The Soil Water Assessment Tool (SWAT)
SWAT is a time-continuous semi-distributed catchment model developed by US Agricultural Research Service to evaluate the impact of land use, agricultural cultivation and climate on water quality and sediment yield of a given catchment (Arnold et al., 2012).
SWAT model uses daily time base data to simulate hydrological processes in a watershed and the watershed divides into multiple sub-watersheds depending upon drainage areas of the tributaries and these sub-watersheds are further divided into hydrological response units (HRUs) based on LULC, management, soil and topographical characteristics.
The hydrological simulation of a watershed in SWAT is done in two separate divisions: the first division is the land phase which controls the amount of water, sediment, nutrient and pesticide loading to the main channel in each sub-basin. The second division is the routing phase which can define the movement of water, sediment, nutrient and organic chemicals through the channel network to the watershed outlet.
In the land phase of hydrological cycle, SWAT simulates the hydrological cycle based on the water balance equation given by:
Where, SWt is final soil water content [mm], SWo is the initial soil water content on day i [mm], t is time in days, Pday is the amount of precipitation on day i [mm], Qsurf is the amount of surface runoff on day i [mm], Eoi is the evapotranspiration on day i, QReturni is amount of return flow on day i [mm] and Perci is the amount of percolation on day i.
Model Input Data
The Input data to setup SWAT model for UBN basin were Digital Elevation Model (DEM), LULC data, soil data, and daily data of climatic variables such as daily data of rainfall, maximum and minimum temperature, relative humidity, wind speed and solar radiation. Additionally, SWAT database files which developed for UBN and lake Tana reservoir parameters are also necessary inputs.
Digital Elevation Model (DEM)
DEM is the major input to perform watershed delineation in SWAT and DEM describe the elevation of a point at specific spatial resolution. A 90 m by 90 m resolution ASTER Global DEM was obtained from the NASA website. The DEM was projected to UTM (Universal Transverse Mercator) on spheroid of WGS84 datum. The UBN basin includes high mountains, rolling ridges, flat grass land areas and valleys. The elevation ranges from 489 m.a.s.l in North West direction near to Sudan border to 4261 m.a.s.l in north east direction.
Land use data
Land use is one of the main input data of the SWAT model to describe the Hydrological Response Units (HRUs) of the watersheds and it is one of the major factor which influence the hydrological property of the watersheds. Land use map of 1000m by 1000m resolution were obtained from the USGS land cover institute-GLC-2000. The most dominant land use in the UBN basin is cropland (agricultural land).
Soil Data
Soil data is the other major input data for the SWAT model with inclusive and chemical properties. The soil map of the study area with resolution of 1000m by 1000m was obtained from Food and Agriculture organization of the United Nation (FAO).
A user soil database which contains textural and chemical properties of soils was prepared for each soil layers and added to the SWAT user soil database to integrate the soil map with SWAT model for the UBN.
Weather Data
SWAT requires weather input data for the simulation. The weather data required for simulation includes daily data of precipitation, maximum and minimum temperature, relative humidity, wind speed and solar radiation. The meteorological stations in the UBN basin are limited due to the inaccessibility and economic limitation. For this reason, Global data from Potsdam Institute for Climate Impact Research (PIK) was collected for the whole Blue Nile basin. The total number of stations was 306 with 0.5 by 0.5ᴼ resolution. From the 306 stations, only 25 stations based on the default watershed delineation of the UBN basin was selected.
Hydrological Data
The stream flow data of the UBN basin is required for the calibration and validation of the model. The daily stream flow data (1961 to 2005) at El Diem (Sudan border) was Republic of Ethiopia, Minister of Water and Energy.
Model Setup
After preparing all the necessary input data for SWAT simulation, SWAT model using ArcSWAT2012 was built up for the UBN. The model set up involves: watershed delineation, HRU Analysis, Write Input Table, Edit SWAT Input, SWAT Simulation. After the model was run successfully parameter Sensitivity Analysis, Calibration and Uncertainty Analysis was also done. SUFI2 algorithm was used for uncertainty analysis. A predefined digital stream network layer was imported and superimposed onto the DEM to accurately delineate the location of the streams. The land use/land cover spatial data were reclassified into SWAT land cover/plant types. The watershed delineation process includes five major steps, DEM set-up, stream definition, outlet and inlet definition, watershed outlets selection and definition and calculation of sub-basin parameters. For the stream definition, the threshold-based stream definition option was used to define the minimum size of the sub-basin. The land use, soil and slope datasets were imported, overlaid and linked with the SWAT databases. Subdividing the sub watershed into hydrological response units (HRU’s), which are areas having unique land use, soil and slope combinations makes it possible to study the differences in evapotranspiration and other hydrological conditions for different land covers, soils and slopes.
Model calibration and uncertainty analysis
Model calibration is the process of adjustment of the model parameters to get a best fit between the observed and the model output by reducing prediction uncertainty. Automatic calibration was done on daily time steps stream flow data of the UBN basin at El Diem station from January 1986 to December 1993 and three year warm up period from 1983 to 1985 were used. Model validation was done from 1994 to 1999. Automatic calibration and uncertainty analysis for this study was done using SUFI2 algorithm which is based on a Bayesian framework and it also determines uncertainties through the sequential and fitting process.
Sequential Uncertainty Fitting—SUFI-2. SUFI-2 is the calibration algorithm developed by Abbaspour et al. (2004, 2007) for the calibration of SWAT model. In SUFI-2, parameter uncertainty accounts for all sources of uncertainties such as uncertainty in driving variables (e.g. rainfall), parameters, conceptual model and measured data (e.g. observed flow, sediment). In SUFI-2, the degree to which all uncertainties are accounted for is quantified by a measure referred to as the p-factor, which is the percentage of measured data bracketed by the 95% prediction uncertainty (95PPU). The 95PPU is calculated at the 2.5% and 97.5% levels of the cumulative distribution of an output variable obtained through Latin hypercube sampling. Latin hypercube sampling is used to draw independent parameter sets (Abbaspour et al., 2007). Another measure quantifying the strength of a calibration/uncertainty analysis is the so-called r-factor, which is the average thickness of the 95PPU band (r) divided by the standard deviation of the measured data. When acceptable values of r-factor and p-factor are reached, the parameter uncertainties are the desired parameter ranges.
Uncertainties related to model parameters and model structures are the two-major source of uncertainty for land cover change prediction models (Ferchichi et al., 2016). For this study uncertainty related to model parameters were assessed. SUFI2 gives several parameters sets which all yields flow; so those parameter sets were used to assess the range of uncertainty for two-hypothetical land use change scenarios. Python code was used to run the land use scenarios for each parameter set. The uncertainty of the predicted change also assesses by taking the minimum and maximum percent change for the difference between the baselines (calibrated model) and the two hypothetical land use scenarios of the average, minimum and maximum flows for all the parameter sets.
∆Min = Min flow for scenarios - Min flow for baseline
∆Avg = Average flow for scenarios - Average flow for baseline
∆Max = Max flow for scenarios - Max flow for baseline
∆Min, ∆Max and ∆Avg are calculated for each parameter sets between the minimum, maximum and average flows of the whole-time series for the two hypothetical scenarios and baseline, respectively.
The goodness of calibration and prediction uncertainty is judged based on the closeness of the p-factor to 100% (i.e. all observations bracketed by the prediction uncertainty), and the r-factor to 1 (i.e. achievement of rather small uncertainty band).
Model Performance Evaluation
The Nash-Sutcliffe Efficiency (NSE), Percentage Bias (PBIAS) and coefficient of determination (R2) are the most commonly used statistics for the evaluation hydrological model. All the statistical evaluation techniques are assessing the relation between the measured and best simulated streamflow of the calibration and validation (Yesuf et al., 2016).
Nash-Sutcliffe Efficiency (NSE):
$$\text{N}\text{S}\text{E}=1-\left[\frac{{\sum }_{\text{i}=1}^{\text{n}}{({\text{O}}_{\text{i}}-{\text{P}}_{\text{i}})}^{2}}{{\sum }_{\text{i}=1}^{\text{n}}{({\text{O}}_{\text{i}}-\stackrel{-}{\text{O}})}^{2}}\right]$$
Percentage of Bias (PBIAS):
Coefficient of determination (R2):
$${\text{R}}^{2}= {\left[\frac{{\sum }_{\text{i}=1}^{\text{n}}({\text{O}}_{\text{i}}-\stackrel{-}{\text{O}})({\text{P}}_{\text{i}}-\stackrel{-}{\text{P}})}{{\left[{\sum }_{\text{i}=1}^{\text{n}}{\left({\text{O}}_{\text{i}}-\stackrel{-}{\text{O}}\right)}^{2}\right]}^{0.5}{\left[{\sum }_{\text{i}=1}^{\text{n}}{({\text{P}}_{\text{i}}-\stackrel{-}{\text{P}})}^{2}\right]}^{0.5}}\right]}^{2}$$
Where, \({O}_{i}\)is the ith observed stream flow value (m3/s), \(\stackrel{-}{O}\) is the average of the measured streamflow for the whole period (m3/s), \({P}_{i}\) is the ith predicted flow value (m3/s), \(\stackrel{-}{P}\) is the average of the predicted flow for the whole period (m3/s) and n is the simulation period.
Land Use Change Scenarios
Land use change affects the catchment water resource either in quality or quantity. Studies are focused on the impact of land use change on the stream flow and sediment of the catchment (Narsimlu et al., 2015; Setegn et al., 2010). In UBN basin, there was a high expansion of agriculture throughout the basin since 1957. For this study, land use change prediction on the catchment stream flow was used for the analysis. Additionally, the seasonal stream flow variability due to LULC dynamics change will be addressed.
To assess the impact of LULC change, the calibrated model was run for two land use scenarios with constant topography and soil data from January 1983 to December 2000 in the UBN basin.
Scenario 1: Partial deforestation: -convert half of forested area in the Upper Blue Nile to agricultural land.
Scenario 2
completely change the forest cover in the Upper Blue Nile basin into agriculture. All the sub-basins which their dominant land use was forest converted to agriculture (94% forest cover from the total deciduous forest of UBN has changed to agriculture and the remaining 6% is dominated by agriculture which is not changed).