Modeling the Effect of Climate Change on Water Stored above a Micro-Dam in an Inland Valley Swamp in Sierra Leone, Using SWAT

Abstract

Many societies have experienced water scarcity resulting from population growth, increased urbanization and industrialization, increased irrigation associated with advances in agriculture productivity, desertification, global warming, or poor water quality. Climate change, and soil heterogeneity has a direct impact on the discharges of many rivers in and around the world. Various hydrological models have been used to characterize the impact of climate and soil properties on hydrology and water resources. The SWAT (Soil and Water Assessment Tool) water balance model, one such model, has been used at a variety of scales. In this instance it was used to model the impact of climate change on water storage in a reservoir at the downstream end of a small (75 ha) watershed. The watershed is the major component of an inland valley swamp, with a valley bottom that receives runoff from the watershed. The SWAT model was calibrated using storage data from 2014/15 and validated with data from 2015/16. Using future ensemble values derived from GCMs, the model predicted a reduction in the storage volume at the beginning of December of every dry season, with the 100-year storage volume down from 10,000 to 6900 cubic meters.

1. Introduction

Advances in economies and standards of living have resulted in a growing dependency on water resources across the globe. Many societies have experienced water scarcity resulting from population growth, increased urbanization and industrialization, increased energy use, increased irrigation associated with advances in agriculture productivity, desertification, global warming, and poor water quality. Improved understanding of how each of these factors influences water supply, demand, and quality requires an understanding of the underlying processes, and their impact, on water availability and use. Such an understanding is best facilitated by a holistic approach that integrates hydrologic processes at the watershed scale, to determine an overall watershed response to user demands, changing climates, or both [1].

The circulation of water within the earth and atmosphere is a complex interaction of energy exchange and transportation pathways [2]. Hydrology addresses the waters of the earth, their occurrence, circulation and distribution, their chemical and physical properties, and their reaction with the environment, including their relation to living things. It also deals with the relationship of water with the environment within each phase of the hydrologic cycle. Factors such as rapid urbanization, industrialization and deforestation, land cover change, irrigation, have precipitated changes in hydrologic systems. Climate change, and soil heterogeneity also has a direct impact on the discharges of many rivers. Different hydrologic phenomena and hydrologic cycles need to be thoroughly studied to characterize these impacts.

Various hydrological models have been developed to characterize the impact of climate and soil properties on hydrology and water resources. Each model has its own unique characteristics, but common inputs used by deterministic models include rainfall, air temperature, soil characteristics, topography, vegetation, hydrogeology and other physical parameters. These models can be applied in very complex and large basins [3]. They have been used to simulate water movement under different conditions. They are used to study, inter alia, the impact of climate change on water availability, land use change on river discharges, and agricultural management strategies on water availability and sediment yield [2]. Central to this effort, watershed modeling is being utilized as a tool to better understand surface and subsurface water movement and the interactions between these water bodies. More importantly, they offer tools to guide decision making on water resources, water quality, and related hazard issues [1].

Models are a type of tool and are used in combination with many other assessment techniques. Models reflect how watershed systems are conceived. As with any tool, the answers they give are dependent on how they are applied, and the quality of these answers reflects how well the systems are understood. Models are also useful for extrapolating from current conditions to potential future conditions. Indeed, it is not possible to monitor the future, so modeling is the default choice. The ability of models to predict future conditions is very useful for projecting the outcomes of various possible management measures and strategies (http://www.epa.gov/watertrain).

The effect of climate change on water resources can be best handled through simulation of the projected hydrological conditions under such change. Such a treatment is essential as hydrological response is a highly complex process governed by a large number of variables such as terrain, landuse, soil characteristics and the state of the moisture in the soil. The SWAT (Soil and Water Assessment Tool) water balance model is one such model that has been used to carry out the hydrologic modeling of river basins [4].

SWAT is a computationally efficient simulator of hydrology and water quality at various scales. It is a mechanistic continuous model that can handle large watersheds in a computationally efficient manner. The “Hydrologic Unit Model for the United States” (HUMUS) has already been used to simulate river discharges at approximately 6000 gauging stations in the United States, with good results [5, 6, 7]. This study was extended within the national assessment of the USDA Conservation Effect Assessment Project (CEAP, (http://www.nrcs.usda.gov/Technical/nri/ceap/ceapgeneralfact.pdf.). SWAT is a comprehensive model that requires a diverse set of input parameters. However, many of the input parameters that are used to simulate special features are not available for all watersheds. SWAT was developed to quantify the impact of land management practices on water, sediment, and agricultural chemical yields in large complex watersheds with varying soils, land use, and management conditions, over long periods of time. The main components of SWAT are hydrology, climate, nutrient cycling, soil temperature, sediment movement, crop growth, agricultural management, and pesticide dynamics [8].

Terink et al. [2] described SWAT as a distributed rainfall-runoff model that divides a catchment into smaller discrete calculation units for which the spatial variation of the major physical properties are limited, and hydrological processes can be treated as being homogeneous. The total catchment behavior is a net result of many sub-catchments. Soil and land cover data are used to generate unique combinations within each sub-catchment, and each combination is considered as a homogeneous physical entity, called a Hydrological Response Unit (HRU). The water balance for HRUs is computed on a daily time basis. Hence, SWAT disaggregates the river basin into units that have similar characteristics in terms of soil and land cover, and that are located in the same sub-catchment. SWAT is capable of predicting the effects of changes in climate and management conditions on basin hydrology. The model includes many modules, including crop growth, groundwater, and river routing the facilitate the required simulations.

The model is semi-distributed, dividing each subbasin into HRUs based on soil type, crop patterns, and management practices [9]. SWAT has produced favorable model results when evaluated on watersheds with a range of conditions in the U.S. and other countries as diverse as Morocco [2], Finland [10], Pakistan [11] and India [4]. Across many of these watersheds, SWAT has shown flexibility in simulating surface runoff. Gassman et al. [12] found that, because of its core strengths, SWAT is primarily used for calibration and/or sensitivity analysis, climate change impacts, GIS interface descriptions, hydrologic assessments, variation in configuration or data input effects, comparisons with other models or techniques, interfaces with other models, and pollutant assessments. The foundation strength of SWAT in many of these applications is the combination of simplified upland and channel processes that are incorporated into the model [12]. The incorporation of the curve number (CN) method and non-spatial HRUs supports model adaptation to virtually any watershed with a wide variety of hydrologic conditions. With a flexible framework, SWAT facilitates the calculation of Total Maximum Daily Loads (TMDLs) and simulation of a wide variety of conservation practices and other BMPs (e.g., fertilizer and manure application rate and timing, irrigation management, and flood prevention structures). Several studies illustrate the success of SWAT in conducting these types of simulations as part of an overall BMP assessment [1].

The object of this study was to use SWAT to determine the effect of projected climate change on water storage and availability in a small reservoir developed by damming the upper reaches of an inland valley swamp in Sierra Leone. The reservoir is used to grow a second rice crop that extends into the dry season. Information on storage can be used to determine irrigable area under different climate change scenarios.

2. Methodology

2.1 Description of study area

The study area is a small 75 ha watershed, located at Njala University, in the Southern Province of Sierra Leone. The watershed is a major component of an inland valley swamp, with a valley bottom that receives runoff from the watershed. In a typical swamp, the stream formed in the valley bottom may be perennial, or ephemeral. The watershed lies between latitudes 8.1140° and 8.1225° North and longitudes 12.0591° and 12.0668° west (Figure 1). The mean annual rainfall at Njala is 2200 mm. The annual Reference Evapotranspiration in 2015 and 2016 were 1322.3 and 1395.1 mm respectively.

Figure 1.

Map of study area.

There are two soils types in the watershed, Pelawahun loam and Njala sloping. Pelewahun soils occur in the drainageways of inland swamps that dissect the higher regions made up of Njala, Mokonde, and Bonjema soils. The terrain is nearly level to depressional, with concave slopes of 1 to 3 percent. Textures are loam to fine sandy loam in the topsoil, changing into clay loam in the upper subsoil, gravelly clay loam in the lower subsoil, and, as the bedrock is approached, clay [13].

2.2 SWAT model setup

SWAT requires data on terrain, land use, soil and weather for assessment of water yield at specified locations in a drainage basin. The selection of scale for spatial data is based on the tradeoff between the availability of required terrain data (in the form of contours data on the topographic maps) and the processing effort required for its preprocessing using the GIS interface. The following sections provide brief description of data elements used and preprocessing performed on them.

2.2.1 Digital elevation model

A digital elevation model (DEM) is a topographic surface made up of elevation values laid out in a regular grid pattern. The DEM was generated using data from a topographic survey conducted over the watershed. The survey was carried out with the aid of Total station. The x-y-z data were then processed with the Quantum GIS (QGIS) into the DEM. QGIS is a public domain GIS that is free, easy to use, and available for Windows, Apple, and Android computers (http://qgis.org/en/site/). It also has the ability to convert maps between coordinate reference systems on the fly.

2.2.2 Delineation of the river basins

Automatic delineation of the river basins is done by specifying the location of the required final outflow point on the DEM as the final pour/drainage point. The watershed was divided into sub-basins using an arbitrarily selected threshold value. The number of sub basins is a function of this threshold, that controls the drainage density of the automatically constructed drainage system and thereby, the number of sub-basins. The DEM along with the main channels are shown in Figure 2.

Figure 2.

Digital elevation model (DEM) with drainage network.

2.2.3 Land cover/land use layer

Maps that outlined the landuses in the watershed were produced. These maps were stored as shapes files in QGIS, and converted to Comma Separated Variable file (.csv) files for input into SWAT. The entire watershed was split into four different landuses: Grass range (RNCH), mixed forest (BUFL), Water (WATR) and agricultural row crops (AGRC). The land use map for the watershed is shown in Figure 3.

Figure 3.

Land use map for the watershed.

2.2.4 Soil layer

A shape file with the various soils in the watershed was also produced. Input tables were prepared, describing various parameters of the soil. These properties include bulk density, particle size distribution, water holding capacity, depth and number of layers of the soil. Two main types of soil were identified in the watershed; Njala sloping (Orthoxic Palehumult) and the Pelewahun loam (Typic Plinthaquult). The soil map is shown in Figure 4.

Figure 4.

Soil map for the watershed.

2.2.5 Weather data input preparation

Daily meteorological data were obtained from a WatchDog Weather Station (the 2009 ET Model, Spectrum Technologies Inc., USA) located next to the study area.

The weather input variables required included: solar radiation, wind speed, relative humidity, and precipitation, minimum and maximum temperatures. Weather data from November 2014 through January 2017 were used to calibrate and validate the model. Average data from nine climate change models were used to predict the change in stored water for mid-century (2046–2064) and the end of century (2081–2100).

2.2.6 Simulated volume of water in the pond

In SWAT, ponds and wetlands are impoundments that are located within the subbasin area. The .pnd file contains parameter information used to model the water, sediment and nutrient balance for ponds and wetlands. The key variables in the subbasin pond input file include: surface area (ha) and volume of water (m³) in ponds when filled to principal spillway (ha), surface area and volume of water stored in ponds when filled to the principal spillway.

2.2.7 Future climatic data

For evaluation of the impact of climate change on water quantity, future climate (monthly precipitation and temperature) data were obtained from Bias Corrected and Downscaled WCRP CMIP3 Climate Projections archive for 2046–2064 and 2081–2100 (http://gdo-dcp.ucllnl.org/downscaled_cmip3_projections/). Weather and seasonal forecasts show improved reliability and consistency, when the outputs from multiple models are combined. Hence, the averages of nine GCM projections were used (ensemble) to drive the SWAT model in this study, rather than using the outputs from a single GCM.

The models were corrected for bias by comparing data predicted by the models for 1980 to 2000, with observed data from a station in Sierra Leone (Figure 5). Mid-century and end of century data produced by the models were then corrected using the ratio of the slopes.

Figure 5.

Comparison of 20 years of simulated and historical rainfall for a 20-year period in Sierra Leone.

3. Result and discussion

3.1 Pond output data

The HRU impoundment output file (.wtr) contains summary information for ponds, wetlands and depressional/impounded areas in the HRUs. The file is written in spreadsheet format. The output data include: precipitation falling directly on the pond (PNDPCP) during the time step (mm), the depth of surface runoff entering the pond (PND_IN) during the time step, evaporation (PNDEVP) from the pond surface during the time step (mm) and volume of water (m³) in the pond (PNDVOL) at the end of time step.

3.2 Observed volume of water in the pond

The stage versus volume of water in the pond plot was used to determine the amount of water in pond at any point in time. A polynomial equation was used to model the relationship between stage (m) and volume of water. The equation was:

V = 9791 h² + 25.818 h – 119.05 (R² = 0.997)

Where,

V = Volume of water in pond (m³)

h = stage (m)

3.3 Coefficient of determination (R2) and Nash- Sutcliffe Efficiency (NSE)

The coefficient of determination (R²) for the simulated and observed data of the volume of water in the pond was determined. The NSE was calculated using Eq. (1)

$$NSE=1-\left[\frac{{\displaystyle {\sum }_{i=1}^n{\left(Y_i^{obs}-Y_i^{sim}\right)}^2}}{{\displaystyle {\sum }_{i=1}^n{\left(Y_i^{obs}-Y^{\mathit{mean}}\right)}^2}}\right]$$E1

Where, Y_i^obs is the ith observation for the constituent being evaluated, Y_i^sim is the ith simulated value for the constituent being evaluated, Y^mean mean is the mean of observed data for the constituent being evaluated, and n is the total number of observations.

3.4 Average rainfall for different periods

The monthly average precipitation totals and average temperatures watershed for the baseline period (1961–2000), mid-century period (2046–2064) and late-century period (2081–2100). The baseline period values were computed using observed data, while the future ensemble values were derived from the GCMs, and corrected for bias, using the ratio of the slopes shown in Figure 5. Baseline and projected climate data are shown in Figure 6.

Figure 6.

Baseline and projected climate data at site.

3.5 Model calibration

The model was calibrated by minimizing the root mean squared deviation (RMSD) between observed and simulated reservoir storage during the 2014/2015 dry cropping season. The two main factors affecting storage during this period was seepage and evaporation from the reservoir. The most sensitive parameter for evaporation in the model was the PNDEVCOEFF; the ratio of reservoir evaporation to potential evaporation, and PND_K, the hydraulic conductivity through the bottom of the pond. Both of these parameters were varied over the allowable ranges in the model, and the results were used to develop RSMD contours. These contours are shown in Figure 7.

Figure 7.

RMSD contours of pond evaporation coefficient and hydraulic conductivity.

Only certain combinations of the two parameters make physical sense. Under most conditions, PNDEVCOEFF varies between 0.6 and 0.8, with 0.7 being typically used as a default value. In this instance the minina band could not be extended down to these levels, as the model imposes a constraint on the PND_K, the hydraulic conductivity through the bottom of the reservoir. This maximum value is suitable for clay soils but is lower than values that are typical for loams, and sandy loams. SWAT was developed for US conditions where ponds are lined with a clay layer or some type of impermeable geotextile material. In this case, however, the bottom of the reservoir was unconsolidated Pelewahun loam. Given the constraints of the model, the calibrated value for PNDEVCOEFF was the value that gave the minimum RMSD value at the maximum allowable value of PND_K. The relationship between RMSD and PNDEVCOEFF when PND_K is fixed at 1 mm/hr. is given in Figure 8. Based on this figure, 1 mm/hr. and 1.75 were used for PND_K and PNDEVCOEFF, respectively, in future simulations.

Figure 8.

RMSD versus pond evaporation coefficient.

3.6 Model validation

Using the calibrated values of the coefficient of pond evaporation (PNDEVCOEFF) and hydraulic conductivity of pond (PND_K), the model was validated against the observed storage volume of 2015/16. The NSE and RMSE values were 0.82 and 1578.2 respectively.

3.7 Storage volume at the start of the dry season

The storage volumes at the start of December which typically marks the beginning of the dry season were fitted to cumulative distribution functions. The normal distribution was a better fit for 1981/2000, but the log normal better fitted the 2080/2099 period.

The baseline (1980–1999) and end of century periods (2081–2099) were utilized. The baseline data (1980–1999) had negative skewness (−0.09) and kurtosis of 1.22 and the end of century data had a skewness of 2.23 and a kurtosis of 5.89.

4. Discussion

The SWAT model has been found to be computationally efficient in its prediction and very useful because of its original design to assess the role of climate, topography, soil, and land use in the hydrologic response of a catchment [9, 14]. Several researches have shown satisfactory results in predicting impact of climate change on flows of streams, rivers and reservoirs [11, 15, 16].

In the calibration of the model, two parameters were considered: the coefficient of pond evaporation and the hydraulic conductivity (K). Although, in most cases K values range from 0.6 to 0.8, this parameter was calibrated at the maximum value of 1 for the model, as the model imposes a constraint on the hydraulic conductivity through the bottom of the reservoir. This could be attributed to the difference in soil texture in the bottom of ponds in the US where the model was developed as against the soil type in Sierra Leone for which the model is calibrated. Similarly, Beharry et al. [17] had to modify hydraulic conductivity as a key parameter in the calibration of the SWAT model.

Despite a high RMSE value for validation, the model also produced a good NSE value of 0.82. This indicates the model has a very high power of predicting the stored volume of water in the reservoir at any point in time. Figure 9 in indicative of a reduction in storage volume of water in the reservoir at any probability level, in the future (2081–2100) as compared to the baseline period (1981–2000). This reduction could be attributed to an increase in evaporation due to high temperatures. This would have a negative impact- reducing the acres of land that can be irrigated. Subsequently, Figure 9 also indicate that the 100-year storage volume will be reduced from a little over 10,000 m³ during the baseline period to less than 8000 m³ by the end of century (2081–2100). This would imply that designing storage for the dry season could involve less cost as storage volume is predicted to be reduced.

Figure 9.

Cumulative distribution functions for storage volume on December 1.

5. Conclusion

Using SWAT, the volume of water stored in a small reservoir was simulated. This was done considering the two main factors affecting storage during this period - seepage and evaporation from the reservoir. The most sensitive parameter for evaporation in the model was the ratio of reservoir evaporation to potential evaporation, and the hydraulic conductivity through the bottom of the pond. The model can be used to predict inflow into dams, water storage and release water management. For evaluation of the impact of climate change on water quantity, future climate (monthly precipitation and temperature) data were obtained from Bias Corrected and Downscaled WCRP CMIP3 Climate Projections archive for 2046–2064 and 2081–2100.

A knowledge of the expected storage volume in the reservoir at the start of the peak dry season is vital for proper planning of supplementary irrigation. Hydrologic modeling tools such as SWAT can help in predicting the hydrologic dynamics of watersheds. However, a challenge to utilizing such a tool will require a calibration process to determine the extent to which parameters can influence the predictive power of the model. The SWAT model was calibrated using the storage volumes of 2014/15 dry season and validated with the 2015/16 dry season data. With future ensemble values derived from the GCMs, the model predicted a reduction in the storage volume at the beginning of December of every dry season, with the 100-year storage volume down from 10,000 m³ to 6900 m³ (Figure 9).

The study highlighted the important role that future climate projections would play in understanding the effect of climate change on hydrologic processes. The use of models such as SWAT would be helpful in investigating land use change on nutrient loading and water quality in streams and rivers especially in sub-Saharan countries.