model, respectively. The simulated stream flow matches well with the observed data during calibration whereas during validation, simulated stream flows were greater during the 23rd–26th months. Linear regression plots were also drawn to check the accuracy level and the value of coefficient of determination during calibration and validation were 0.85 and 0.82, respectively and presented in Figure 14.

In this model, monsoonal stream flow data from the months June to October for all the years were used for calibration and validation. Figures 15 and 16 present the comparison of simulated and observed monsoonal stream flow during calibration and validation, respectively. The simulated stream flows were in good agreement with the observed data during calibration, whereas during validation, simulated stream flows were slightly higher in some months. Figure 17 presents the linear regression plots between simulated and observed monsoonal data. The values of coefficient of determination were 0.78 and 0.82 during calibration and validation, respectively.

Table 4 presents the statistical parameters R², NSE, PBIAS and RSR for the analysis of the three models (daily, monthly and monsoonal) in HEC-HMS. R² values for all the models are above 0.75, thus it shows good calibration of the model. NSE values for the daily model lies between 0.65 and 0.75, thus show good result whereas for monthly and monsoonal models, it is greater than 0.75, and thus it shows very good results. PBIAS values are less than 10 in all the models, thus show very good results. RSR values for daily and monsoonal models are less than 0.5, which shows very good results, whereas for the monthly model, it is slightly greater than 0.5 and it shows good results. The performance evaluation of all the models shows that the monthly model performs better than the daily and monsoonal models.

A HEC-HMS based rainfall runoff model has been developed for the Punpun river basin. Three models have been developed using daily, monthly and monsoonal data. The model has been calibrated and validated using data from the years 2005 to 2012 and 2013 to 2017, respectively. Performance of the model has been evaluated using statistical parameters: R², NSE PBIAS and RSR. Based on R², NSE and PBIAS values, the monthly model performs better both in calibration and validation as compared to the daily and monsoonal models but RSR values are slightly higher than the other models. Thus it can be concluded that the monthly model is better than the daily and monsoonal models. It can also be concluded that HEC-HMS based rainfall runoff model can be used to forecast the stream flow in Punpun river basin. The assumption of this study is that only one LULC scenario in the mid-time period has been considered in whole study period due to the major agricultural land use area, which did not change much on a yearly basis.

Data cannot be made publicly available; readers should contact the corresponding author for details.