Evaluation of land use/land cover effect on streamflow: a case of Robigumero watershed, Abay Basin, Ethiopia

The land and water resource of the watershed and its ecosystem are danger due to the nature of the watershed, rapid population growth, deforestation, overgrazing, and soil erosion or soil detachment from the surface are the serious problems in Nile basin (Mengie et.al. 2019). The land use and land cover of the certain basin is subjected to the given change from one land use to the other land use from time to time (Lambin et al. 2003; Welde and Gebremariam 2017a; Bewket and Woldeamlak 2002). The change in land use and land cover are the direct and indirect consequence of human activities (Hassen and Assen 2017; Tadele and Förch 2007). Land use and land cover also has impact on hydrology the basin and these impact are integrated strongly (Hassen and Assen 2017; Ayele et.al. 2017; Getachew and Melesse 2012). In Ethiopia where nearly about 85% of the population is engaged primarily in agriculture and depends heavily on available water resources, the assessment and management of available water resources is a matter of prime importance. Surface water flow modeling is an important tool frequently used in studies in surface water system and watershed management (Bezawit A., 2019). The land use land cover condition is dynamically changing, especially in developing country like, Ethiopia, whose economy depending on agriculture. In particular, the forest land, shrubs, and grass land changed to agricultural and settlement land in most part of the country (Ayele et.al. 2017; Tadele and Förch 2007). For example, studies conducted in Gilgel Abay watershed of Blue Nile basin show that there was the redaction of forest land and shrub land with the increment of agricultural land. About 570 km² of forest and shrub land converted to agriculture and settlement in the year between 1973 to 2001 (Getachew and Melesse 2012). There are several available and important hydrological models which consider physical environment or land use land cover condition to estimate the stream flow or surface runoff including HEC-HMS, MIKE SHE, SWAT, etc. (Tadele and Förch 2007). There are many hydrological models within each class of modeling. Hence choosing the particular model is one of the challenge of model use community. the Two criteria in order to select the hydrological model structure are suggested by (Lambin et al. 2003; Mohammed and Thatiparthi 2020; Jain et al. 2017; Nicótina et al. 2008; Ghonchepour et.al., 2003). The model must be readily and freely available within available documentation and should be applied over arrange of watershed size from large to global (Ghonchepour et.al., 2003). Based on the above criteria Soil and Water Assessment Tool (SWAT) model was selected and used for many studies in Ethiopia. The Soil and Water Assessment Tool (SWAT) model was examined for its applicability to the assessment of water resources in the Upper Awash watershed by (Chekol et al. 2007). In the last thirty years, the land use land cover change was huge which were due to the increment of agricultural land and reduction of forest and grass land in the Robigumero watershed. Several visible change in stream flow and surface run off were observed in the form of flooding and soil erosion during rainy season while reduction of stream flow in dry season in the study area. However, these change of stream flow were not well understood that what couse the change in the watershed. In the study area the major cause of altering streamflow is observed primarily the change in land use land cover including deforestation activities and conversion of grass land to agricultural land. This causes various effects on resource bases like deforestation and agricultural land this leads to the changes in hydrology of the watershed and sediments deposited in stream channels reduce flood carrying capacity, resulting in more frequently over flows and greater floodwater damage to adjacent properties. The main objective of this study is to evaluate land use land cover change effects of on stream flow of Robigumero Watershed. Moreover, this research tried to evaluate the land use land cover changes between 1996, 2006 and 2016 and its implication on stream flow. The outcome of this study befits the stakeholders, water resource planers, farmers, residents’ decision makers and beneficiaries to get aware of the land use land cover change in the watershed and further adaptive important measures to control and protect the negative impact of land use land cover change on the stream flow in the study area.

The Jemma River is one of the biggest tributaries of the Blue Nile (Abay River) Basin and founds in the central highlands of Ethiopia, 180 km North of Addis Ababa. It includes parts of the Wollo, North Shewa Zones of the Amhara, and Oromia Regions. Jemma River is located in the East of the Blue Nile River Basin between 9° 05′ 37’’–11° 10′ 07’’ N latitude to 37° 12′ 07’’− 40° 0′ 01’’ E longitude and cover an area of 15,720 km². From the number of small tributaries flowing from the east of the basin into the Jemma River, the Robigumero River is one of the major gauged tributaries. It covers, the catchment area of 914.7 km² in between 9° 25’–9° 55’ N and 38° 54’’– 39° 20’ East position.

The watershed’s altitude varies from slightly over 2546 m above mean sea level (m.a.s.l) in the southern part to 3624 m a.s.l. The study site has two major seasons: a wet season from May to October and a dry season that extends from November to April. Based on the records from 31 years (1988–2018) at nearby meteorological stations, the annual rainfall depth ranges from 986.7 to 1266.7 mm. More than 85% of the rains fall during the wet season.

According to the FAO soil classification, the dominant soil for the Robigumero watershed was grouped as Calcic Vertisols, Eutric Leptosols, Eutric Vertisols, and Eutric cambisols The most common soil texture for these soil types i.e. for Calcic Vertisols, Eutric Vertisols is clay, for Eutric cambisols is clay-loam, and for Eutric Leptosols is loam.

In this study, according to the Minister of Water Resource, Irrigation, and Electricity land uses land cover data the dominant land covers of the study area are Agricultural land, grassland, deciduous forest, and the catchment area < 1% covered with the Urban area.

Physically based Soil and Water Assessment Tool (SWAT) was used for watershed delineation, hydrologic response unit analysis (HRUs), weather data write up, sensitivity analysis and other watershed characteristic determinations. The watershed delineation operation uses and expands Arc-GIS and spatial analyst extension functions to perform watershed delineation (Easton et al. 2010; Tadele & Förch, 2007; Khalid et al. 2016a, b; Wheater 2007; Bekele et al. 2021). The initial stream network and sub-basin outlets were defined based on drainage area threshold approach. Multiple Hydrological response units (HRU) of the watershed were formed using 20%/10%/20% threshold levels of land use, soil and slope classes respectively (Neitsch et al. 2011; Arnold et al. 2012). After creating multiple HRUs weather write up and simulation of the model follows (Neitsch et al. 2011; Setegn et al. 2008). The swat model simulates land phase of the hydrologic cycle based on the balance equation (Arnold et al. 2012; Githui and Mutua 2009; Neitsch et al. 2011).where, SWt: the final soil water content(mm), SWo: the initial soil water content(mm), t: the time (days), Rday: the amount of precipitation on day(mm), Qsurf: the amount of surface runoff on day(mm), Ea: the amount of evapotranspiration on day(mm), Wseep: the amount of water entering the vadose zone from the soil profile on day (mm), and, Qgw = the amount of return flow on day (mm).

Runoff in SWAT model may be estimated by ether the soil conservation curve number (Mohammed and Thatiparthi 2020) or green and Ampt infiltration method (Green& Ampt, 1911). For this study, the curve number method was employed. because of it is efficient and most popularly used in estimation of runoff (Bewket and Woldeamlak 2002); Mengie et al. 2019) mainly based on the physical characteristics including the land use, soil and the slope of the study area and the hydrology condition (Githui and Mutua 2009; Githui et al. 2010; Tang et al. 2012). Soil conservation curve number method estimates the runoff based on the Eq. 2 (Narsimlu et al. 2015; Nicótina et al. 2008); Alemu 2013); Sloan and Sayer 2015).where, Rday: the amount of precipitation on day(mm), Qsurf: the amount of surface runoff on day(mm), s: retention parameter on day (mm).

The retention parameter(S) is given by the Eq. (3) (Bewket and Woldeamlak 2002; Setegn et al. 2008; Chaubey et al. 2005; Narsimlu et al. 2015).

The topography is defined by DEM, which describes the elevation of any point in a given area at a specific spatial resolution, which is used for watershed delineation. A 30 by 30-m resolution DEM was collected from Ministry of water, Irrigation and Energy of Ethiopia.

Soil data is one of the major input for SWAT model with inclusive and chemical properties (WaleWorqlul et al. 2018; Barbalho 2014; Pontes et al. 2016). The soil map of the study area was also obtained from Ministry of Water, Irrigation and Energy of Ethiopia. According to the FAO soil classification, the dominant soil for the Robigumero watershed was grouped as Calcic Vertisols, Eutric Leptosols, Eutric Vertisols, and Eutric cambisols and summarized in Table 1. To integrate the soil map with SWAT model, a user soil data base which contains textural and chemical properties of soils was prepared for each soil layers and added to the SWAT user soil data bases.

Meteorological data is needed by the SWAT model to simulate the hydrological conditions of the watershed. The meteorological data required for this study were collected from National Meteorological Agency of Ethiopia. The meteorological data collected were precipitation, maximum and minimum temperature, relative humidity, and wind speed and sunshine hours for four stations (Debrebirhan, Chacha, Deneba and Lemi) from the year 1988 -2018.In this study, the weather generating station was Debrebirhan rain gauge station. The monthly statistical weather parameters needed when WGEN was prepared from daily weather data are rainfall parameters (PCPMM, PCPSTD, PCPSKW, PCP_W1, PCP_W2, PCPD, RAINHHMX), temperature parameters (TMPMX, TMPMN, TMPSTDMX, TMPSTDMN), solar radiation parameters (SOLARAV), wind parameters (WNDAV) and dew point temperature parameters (DEWPT). The rainfall parameters were calculated by using pcpSTAT.exe, whereas the dew temperature parameters were calculated using dewp02.exe (Neitsch et al. 2011).where: dew is dew point temperature [0c], es is saturation vapor pressure [hpa], ea is actual vapor pressure [hpa], RF is relative humidity [%], T is temperature [0c] e = 2.7183 (base natural logarithm).

Data were missing from a particular gauge site or representative precipitation is necessary at a point of interest. There are different methods for filling the missing data from those methods station average and normal ratio method were used for the rainfall in this study (Rientjes et al., 2011). All of the rainfall recorded from the stations has missing data with ranging greater than 10% of missing. Therefore, before using the data to runoff modeling it was first essential to apply a gap filling techniques.where PX is the missing data at station x, Nx is the missing data stations normal annual rainfall, Ni is normal annual rainfall at station i, and n is number of nearby gauges The station-average method for estimating missing data uses n gages from a region to estimate the missing point rainfall Fig. 1.

The consistent record is one where the characteristics of the record have changed with time. Adjusting for gauge consistency involves the estimation of the effect rather than a missing value (Pontes et al. 2016; Richards 1998; Nicótina et al. 2008). For this study double mass curve method was used in order to estimate the consistency of four stations in the study area and as shown in Figs. 2, 3 below the station rainfall dates were consistent.

Homogeneity analysis was used to identify a change in the statistical property of the time series (Neitsch et al. 2011; Arnold et al. 2012). The cause may be either natural or manmade. Therefore, to select the representative metrological station for the analysis of areal rainfall estimation, checking the homogeneity of group is essential. The RAINBOW software is used based on the cumulative deviation from the mean (Wheater 2007; Neitsch et al. 2011).

The average rainfall over an area may be considered as the main input on the watershed modeling process, especially of those which deal with surface runoff because, the rain is the only climatic variable that can explain fast increasing flow (Anctil et al. 2006; Wheater 2007). According to Andréassian et al. 2001; Nicótina et al. 2008;Younger 2010), spatial variability of rainfall over the basin and their distribution pattern, as well as its interaction with the basin, have a considerable effect on runoff response generated. There are different methods used to calculate the mean annual rainfall which represents its distribution on the watershed (Tadele and Förch 2007; Chaubey et al. 2005). However, The Thiessen-polygon method is the best technique that shows the convergence for increasing the rain gauge density in the basin (Barbalho (2014). The average rainfall over the catchments was calculated as Equation below.where Pav is mean areal precipitation (mm), Pi is mean annual precipitation (mm) and Ai is coverage area at itℎ tℎe station, within Thiessen polygon respectively.

The observed daily streamflow data is the required data for calibration and validation of the simulated streamflow from the Watershed (Rientjes et al., 2011; Getachew and Melesse 2012; Githui et al. 2010). The streamflow in the Blue Nile basin including the Robigumero watershed was recorded by the Ministry of Water, Irrigation, and Energy (MoWIE). The available observed daily streamflow data recorded at Robigumero gauging station from 1990–2009 years was collected from the Ministry of Water, Irrigation, and Electricity.

Land use is also The most important factor that affects runoff, evapo-transpiration and surface erosion in a watershed (van Griensven et al. 2006;Pontes et al. 2016). There are many studies on land use and land cover change in the districts and catchments of the Blue Nile basin. These studies support this study in many aspects especially in the continuous expansion of farm land (WaleWorqlul et al. (2018). The Land use and land cover change studies usually need the development of land cover units before the analysis is started (Nicótina et al. 2008; Sloan and Sayer 2015). The Three different year’s land use/land cover data were collected from mister of water irrigation and energy.

A model sensitivity analysis can be help full in understanding which model input are the most important. Sensitivity analysis is a method of identifying the most sensitive parameters that significantly affects the model calibration and validation (Neitsch et al. 2011; Tang et al., 2012 ;Abbaspour, 2013). Sensitivity analysis describes how model output varies over a range of a given input variable (Khalid et al. 2016a, b; Welde and Gebremariam 2017b; Andualem and Gebremariam 2016).So that twenty-six flow, parameters were checked for sensitivity (Garzanti et al. 2006; Khalid et al. 2016a, b). For this study, the global sensitivity analysis was employed in SWAT-CUP 2012 and the p-value were used to select the sensitive parameters (Abbaspour 2012; Arnold et al. 2012).

Calibration is the process whereby model parameters are adjusted to make the model output match with the observed data (Rientjes et al. 2011). The period from 1990 to 2002 was used as a calibration period since the data for this period was with little missing data or representative data. Validation is the comparison of the model outputs with in independent data set without making any adjustment. The purpose of model validation is to check whether the model can predict flow for another range of period (Tang et al. 2012). The period from 2003 to 2009 was used as a validation period.

Model evaluation is an essential measure to verify the robustness of the model. In this study, two model evaluation methods were used, which were Nash–Sutcliffe efficiency (NSE) and coefficient of determination (R²) Barbalho (2014).where Si and Oi are simulated and observed values during model evaluation at time step ith respectively, O_min is the average observed value, and “n” is the number of values.

The coefficient of determination (R²) describes the proportion the variance in measured data by the model. It is the magnitude linear relationship between the observed and the simulated values. R² ranges from 0 (which indicates the model is poor) to1 (which indicates the model is good), with higher values indicating less error variance, and typical values greater than 0.6 are considered acceptable according to (Barbalho 2014).where Si and Oi are simulated and observed values during model evaluation at time step ith respectively, O_min and Smin is the average observed and simulated value, and “n” is the number of values Fig. 4.

The Three-land use cover maps of 1996, 2006 and 2016 were collected from minster of water irrigation and energy (Fig. 5). It is easily shown that there is an increase of agricultural land, and urbanization and decrease of forested areas, and grassland over 21 years. In general, during 21-year period the Agricultural land increases at about 22.4% whereas the forested area decreased by 5.3%. For the individual class area and change statistics for the three periods are summarized as follows (Table 2).

The land use land cover map of 1996 (Fig. 5) showed that the total agricultural land coverage was about 63.3% of the sub basin and increased rapidly to 85.7% of the Watershed in 2016 (Tables 3, 4 and Fig. 3). The reason is mainly the growth of the population that caused the increase in demand for new Agricultural land and settlement, which in turn resulted shrinking of other types of land use percentage of the watershed. On the other hand, the total forest coverage in 1996 was about 14.9% and then reduced to 9.6% in 2016. This was due to deforestation activities that have taken place for the purpose of agriculture, firewood and new settlement.

After sensitivity analysis has been done, the calibration of stream flow was done automatically. The result of calibration for the average monthly stream flow showed a very good agreement between observed and simulated stream flow (Fig. 6) with Nash –Sutcliffe simulation efficiency of 0.81 and coefficient of determination (R²) of 0.83.

After calibration was done manually and getting acceptable values of NSE and R², validation was checked using monthly-observed flows. The model validation also showed a very good agreement between simulated and measured monthly flow (Fig. 7) with the NSE value of 0.86 and R² 0.87.

The calibrated and validated stream flow result shows a very good agreement between observed and simulated stream flow. Therefore, the results of stream flows (Table 6) indicate that SWAT model is a very good predicator for stream flow of Robigumero Watershed.

Different studies that were conducted in the upper Blue Nile basin also showed similar result. For example, The SWAT model showed a good match between measured and simulated flow of Gumara watershed both in calibration and validation periods with (NSE = 0.76 and R² = 0.87) and (NSE = 0.68 and R² = 0.83), respectively (Awlachew, 2006).This indicates that SWAT can give sufficiently reasonable result in the upper Blue Nile basin. The following figure shows that the scatter plots of observed and simulated value for both calibration and validation (Fig. 8, 9). This shows good linear correlation between observed and simulated values.

This study assessed the impact of LULC change on streamflow in Robigumero watershed. Also, seasonal variability of streamflow was evaluated on wet (July, August, and September) and dry (Jan, Feb, and March) months. The simulation results of mean monthly streamflow for 1996, 2006 and 2016 LULC maps are shown in Fig. 10. The wet and dry mean monthly streamflow of 1996, 2006 and 2016 LULC and its variability during the study period are presented in Table 7, 8, 9. The results indicated that mean monthly streamflow was increased in the wet months (27.9%) and increase in dry months (1.9%) in the year 1996 and 2006 (Table 7, 8, 9). This was attributed to increase in the area under agriculture and decrease of forest land in the Robigumero watershed. This is due to rainfall satisfies soil moisture deficit more quickly in the agricultural land than forest there by generating more runoff in agricultural land. As a result, more runoff was generated due to streamflow in the year 2006 than 1996 (Fig. 11). Moreover, expansion in agricultural land decreased rainfall infiltrated into the soil and increase surface runoff. Therefore, the streamflow was increased in wet months and decreased in dry months. The streamflow was contributed more in wet months from surface runoff while in dry months, it was contributed more from groundwater. However, streamflow was increased in 2016 both in wet (16.2%) and dry (0.4%) seasons as compared to 2006 due to LULC change (Fig. 8). Besides, a slight decrease in land under grassland which contributed to increases of groundwater in the watershed. It generates more surface runoff in grassland due to less infiltration.

The result indicates that mean monthly streamflow was increased by 6% in the year 1996 to 2006 and 2.3% between the years 2006 to 2016 (Table 8). The dominant land cover in the year 2006 was agriculture and there was high agricultural expansion at the expense of other land use from the year 1996 to 2016. As a result, high runoff was generated during this period; this increases streamflow of 2006 as compared to 1996. In the year 2016, there was a further expansion of the land under agriculture and decrease of the grass land with slightly increase in forest land. Therefore, for the same reason, the streamflow was increased in 2016 as compared to 2006. Generally, during the study period, Robigumero watershed experienced an increase of streamflow due to extreme LULC change.