This study, conducted in the Black Volta basin of Ghana, determined how well TRMM Multi-Satellite Precipitation Analysis (TMPA) data compare with rain gauge measurements. The potential of using the TMPA data as inputs into a hydrological model for runoff simulation in a data-scarce basin was also assessed. Using a point-to-grid approach, accumulations of ground measured rainfall on daily, monthly and annual time scales were compared with accumulations derived from TMPA daily rainfall grids. The TMPA derived data together with other free global data were used as input into the soil and water assessment tool (SWAT) to set up a hydrological model for the basin. This model was calibrated and validated using streamflow data from a station located downstream of the basin. The study results showed a correlation from 0.85 to 0.92 for the monthly accumulated rainfall. Also, good Nash–Sutcliffe efficiencies of 0.94 and 0.67 were obtained for calibration and validation, respectively, on monthly scale. Moreover, simulation of streamflow was ‘satisfactory’ to ‘very good’ in terms of trends and residual variation. The study, therefore, shows that the use of satellite rainfall in the basin would be of great benefit considering the difficulties in accessing data across the basin.

INTRODUCTION

Hydrological modelling in data-scarce basins across the globe has always been a challenge. This situation is compounded when dealing with large-scale and trans-boundary basins. Many experts in the field of hydrology believe most of the catchments in the world are either ungauged or poorly gauged (Kundzewicz 2007). Hence, there have been initiatives by the International Association of Hydrological Sciences to improve on this situation with the introduction of the Prediction in Ungauged Basins (PUB) initiative (Sivapalan et al. 2003; Hrachowitz et al. 2013). Even though this initiative has yielded remarkable results (Franks 2005; Raju 2006; Sivapalan 2006), a great deal more still needs to be done in the area of the discovery of methodologies and data sources that can help in hydrological predictions and studies in poorly gauged basins.

The Volta basin is an important trans-boundary catchment in West Africa with very significant resources for its riparian countries. Even though this basin is important, adequate ground measurements of both meteorological and hydrological parameters are not available (GLOWA Volta Project (GVP) 2007; Taylor et al. 2007; Schuol et al. 2008), and in cases where there are measurements, they are often plagued with a great deal of missing data (Adjei et al. 2012). Also, access to data collected by institutions, researchers and other stakeholders from riparian countries is very difficult. Even though there have been some initiatives like the Global Change in the Hydrological Cycle (GLOWA) Volta project (Rodgers et al. 2007), the Volta Basin Authority and others to help bridge this gap, there still are challenges in respect of access to basin-wide hydrological and meteorological measurements. In recent times, various global data sets with durations significant for hydrological analysis, especially in poorly gauged catchments, have been made available. Current sources of some of these data sets are the Rainfall Estimates (USGS 2012), Global Land Data Assimilation System (Fang et al. 2009) and the TRMM Multi-Satellite Precipitation Analysis (TMPA) (Adler et al. 2007).

Satellite rainfall has been validated (Islam & Uyeda 2007; Adjei et al. 2012; Li et al. 2012; Yong et al. 2012) at different scales across the globe and has been used alone or as a complementary data source, satisfactorily, in some hydrological modelling studies (Tobin & Bennett 2009; Li et al. 2012, 2013). However, not much has been done in the use of satellite rainfall for hydrological studies in the Black Volta basin. Earlier studies in the basin obtained satisfactory results for the validation of the TMPA version 6 data set for the basin (Adjei et al. 2012). However, in this study, version 7 of the TMPA data set is used to set up a semi-distributed hydrological model for the Black Volta basin. Satellite-based rainfall studies are especially good for data-scarce basins because of their frequency of update (Li et al. 2013). Research carried out by Li et al. (2013) in the southern African region and Jung et al. (2012) in the Volta basin underscored the increasing use of large-scale hydrological models as the main assessment tool for global/regional water resources and the challenges associated with basins in Africa, which usually have sparse data and are poorly gauged. The availability of high quality and consistent data is usually a challenge making it difficult to set up regional models for water resources assessment. Thus, the aim of the research was: (1) to validate satellite-derived rainfall estimates by comparing them with ground-measured rainfall; and (2) to assess the suitability of using the satellite rainfall with other free global data sets as inputs into soil and water assessment tool (SWAT) to set up a hydrological model for runoff simulation in the Black Volta basin, which has been described as data scarce (Van de Giesen et al. 2001).

The SWAT model used in this study was generally developed for predicting 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. It is a continuous time model, semi-distributed and physically based (Arnold et al. 1998; Neitsch et al. 2011). Due to lack of data in some parts of the basin and the sparse distribution of rain gauges, rainfall estimates provided by the USGS were used to model streamflow in the Black Volta basin.

TMPA rainfall has been used for the hydrological model in a few reported studies including a study by Mango et al. (2011) in the trans-boundary Mara River basin, which is shared between Kenya and Tanzania in East Africa. In their study, Mango et al. (2011) demonstrated that a semi-distributed hydrological model such as SWAT can be set up and calibrated in a poorly gauged rural African catchment to yield useful results from satellite-based rainfall estimates, which can be used to study catchment changes and management scenarios useful for decision-making (Mango et al. 2011). Another study by Schuol & Abbaspour (2006), using data from CRU, demonstrated that global data sources could be used to calibrate and validate large-scale hydrological models in data-scarce regions like West Africa. The two studies above (i.e., Schuol & Abbaspour 2006; Mango et al. 2011) present an opportunity to further explore how well global data sources can be used for large-scale hydrological modelling and, even, in small-scale basins where data are scarce. Also, these two studies show the possibility of using global data sets for large-scale hydrological modelling at a finer spatial resolution. However, other critical inputs to the SWAT model such as land use/cover, digital elevation model (DEM) and others could also be improved with fine resolution data sets like the SRTM (US Geological Survey 2012), Advanced Spaceborne Thermal Emission and Reflection Radiometer (US Geological Survey 2012) and GlobCover 2009 (Bontemps et al. 2011) data sets. Thus, other methods and approaches could be applied to understand the best way for using these global data sets for hydrological modelling in data-scarce regions.

STUDY AREA AND DATA

The Black Volta basin, with a total area of about 155,000 km2 (Figure 1) at the hydrological station in Bamboi, is shared by four West African countries. These are Burkina Faso, Cote d'Ivoire, Ghana and Mali. It is also one of the three main sub-basins of the Volta basin. The average runoff from the Black Volta basin is about 243 m3/s and it contributes about 18% of the total flows to Lake Volta. The specific suspended sediment yield in this basin ranges from 8.0 to 12.0 tonnes/yr/km2. Current surface water use is estimated to be only about 0.03 m3/s for domestic water supply (FAO 1997; Barry et al. 2005).

Figure 1

Map of the study area showing the riparian countries and the Volta basin.

Figure 1

Map of the study area showing the riparian countries and the Volta basin.

The climate of the Black Volta basin is semi-arid to sub-humid with reference evaporation exceeding precipitation. There are two distinct seasons in the basin, namely the dry and rainy seasons, which are largely influenced by the West African monsoon (Global Water Partnersip 2012). The annual rainfall in the basin ranges from 600 mm to 1,348 mm while the reference evaporation ranges from 1,450 mm to 1,800 mm. The basin has a mean annual runoff of 7,673 × 106 m3 and a length of about 1,363 km (Barry et al. 2005).

The major land use in the basin is agriculture with food crop cultivation under extensive bush fallow. The major food crops include yam, cassava, maize, sorghum, millet, groundnuts and beans. Animal grazing on free range is a significant activity. In the northwest of the basin, lands are highly degraded both in terms of physical status and fertility levels, and can barely support meaningful crop cultivation (Barry et al. 2005).

To set up a SWAT model, the DEM, digital land cover/land use map, digital soil map with physical properties, meteorological data (precipitation, dew point or relative humidity, maximum and minimum temperature, potential evapotranspiration, if available on daily time scale) and stream flow measurements are required. Some of these required data can be simulated if a custom weather generator has been created for the basin. Hence, the available limited data obtained from the Ghana Meteorological Agency (GMA) were used to create a custom weather generator for the basin. In all, three meteorological stations, Bole, Sunyani and Wenchi (see Figure 1), which had measurements for the parameters required for the creation of the weather generator, were used. The other stations (Berekum, Bui, Damongo and Kintampo) were added to the three and used for the validation of the satellite-derived rainfall. To set up the model, free global data from various sources listed in Table 1 were used.

Table 1

Data and source

Data type Source Remarks 
Digital elevation model NASA Space Shuttle Radar Topography Mission (SRTM) 90 m × 90 m 
Land cover data European Space Agency (ESA)- GlobCover 2009 project 300 m × 300 m 
Soil data FAO-UNESCO Soil Map of the World (Batjes 19971:500,0000 scale 
Measured (rainfall, temperature, wind speed, relative humidity) Ghana Meteorological Service 1998–2010a 
Satellite rainfall (TMPA) USGS Data Portal 1998–2010 
Discharge data Ghana Hydrological Service 1998–2010a 
Data type Source Remarks 
Digital elevation model NASA Space Shuttle Radar Topography Mission (SRTM) 90 m × 90 m 
Land cover data European Space Agency (ESA)- GlobCover 2009 project 300 m × 300 m 
Soil data FAO-UNESCO Soil Map of the World (Batjes 19971:500,0000 scale 
Measured (rainfall, temperature, wind speed, relative humidity) Ghana Meteorological Service 1998–2010a 
Satellite rainfall (TMPA) USGS Data Portal 1998–2010 
Discharge data Ghana Hydrological Service 1998–2010a 

aWith missing values.

METHODOLOGY AND TOOLS

The data were obtained from the TRMM Online Visualization and Analysis System. The downloaded TMPA data were first compared with measured rainfall from some selected rain gauge (RNG) stations within the basin with minimal missing values before pre-processing them for use in ARCSWAT. A point-to-grid comparison was used to extract the daily time series from the data sets. Thus, the value for the grid within which a meteorological station is located was extracted and used for analyses. Using this approach, the time series data were extracted for the selected representative stations across the basin and compared with their respective RNG measurements.

The TMPA data set is produced as daily grids and has a spatial resolution of 0.25° × 0.25°. SWAT, however, does not accept gridded meteorological data but uses the stations closest to the centroid of a sub-basin as the meteorological station for that sub-basin. Thus, to use the TMPA data in SWAT, the centroids of the grids were calculated and used as virtual rainfall stations for the distributed nature of the satellite rainfall to be effectively utilized. SWAT requires other data sets in gridded formats; hence, the other required gridded inputs were obtained from freely available global sources (Table 1) and used for setting up the model.

The model calibration and validation were carried out using the sequential uncertainty fitting algorithm (SUFI-2) SWAT calibration and uncertainty procedures (SWAT-CUP) (Abbaspour et al. 2004, 2007; Schuol & Abbaspour 2006) with the aid of streamflow data from the Bui hydrological station. The data used for the modelling were from 1998 to 2010 (Table 1). The calibration of the model parameters was carried out using two approaches; namely, the value and relative approaches. A code was placed beside each parameter to be calibrated to indicate the type of change to be applied to the parameter. In the case of the parameters whose initial values were obtained at the hydrologic response unit (HRU) level based on soil, land use or slope map, the relative (r) change method was used to preserve the variability in the values across the different sub-basins. The value (v) change method was, however, used for those parameters having a single value for the basin. The ‘v’ change method means the existing parameter value is to be replaced by the given value while the ‘r’ change method means the existing parameter value is to be multiplied by (1+ a given value).

In the model validation process, the final parameter set of the simulation that gave the best fit and, thus, had the highest Nash–Sutcliffe efficiency (NSE) was used to re-run the SWAT model without adjusting the parameters. The performance of the model in both calibration and validation processes were evaluated using the recommended statistics by Moriasi et al. (2007). These recommended statistics are the NSE in Equation (1), percent bias (PBIAS) in Equation (2) and the root mean square error observations standard deviation ratio (RSR) in Equation (3). Their study (Moriasi et al. 2007) recommended ranges of these statistics (Table 2) that can be used to assess the performance of a calibrated model based on streamflow. 
formula
1
 
formula
2
 
formula
3
Table 2

General performance ratings of recommended statistics for a monthly time step (Moriasi et al. 2007)

Performance rating RSR NSE PBIAS (%) 
Very good 0.00 < RSR <0.50 0.75 < NSE <1.00 PBIAS < ±10 
Good 0.50 < RSR <0.60 0.65 < NSE <0.75 ± 10 < PBIAS < ±15 
Satisfactory 0.60 < RSR <0.70 0.50 < NSE <0.65 ± 15 < PBIAS < ±25 
Unsatisfactory RSR >0.70 NSE <0.50  PBIAS > ±25 
Performance rating RSR NSE PBIAS (%) 
Very good 0.00 < RSR <0.50 0.75 < NSE <1.00 PBIAS < ±10 
Good 0.50 < RSR <0.60 0.65 < NSE <0.75 ± 10 < PBIAS < ±15 
Satisfactory 0.60 < RSR <0.70 0.50 < NSE <0.65 ± 15 < PBIAS < ±25 
Unsatisfactory RSR >0.70 NSE <0.50  PBIAS > ±25 

RESULTS AND DISCUSSION

Comparison of daily rainfall

The results from the comparison of the daily rainfall from the TMPA data set and the rain gauge measurements show very weak correlations, ranging from 0.25 to 0.40, between the two data sets (Table 3). This poor correlation is as a result of the differences in the time of accumulation for the two different data sets. The daily accumulated TMPA begins at 00Z and ends at 21Z, which is equivalent to the UTC time zone. However, the ground measured accumulations are from 09:00am to 09:00am UTC. This difference in the accumulation periods is one of the main reasons for the very poor correlation between the data sets. This was further corroborated when the correlation coefficients between the two data sets improved when a 2-day running average was carried out on the two data sets (Table 3). Another possible source of error for the low daily correlations and underestimations may be due to the nature of rain that falls within the Volta basin, which is rainfall controlled by the Inter-tropical Convergence Zone. This kind of rain is mainly convective in nature and has high intensity, but of short duration, usually 20–60 min (Salako 2007; Van de Giesen et al. 2011). The low daily correlation difficulties observed are not really peculiar to this study but were also observed by Islam & Uyeda (2007), who found that the estimation patterns of satellite rainfall showed good similarity with the RNG values, but the numbers of deviations were very high. However, they attributed the deviations to the short life of the cloud cells and the frequency of satellite passes during the 24-hour period.

Table 3

Pearson correlation between rain gauge and TMPA satellite measurements

 Pearson correlation (R)  
Station Original data set 2-day running average Mean annual difference (mm)a 
Berekum 0.27 0.45 32 
Bole 0.34 0.48 −39 
Bui 0.25 0.44 −34 
Damongo 0.34 0.51 20 
Kintampo 0.32 0.48 −72 
Sunyani 0.36 0.52 16 
Wenchi 0.31 0.48 −32 
 Pearson correlation (R)  
Station Original data set 2-day running average Mean annual difference (mm)a 
Berekum 0.27 0.45 32 
Bole 0.34 0.48 −39 
Bui 0.25 0.44 −34 
Damongo 0.34 0.51 20 
Kintampo 0.32 0.48 −72 
Sunyani 0.36 0.52 16 
Wenchi 0.31 0.48 −32 

aMean annual difference = TMPA – RNG.

Comparison of monthly rainfall

The correlation coefficients for the monthly accumulations, which were obtained from the daily data set, ranged from 0.85 to 0.92 with a mean monthly rainfall of all the stations having a correlation coefficient of 0.97 (Table 4). These were tremendous improvements in comparison to the daily accumulation correlations in the previous section because the time of accumulation was not very significant on a monthly basis. This shows that whenever a TMPA data set is to be compared with rain gauge data, the time of accumulation (24-hour meteorological day definition) needs to be chosen appropriately to avoid arriving at any wrong conclusions in the two data sets.

Table 4

Correlation between monthly rain gauge and TMPA satellite accumulations

Station Pearson correlation (R) 
Berekum 0.87 
Bole 0.90 
Bui 0.85 
Damongo 0.90 
Kintampo 0.86 
Sunyani 0.92 
Wenchi 0.87 
Average 0.97 
Station Pearson correlation (R) 
Berekum 0.87 
Bole 0.90 
Bui 0.85 
Damongo 0.90 
Kintampo 0.86 
Sunyani 0.92 
Wenchi 0.87 
Average 0.97 

Also, in the monthly accumulations, the overestimations (positive) and underestimations (negative) cancel out to give an improved correlation. The stations at the middle part of the catchment (i.e., Bole and Damongo) showed relatively higher correlations in comparison to the stations at the lower belt, except for Sunyani station.

The long-term monthly mean data sets of both the TMPA and RNG follow the same trends as those observed in Figure 2. The TMPA overestimated, slightly, by 0.42 to 19.0 mm/month in seven months at a mean overestimation of 6.52 mm/month. This result indicates that the TMPA can be used to describe the long-term rainfall characteristics of the basin.

Figure 2

Plot of accumulated monthly rainfall of TMPA versus RNG.

Figure 2

Plot of accumulated monthly rainfall of TMPA versus RNG.

Both underestimations and overestimations were recorded at all the stations (Figure 3). However, the TMPA is more likely to overestimate in the basin. The number of months in which TMPA overestimated was higher than 50% at all the stations except at Kintampo station, which was slightly lower than 50% (Table 5). The bulk of underestimations or overestimations at most stations was below 30 mm with just a few stations above 30 mm.

Table 5

Comparison of TMPA and RNG performances for selected stations from 1998 to 2010

  Number of months within bias range for station 
 Bias range (mm)/month Berekum Bole Bui Damongo Wenchi Sunyani Kintampo Average 
Overestimation > 30 20 13 27 23 22 21 26 
 30–20 18 11 13 16 10 10 
 20–10 12 18 16 18 16 17 14 15 
 0–10 17 50 28 36 30 27 19 47 
Total number of months (count)  67 89 82 90 74 81 69 76 
Total number of months (%)  55.8 57.1 52.6 57.7 51.4 51.9 44.2 48.7 
Total number of months (count) Equal monthly rain 
Total number of months (%)  2.5 1.3 3.2 4.5 1.4 3.8 3.2 0.0 
Underestimation 0–10 10 19 16 19 13 27 20 48 
 20–10 13 15 13 18 20 14 
 30–20 10 11 
 > 30 19 25 33 24 33 23 33 11 
Total number of months (count)  50 65 69 59 68 69 82 80 
Total number of months (%)  41.7 41.7 44.2 37.8 47.2 44.2 52.6 51.3 
  Number of months within bias range for station 
 Bias range (mm)/month Berekum Bole Bui Damongo Wenchi Sunyani Kintampo Average 
Overestimation > 30 20 13 27 23 22 21 26 
 30–20 18 11 13 16 10 10 
 20–10 12 18 16 18 16 17 14 15 
 0–10 17 50 28 36 30 27 19 47 
Total number of months (count)  67 89 82 90 74 81 69 76 
Total number of months (%)  55.8 57.1 52.6 57.7 51.4 51.9 44.2 48.7 
Total number of months (count) Equal monthly rain 
Total number of months (%)  2.5 1.3 3.2 4.5 1.4 3.8 3.2 0.0 
Underestimation 0–10 10 19 16 19 13 27 20 48 
 20–10 13 15 13 18 20 14 
 30–20 10 11 
 > 30 19 25 33 24 33 23 33 11 
Total number of months (count)  50 65 69 59 68 69 82 80 
Total number of months (%)  41.7 41.7 44.2 37.8 47.2 44.2 52.6 51.3 
Figure 3

Average monthly rainfall of the TMPA versus RNG for all the stations from 1998 to 2010.

Figure 3

Average monthly rainfall of the TMPA versus RNG for all the stations from 1998 to 2010.

Comparison of annual rainfall

The results of the annual rainfall calculation for each station are shown in Tables 6 and 7. The result indicates both underestimation (negative bias) and overestimation (positive bias) of the annual RNG rainfall by the TMPA annual accumulations.

Table 6

Comparison of TMPA and RNG performances in the estimation of annual rainfall for some stations from 1998 to 2010

  Berekum Bole Bui Damongo 
Year TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias 
1998 1065 1024 4.0 885 950 − 6.8 965 833 15.9 874 716 22.0 
1999 1303 1102 18.3 1273 1163 9.4 1223 1219 0.3 1309 1275 2.7 
2000 1084 1056 2.6 1178 1226 − 4.0 1051 1150 − 8.6 1100 1117 − 1.5 
2001 1171 1125 4.1 915 904 1.2 1084 1052 3.0 889 747 19.0 
2002 1142 1363 − 16.2 1073 1153 − 6.9 1014 1303 − 22.2 1015 1010 0.5 
2003 1321 952 38.8 1161 1251 − 7.2 1130 1110 1.8 1234 1127 9.5 
2004 1270 1333 − 4.7 1110 1245 − 10.8 1225 981 24.9 1165 1085 7.4 
2005 1235 1183 4.4 1159 1187 − 2.3 1180 1328 − 11.1 1100 1073 2.5 
2006 1209 1319 − 8.3 1069 1108 − 3.5 1119 1048 6.9 990 973 1.7 
2007 1237 1256 − 1.5 1057 949 11.4 893 955 − 6.5 1229 996 23.3 
2008 1230   1086 1089 − 0.3 1077 1261 − 14.6 1141 1378 − 17.2 
2009 1088   1132 1135 − 0.3 1104 1121 − 1.5 989 1143 − 13.5 
2010 1351   1151 1393 − 17.3 1134 1285 − 11.7 1160 1290 − 10.1 
  Berekum Bole Bui Damongo 
Year TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias 
1998 1065 1024 4.0 885 950 − 6.8 965 833 15.9 874 716 22.0 
1999 1303 1102 18.3 1273 1163 9.4 1223 1219 0.3 1309 1275 2.7 
2000 1084 1056 2.6 1178 1226 − 4.0 1051 1150 − 8.6 1100 1117 − 1.5 
2001 1171 1125 4.1 915 904 1.2 1084 1052 3.0 889 747 19.0 
2002 1142 1363 − 16.2 1073 1153 − 6.9 1014 1303 − 22.2 1015 1010 0.5 
2003 1321 952 38.8 1161 1251 − 7.2 1130 1110 1.8 1234 1127 9.5 
2004 1270 1333 − 4.7 1110 1245 − 10.8 1225 981 24.9 1165 1085 7.4 
2005 1235 1183 4.4 1159 1187 − 2.3 1180 1328 − 11.1 1100 1073 2.5 
2006 1209 1319 − 8.3 1069 1108 − 3.5 1119 1048 6.9 990 973 1.7 
2007 1237 1256 − 1.5 1057 949 11.4 893 955 − 6.5 1229 996 23.3 
2008 1230   1086 1089 − 0.3 1077 1261 − 14.6 1141 1378 − 17.2 
2009 1088   1132 1135 − 0.3 1104 1121 − 1.5 989 1143 − 13.5 
2010 1351   1151 1393 − 17.3 1134 1285 − 11.7 1160 1290 − 10.1 
Table 7

Comparison of TMPA and RNG performances in the estimation of annual rainfall for some stations from 1998 to 2010

  Kintampo Sunyani Wenchi Average 
Year TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias 
1998 1093 1290 − 15.2 1091 914 19.4 1101 1053 4.6 1011 969 4.4 
1999 1359 1650 − 17.6 1296 1178 10.0 1367 1289 6.1 1304 1268 2.9 
2000 1272 1249 1.9 1079 1018 6.0 1097 1187 − 7.5 1123 1143 − 1.8 
2001 954 869 9.8 1223 1238 − 1.3 1179 987 19.4 1059 989 7.1 
2002 1342 1635 − 17.9 1239 1224 1.2 1284 1413 − 9.1 1158 1300 − 10.9 
2003 1308 1226 6.7 1344 1326 1.4 1306 1396 − 6.5 1258 1198 4.9 
2004 1403 1531 − 8.4 1270 1286 − 1.2 1264 1350 − 6.3 1244 1259 − 1.2 
2005 1350 1294 4.3 1203 1084 11.0 1201 1330 − 9.7 1204 1211 − 0.6 
2006 1064 1235 − 13.8 1201 1156 3.9 1219 1196 1.9 1125 1148 − 2.0 
2007 1196 1358 − 11.9 1307 1459 − 10.5 1356 1216 11.5 1182 1170 1.0 
2008 1345 1112 20.9 1281 1327 − 3.4 1228 1311 − 6.3 1193 1246 − 4.3 
2009 988 1088 − 9.2 1111 1307 − 15.0 1052 1201 − 12.4 1063 1166 − 8.9 
2010 1286   1478 1404 5.3 1404 1549 − 9.3 1266 1384 − 8.6 
  Kintampo Sunyani Wenchi Average 
Year TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias TMPA RNG %Bias 
1998 1093 1290 − 15.2 1091 914 19.4 1101 1053 4.6 1011 969 4.4 
1999 1359 1650 − 17.6 1296 1178 10.0 1367 1289 6.1 1304 1268 2.9 
2000 1272 1249 1.9 1079 1018 6.0 1097 1187 − 7.5 1123 1143 − 1.8 
2001 954 869 9.8 1223 1238 − 1.3 1179 987 19.4 1059 989 7.1 
2002 1342 1635 − 17.9 1239 1224 1.2 1284 1413 − 9.1 1158 1300 − 10.9 
2003 1308 1226 6.7 1344 1326 1.4 1306 1396 − 6.5 1258 1198 4.9 
2004 1403 1531 − 8.4 1270 1286 − 1.2 1264 1350 − 6.3 1244 1259 − 1.2 
2005 1350 1294 4.3 1203 1084 11.0 1201 1330 − 9.7 1204 1211 − 0.6 
2006 1064 1235 − 13.8 1201 1156 3.9 1219 1196 1.9 1125 1148 − 2.0 
2007 1196 1358 − 11.9 1307 1459 − 10.5 1356 1216 11.5 1182 1170 1.0 
2008 1345 1112 20.9 1281 1327 − 3.4 1228 1311 − 6.3 1193 1246 − 4.3 
2009 988 1088 − 9.2 1111 1307 − 15.0 1052 1201 − 12.4 1063 1166 − 8.9 
2010 1286   1478 1404 5.3 1404 1549 − 9.3 1266 1384 − 8.6 

The overestimations and underestimations were mostly below 20% of the RNG measured rainfall with just a few exceeding 20% at some stations. These results show that the TMPA data can be used to describe the long-term climate of a basin when gauged data are not available since the biases were relatively small in most of the years. The correlation between the two data sets can be improved through error modelling in the catchment based on a dense network or wide distribution of meteorological stations.

Selection of pre-calibration parameters

The selection of pre-calibration parameters was done through a review of the literature on the application of SWAT in different catchments across the globe and, especially, in catchments, which have characteristics similar to the Black Volta basin (Schuol et al. 2008; Rouhani et al. 2009; Fadil et al. 2011; Mango et al. 2011; Nie et al. 2011; Obiero et al. 2011; Güngör & Göncü 2013). Further to the selection of the parameters, the SUFI-2 of the SWAT-CUP tool which uses the Latin Hypercube (McKay et al. 1979) sampling method for parameter calibration was used to narrow down the number of parameters for model calibration. Many of the parameters used to calibrate SWAT, like curve number (CN2), available water capacity (AWC) or the bulk densities (BD) are closely related to different land use classes and soil textures; hence, their parameter values change accordingly. In all, a total of 28 parameters (13 global parameters, one parameter with a separate value for each of the six dominant land use types and three parameters with a separate value for each of the dominant soil textures, i.e., AWC, K and Z for sand, sandy loam and sandy clay loam soils (Table 9)) were selected for the calibration of the model after the sensitivity approach.

Sensitivity analysis of parameters

The sensitivity of the parameters used in the modelling are summarized in Table 8. The ranking of the parameters are based on t-stat from the sensitivity analysis of all the parameters for the whole basin. Parameter sensitivities were determined by calculation, using the multiple regression system, which regresses the Latin Hypercube generated parameters against the objective function values. The t-stat was used as a measure of sensitivity (larger in absolute values are more sensitive) and the p-values determined the significance of the sensitivity (values close to zero have more significance) (Abbaspour 2011). Analyses of the sensitivity results (Table 8) show the most sensitive parameters were the SCS curve number, soil available water storage capacity and the soil evaporation compensation factor. These parameters go to show that more emphasis should be placed on getting very good land use/land cover and soil properties data for the basin since they have a high impact on the model performance.

Table 8

Results of sensitive analysis of calibration parameters with their sensitivity ranking

Parameter type Definition Sensitivity rank 
Global 
SURLAGa Surface runoff lag coefficient [days] 15 
CH_K2a Effective hydraulic conductivity in the main channel [mm/h] 10 
CH_N2a Manning's n value for the main channel 12 
ESCOa Soil evaporation compensation factor [–] 
SOL_BDr Moist soil BD [g/cm3
ALPHA_BFa Baseflow alpha factor [day] 
GW_DELAYa Groundwater delay time: lag between the time that water exits the soil profile and enters the shallow aquifer [days] 14 
GW_REVAPa Groundwater ‘revap’ coefficient [–] 
GWQMNa Threshold depth of water in the shallow aquifer required for return flow to occur [mm H2O] 
RCHRG_DPa Deep aquifer percolation fraction [–] 13 
REVAPMNa Threshold depth of water in the shallow aquifer for ‘revap’ or percolation to the deep aquifer [mm H2O] 11 
Land use/cover based   
CN2b SCS runoff curve number [–] 
Soil type based   
SOL_AWCb Soil available water storage capacity [mm H2O/mm soil] 
SOL_Kr Soil conductivity [mm/h] 
SOL_Zb Soil depth [mm] 10 
Parameter type Definition Sensitivity rank 
Global 
SURLAGa Surface runoff lag coefficient [days] 15 
CH_K2a Effective hydraulic conductivity in the main channel [mm/h] 10 
CH_N2a Manning's n value for the main channel 12 
ESCOa Soil evaporation compensation factor [–] 
SOL_BDr Moist soil BD [g/cm3
ALPHA_BFa Baseflow alpha factor [day] 
GW_DELAYa Groundwater delay time: lag between the time that water exits the soil profile and enters the shallow aquifer [days] 14 
GW_REVAPa Groundwater ‘revap’ coefficient [–] 
GWQMNa Threshold depth of water in the shallow aquifer required for return flow to occur [mm H2O] 
RCHRG_DPa Deep aquifer percolation fraction [–] 13 
REVAPMNa Threshold depth of water in the shallow aquifer for ‘revap’ or percolation to the deep aquifer [mm H2O] 11 
Land use/cover based   
CN2b SCS runoff curve number [–] 
Soil type based   
SOL_AWCb Soil available water storage capacity [mm H2O/mm soil] 
SOL_Kr Soil conductivity [mm/h] 
SOL_Zb Soil depth [mm] 10 

aExisting parameter value is to be replaced by the given value.

bExisting parameter value is multiplied by (1+ a given value).

Calibration and validation

Using the NSE as the main evaluation criteria, the model was calibrated using monthly observed flows at the Bui discharge station. Although inputs in the model were on a daily time scale the model was calibrated and validated using the monthly outputs. Also, the quality of the final parameter ranges and the fitted values for the last iteration of SUFI-2 are shown in Table 9. The values obtained are consistent with results obtained from SWAT modelling in some major basins in West Africa (Schuol et al. 2008).

Table 9

The 28 SWAT model parameters included in the final calibration, their fitted value and final ranges

   Final parameter range 
 Parameter name Fitted value Min. value Max. value 
CN2______AGRLa − 0.18 − 0.31 0.08 
CN2______CRGRa − 0.67 − 0.71 −0.47 
CN2______SHRBa − 0.03 − 0.20 0.03 
CN2______SAVAa 0.10 − 0.11 0.37 
CN2______FODBa − 0.53 − 0.79 −0.41 
CN2______GRASa 0.28 0.04 0.32 
ALPHA_BFb 0.52 0.40 0.62 
GW_DELAYb 48 43 105 
GWQMNb 234 400 
10 GW_REVAPb 0.13 0.12 0.17 
11 RCHRG_DPb 0.82 0.33 0.87 
12 REVAPMNb 305 303 381 
13 ESCOb 0.45 0.12 0.45 
14 CH_K2b 114 79 121 
15 SOL_Z____SANDY_CLAY_LOAMa − 0.05 − 0.07 0.06 
16 SOL_Z____SANDY_LOAMa − 0.17 − 0.20 −0.03 
17 SOL_Z____SANDa − 0.31 − 0.46 −0.30 
18 SOL_AWC____SANDa − 0.20 − 0.25 0.00 
19 SOL_AWC____SANDY_LOAMa 0.10 − 0.10 0.17 
20 SOL_AWC____SANDY_CLAY_LOAMa − 0.12 − 0.19 −0.10 
21 SOL_K____SANDa 0.28 0.25 0.71 
22 SOL_K____SANDY_LOAMa − 0.11 − 0.21 −0.04 
23 SOL_K____SANDY_CLAY_LOAMa 0.02 − 0.16 0.13 
24 SURLAGb 2.0 0.09 2.21 
25 HRU_SLPa 0.20 0.13 0.40 
26 OV_Na − 0.32 − 0.35 −0.07 
27 SLSUBBSNa − 0.37 − 0.39 −0.21 
28 SOL_BDa 0.39 0.13 0.51 
   Final parameter range 
 Parameter name Fitted value Min. value Max. value 
CN2______AGRLa − 0.18 − 0.31 0.08 
CN2______CRGRa − 0.67 − 0.71 −0.47 
CN2______SHRBa − 0.03 − 0.20 0.03 
CN2______SAVAa 0.10 − 0.11 0.37 
CN2______FODBa − 0.53 − 0.79 −0.41 
CN2______GRASa 0.28 0.04 0.32 
ALPHA_BFb 0.52 0.40 0.62 
GW_DELAYb 48 43 105 
GWQMNb 234 400 
10 GW_REVAPb 0.13 0.12 0.17 
11 RCHRG_DPb 0.82 0.33 0.87 
12 REVAPMNb 305 303 381 
13 ESCOb 0.45 0.12 0.45 
14 CH_K2b 114 79 121 
15 SOL_Z____SANDY_CLAY_LOAMa − 0.05 − 0.07 0.06 
16 SOL_Z____SANDY_LOAMa − 0.17 − 0.20 −0.03 
17 SOL_Z____SANDa − 0.31 − 0.46 −0.30 
18 SOL_AWC____SANDa − 0.20 − 0.25 0.00 
19 SOL_AWC____SANDY_LOAMa 0.10 − 0.10 0.17 
20 SOL_AWC____SANDY_CLAY_LOAMa − 0.12 − 0.19 −0.10 
21 SOL_K____SANDa 0.28 0.25 0.71 
22 SOL_K____SANDY_LOAMa − 0.11 − 0.21 −0.04 
23 SOL_K____SANDY_CLAY_LOAMa 0.02 − 0.16 0.13 
24 SURLAGb 2.0 0.09 2.21 
25 HRU_SLPa 0.20 0.13 0.40 
26 OV_Na − 0.32 − 0.35 −0.07 
27 SLSUBBSNa − 0.37 − 0.39 −0.21 
28 SOL_BDa 0.39 0.13 0.51 

aExisting parameter value is multiplied by (1+ a given value).

bExisting parameter value is to be replaced by the given value.

The performance of the model (Table 10) according to the main efficiency criteria used can be described as very good (NSE of 0.94). The other performance indicators also show the model to be performing very well. The calibrated model showed marginal (PBIAS 3.34%) underestimation of the observed flows.

Table 10

Summary results for SWAT model calibration and validation using TMPA data

Criteria Calibration Validation 
R2 0.94 0.69 
NSE 0.94 0.67 
RSR 0.25 0.57 
PBIAS (%) 3.34 17.87 
Criteria Calibration Validation 
R2 0.94 0.69 
NSE 0.94 0.67 
RSR 0.25 0.57 
PBIAS (%) 3.34 17.87 

The performance of the model validation, according to the criteria used, can be described as good since the NSE is 0.67. However, the RSR, PBIAS and the R2 obtained for the model puts its performance as satisfactory. Since the main criterion used in the calibration of the model was the NSE, the performance of the model, in general, can be described as good. The model's ability to simulate very high flows during the validation period was not very good since it underestimated the very high peaks for the two events recorded (Figure 4). This observation is not unique to this study but has been observed in similar studies where the TMPA data set was used. For example, a study by Li et al. (2012) found that a model with TMPA data set input is unable to simulate the peak flows very well like the model with rain gauge measured rainfall as input. The model underestimated the mean annual flow by 10% of the 292 m3/s observed flow.

Figure 4

A plot of monthly calibration and validation results for the Bui station.

Figure 4

A plot of monthly calibration and validation results for the Bui station.

The simulated flows showed a narrower standard deviation of 389 m3/s compared to 462 m3/s for the observed flows. This variation is due to the model's inability to simulate very well the highs flows, hence, the narrower standard deviation.

The overall performance of the model in the estimation of the flows above 100 m3/s during the validation can be described as very good when compared with the low flows (Figures 5 and 6). This observation is as a result of the objective function used, which is more inclined towards high flows (Legates & McCabe 1999; Moriasi et al. 2007). However, the impact of the contribution of the low flows to the total streamflow compared to that of the high flows is less significant. Thus the use of NSE as the main evaluation criterion is still appropriate. The representation of groundwater flow in SWAT is simplified; hence, the appropriate calibration of low flows is, sometimes, a challenge for basins, which have significant periods where flows are mainly as a result of groundwater contribution.

Figure 5

Scatter plot of average observed monthly discharge against average simulated monthly discharge for the Bui station during calibration.

Figure 5

Scatter plot of average observed monthly discharge against average simulated monthly discharge for the Bui station during calibration.

Figure 6

Scatter plot of average observed monthly discharge against average simulated monthly discharge for the Bui station during validation.

Figure 6

Scatter plot of average observed monthly discharge against average simulated monthly discharge for the Bui station during validation.

As can be observed in Figures 4 and 6, the model is not good at simulating very high flows. Therefore, flows above 2,000 m3/s, which were two points out of the 60 points used for the validation, were excluded in the computation of the performance indicators to obtain the following: R2 (0.71), RSR (0.55), PBIAS (7.69) and NSE (0.70). Marginal improvements were observed in all the performance indicators except the PBIAS, which showed significant improvement from 17.87 to 7.69%. The changes observed in the indicators show that the model performance in the simulation of the overall flow is good. Hence it can be used for assessment of catchment response in cases where interest is not in the simulation of very high flows.

CONCLUSION

This study evaluated how well satellite-derived rainfall estimates compare with ground-measured rainfall. It also assessed the potential of using the satellite-derived rainfall together with other free global data sets as inputs into SWAT for runoff simulation in the Black Volta basin, which is a large-scale trans-boundary basin in a data-scarce region.

Analysis of TMPA data set revealed unsatisfactory daily correlation between the rain gauge measurements and the satellite-derived rain estimates. However, there was an appreciable improvement in the correlations when the monthly accumulations were used. The correlations could be described as ‘very good’, with monthly correlation coefficients ranging from 0.85 to 0.92. On the other hand, analysis of the annual rainfall for the stations showed that the TMPA both overestimated and underestimated the rain gauge measurements at all stations in the various years with no clear trend observed.

The use of the satellite-derived rainfall estimates as input into the SWAT model for the basin indicated that a model set-up using the data would not be good for flood analysis since its peak discharge estimates were not very good. However, the model showed indications of ‘very good’ to ‘satisfactory’ performance according to the PBIAS and the RSR at the downstream station. Also, the model simulation of the overall streamflow was ‘good’ to ‘very good’ in terms of trends and magnitude (R and NSE). Therefore, the use of satellite data in the study basin for water resources assessment would be of great benefit considering the difficulties in accessing data both locally and from the other riparian countries. The methods applied in this study can also be replicated in other data-scarce basins across the globe for hydrological studies.

This study had some limitations, which included our inability to access ground meteorological measurements across the whole basin. Also, some of the discharge data obtained had some reliability issues during verification and cross-validation and, thus, only one station which was found to be reliable was used for the model calibration and validation. These limitations made it difficult to set up a model solely based on ground measurements for comparison with that of the satellite-based model.

One of the ways which can greatly help in the improvement of satellite-derived rainfall estimates, to enable its direct application in hydrological modelling, is to encourage a great deal more localized validation of the rainfall estimates. This will help in the assessment of the performance of the estimates in different climatic zones and regions. This will also lead to the development of more refined algorithms incorporating more localized data and conditions leading to better estimates.

ACKNOWLEDGEMENTS

This work was supported by the Innovative Research Team Project under Grant No. 2009585412 by Basic Research Funds for National University at State Key Laboratory of HYDROLOGY-Water Resources and Hydraulic Engineering and also by the Ministry of Education and State Administration of Foreign Experts Affairs, P. R. China through the 111 Project under Grant No. B08048. The images and data used in this study were acquired using the Goddard Earth Sciences (GES)-Data and Information Services Center (DISC) Interactive On-line Visualization and Analysis Infra-structure (Giovanni) as part of NASA's GES DISC. We are also grateful to the GMA for providing us with the rainfall data and the European Space Agency TIGER II Capacity Building Facility for Africa for supporting this project through training and data facilitation.

REFERENCES

REFERENCES
Abbaspour
K. C.
2011
SWAT-CUP4: SWAT Calibration and Uncertainty Programs – A User Manual
.
Swiss Federal Institute of Aquatic Science and Technology
,
Eawag, Duebendorf, Switzerland
.
Abbaspour
K. C.
Yang
J.
Maximov
I.
Siber
R.
Bogner
K.
Mieleitner
J.
Zobrist
J.
Srinivasan
R.
2007
Modelling hydrology and water quality in the pre-alpine/alpine Thur watershed using SWAT
.
J. Hydrol.
333
,
413
430
.
Adjei
K. A.
Ren
L.
Appiah-Adjei
E. K.
Kankam-Yeboah
K.
Agyapong
A. A.
2012
Validation of TRMM data in the Black Volta Basin of Ghana
.
J. Hydrol. Eng.
17
,
647
654
.
Adler
R.
Braun
S.
Stocker
E.
Marius
J.
2007
Tropical Rainfall Measuring Mission, TRMM, Senior Review Proposal
.
Laboratory for Atmospheres, NASA Goddard Space Flight Centre
,
Greenbelt, MD
.
Arnold
J. G.
Srinivasan
R.
Muttiah
R. S.
Williams
J. R.
1998
Large area hydrologic modeling and assessment part I: model development 1
.
J. Am. Water Resour. Ass.
34
,
73
89
.
Barry
B.
Obuobie
E.
Andreini
M.
Andah
W.
Pluqet
M.
2005
The Volta River Basin: Comprehensive Assessment of Water Management in Agriculture Comparative Study on River Basin Development and Management
.
International Water Management Institute (IWMI)
,
Accra
.
Bontemps
S.
Defourny
P.
Bogaert
E.
Van, Arino
O.
Kalogirou
V.
Perez
J. R.
2011
GLOBCOVER 2009 Products Report Description and Validation Report
.
Fadil
A.
Hassan
R.
Kaoukaya
A.
Kharchaf
Y.
Bachir
O. A.
2011
Hydrologic modeling of the Bouregreg watershed (Morocco) using GIS and SWAT model
.
J. Geogr. Inform. Syst.
3
,
279
289
.
Fang
H.
Beaudoing
H. K.
Rodell
M.
Teng
W. L.
Vollmer
B. E.
2009
Global Land Data Assimilation System (GLDAS) Products, Services and Application from NASA Hydrology Data and Information Services Center (HDISC)
. In:
ASPRS 2009 Annual Conference. American Society of Photogrammetry and Remote Sensing
,
Baltimore, MD
.
FAO
1997
FAO Land and Water Bulletin
,
4th edn
.
Irrigation potential in Africa: A basin approach. Food and Agriculture Organization of the United Nations
.
Rome
.
Franks
S. W.
2005
Predictions in Ungauged Basins: International Perspectives on the State of the Art and Pathways Forward
.
IAHS Publication Series. IAHS Press
,
Wallingford
.
Global Water Partnership
2012
.
GLOWA Volta Project (GVP)
2007
Partners' Capacity Needs Assessment Workshop
.
Accra
.
Hrachowitz
M.
Savenije
H. H. G.
Blöschl
G.
McDonnell
J. J.
Sivapalan
M.
Pomeroy
J. W.
Arheimer
B.
Blume
T.
Clark
M. P.
Ehret
U.
Fenicia
F.
Freer
J. E.
Gelfan
A.
Gupta
H. V.
Hughes
D. A.
Hut
R. W.
Montanari
A.
Pande
S.
Tetzlaff
D.
Troch
P. A.
Uhlenbrook
S.
Wagener
T.
Winsemius
H. C.
Woods
R. A.
Zehe
E.
Cudennec
C.
2013
A decade of predictions in ungauged basins (PUB) — a review
.
Hydrol. Sci. J.
58
,
1198
1255
.
Kundzewicz
Z. W.
2007
Prediction in ungauged basins – a systemic perspective
. In:
Predictions in Ungauged Basins
.
IAHS Publications
,
Wallingford
,
309
, pp.
38
44
.
McKay
M. D.
Beckman
R. J.
Conover
W. J.
1979
Comparison of three methods for selecting values of input variables in the analysis of output from a computer code
.
Technometrics
21
,
239
245
.
Moriasi
D. N.
Arnold
J. G.
Van Liew
M. W.
Bingner
R. L.
Harmel
R. D.
Veith
T. L.
2007
Model evaluation guidelines for systematic quantification of accuracy in watershed simulations
.
Trans. ASABE
50
,
885
900
.
Neitsch
S. L.
Arnold
J. G.
Kiniry
J. R.
Williams
J. R.
2011
Soil & Water Assessment Tool Theoretical Documentation Version 2009. Texas Water Resources Institute Technical Report No. 406
.
College Station
,
TX
.
Obiero
J. P. O.
Hassan
M. A.
Gumbe
L. O. M.
2011
Modelling of streamflow of a catchment in Kenya
.
J. Water Resour. Protect.
3
(
9
),
667
677
.
Raju
K. S.
2006
Predictions In Ungauged Basins For Sustainable Water Resources Planning And Management
.
Jain Brothers
,
New Delhi
.
Rodgers
C.
Vlek
P. L. G.
Eguavoen
I.
Amtz
C.
(eds)
2007
GLOWA Volta Phase II Completion Report
.
GLOWA Volta
,
Bonn
.
Salako
F. K.
2007
Temporal variation of rainfall erosivity in southern Nigeria
.
ASSET
7
,
190
202
.
Sivapalan
M.
(ed.)
2006
Predictions in Ungauged Basins: Promise and Progress
.
IAHS Publication No. 303, IAHS Press
,
Wallingford
.
Sivapalan
M.
Takeuchi
K.
Franks
S. W.
Gupta
V. K.
Karambiri
H.
Lakshmi
V.
Liang
X.
Mcdonnell
J. J.
Mendiondo
E. M.
Connell
P. E. O.
2003
IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012: Shaping an exciting future for the hydrological sciences
.
Hydrol. Sci. J.
48
,
857
880
.
Taylor
J. C.
Van de Giesen
N.
Steenhuis
T. S.
2007
West Africa: Volta discharge data quality assessment and use
.
J. Am. Water Resour. Ass.
42
,
1113
1126
.
US Geological Survey
2012
Earth Explorer. http://earthexplorer.usgs.gov/ (accessed 1.25.12)
.
USGS
2012
FEWS NET Africa Data Portal. http://earlywarning.usgs.gov/fews/africa/index.php (accessed 2.1.13)
.
Van de Giesen
N.
Andreini
M.
Van Edig
A.
Vlek
P.
2001
Competition for water resources of the Volta basin
. In:
Regional Management of Water Resources
.
IAHS Press
,
Wallingford
, pp.
199
206
.
Van de Giesen
N.
Stomph
T.-J.
Ajayi
A. E.
Bagayoko
F.
2011
Scale effects in Hortonian surface runoff on agricultural slopes in West Africa: Field data and models
.
Agric. Ecosyst. Environ.
142
,
95
101
.
Yong
B.
Hong
Y.
Ren
L.
Gourley
J. J.
Huffman
G. J.
Chen
X.
Wang
W.
Khan
S. I.
2012
Assessment of evolving TRMM-based multisatellite real-time precipitation estimation methods and their impacts on hydrologic prediction in a high latitude basin
.
J. Geophys. Res.
117
,
D09108
.