Abstract
This work compares future projections of rainfall over the Pra River Basin (Ghana) using data from five climate models for the period 2020–2049, as referenced to the control period 1981–2010. Bias-correction methods were applied where necessary and models' performances were evaluated with Nash–Sutcliffe Efficiency, root-mean-square error and coefficient of determination. Standardised Anomaly Index (SAI) was used to determine variability. The onset and cessation dates and length of the rainy season were determined by modifying the Walter–Olaniran method. The ensemble means of the models projected a 1.77% decrease in rainfall. The SAI showed that there would be drier than normal years with the likelihood of drought occurrence in 2021, 2023, 2031 and 2036. The findings showed that high-resolution models (≤25 km) were more capable of simulating rainfall at the basin scale than mid-resolution models (26–150 km) and projected a 20.13% increase. Therefore, the rainfall amount is expected to increase in the future. However, the projected increase in the length of the dry season by the ensemble of the models suggested that alternative sources of water would be necessary to supplement rainfed crop production for food security.
INTRODUCTION
Recently, the impact of climate change on water resources and its services are increasingly evident, especially in developing countries (IPCC 2007, 2014; Boon & Ahenkan 2012; Obuobie et al. 2012; Cubasch et al. 2013). Projections indicate that global temperature will continue to increase and will impact on rainfall patterns via its effects on evaporation and evapotranspiration from land, water and plants into the atmosphere (Marcé et al. 2010; López-Moreno et al. 2011). This will lead to natural hazards like flooding and droughts, which may not have a severe impact on populations with better economic and political stability, and improved agricultural technology (Davis et al. 2015). However, the impact will be significant in Africa due to poverty and political instability, amongst others (Burke et al. 2009; Niang et al. 2014). As a result, poverty is expected to increase, food will be scarcer and development will be less sustainable (UNEP 2002; Bo et al. 2004; FAO 2009; Serdeczny et al. 2016; Welborn 2018). A major determinant of agricultural production globally is rainfall. Climate change has affected the seasonal and annual characteristics of rainfall (Mawunya et al. 2011) leading to fluctuations in crop yields across many parts of the world (Brahic 2007; Rasul et al. 2011; Knox et al. 2013; Hatfield & Prueger 2015; UNEP 2017).
Obuobie et al. (2012) used the fourth-generation European Centre Hamburg Model (ECHAM4) and the Commonwealth Scientific and Industrial Research Organization, Australia (CSIRO) joint model and projected a decrease in rainfall by 12.3% and 17.8% by 2020 (2006–2035) and 19.6% and 25.9% by 2050 (2036–2065) in the White Volta and Pra basins, respectively. In addition, the Fourth Assessment Report (AR4) climate change scenarios of central Ghana by 2050 indicated a decrease of 10% in precipitation (WRC 2012). Nutsukpo et al. (2013) assessed precipitation change over Ghana using the A1B scenario with four downscaled general circulation models (GCMs). Two of the models, namely the National Meteorological Research Center – Climate Model 3 and ECHAM5 projected an increase of precipitation in the extreme southern part and south-eastern part of the country, respectively. The CSIRO Mark 3 projected a general decrease in precipitation, while the Model for Interdisciplinary Research on Climate (MIROC 3.2) projected increasing precipitation in the north and decrease in the south of Ghana (Nutsukpo et al. 2013). Projections of precipitation are influenced by the resolution of models, their associated uncertainties and bias of downscaling (Souvignet et al. 2010). There is a need to assess the climate impact with multiple climate data using different resolution models to get the moderate scenario for the Pra River Basin, which is dominated by tuber crop and cash crop production as well as the potential damming site in Ghana (WRC 2012; Nutsukpo et al. 2013).
Agriculture and its related economies employ more than 63% of the population in the basin with cocoa farming, the main commercial crop, which accounts for over 70% of household income (WRC 2012). Cocoa is a major export agricultural product, which contributed to 8.2% of the total Agriculture GDP of Ghana in 2010 (MoFA 2011). Peasant farmers in the basin mostly cultivate food crops like cassava, maize, plantain and especially tuber crops, which ensures national food security (Nutsukpo et al. 2013). Unfortunately, agriculture is mainly rainfed, with few irrigation technologies emerging from government and international projects (WRC 2012). Water availability to sustain the livelihood of farmers and the economy of the nation is critical due to the changing climate. Moreover, about six of the potential hydro-dam sites are located in the basin (Ahiataku-Togobo 2014). Knowledge of future rainfall in the study area is currently necessary for policy and adaptation planning.
The aim of this study was to project rainfall variability and change in the period 2020–2049 with five different climate models at different spatial resolutions under the Representative Concentration Pathway 4.5 (RCP4.5) emission scenario for the Pra River Basin of Ghana. Rainfall onset, cessation and length of the rainy season (LRS) for the observed (1981–2010) and future period (2020–2049) were also determined.
METHODS
Study area
The Pra River Basin is the largest south-western drainage basin in Ghana, covers an area of 23,321 km2 and lies between latitudes 4°58′ N and 7°11′ N and longitudes 0°25′ W and 2°13′ W, as shown in Figure 1 (Kusimi et al. 2015). It stretches to the Gulf of Guinea at Shama town in the Western Region. The annual mean temperature ranges between 26 °C and 27 °C (Bessah et al. 2018). The basin supplies water to the capitals of Ashanti, Western and Central Region for both domestic and industrial uses. About 41 districts and over 1,300 towns also benefit from the water sources in the basin (Kusimi et al. 2015). Water resources were available all year round until the increase in anthropogenic activities such as agricultural expansion, mining (legal and illegal) and logging, which are adversely degrading the quality of water resources in the basin (WRC 2012). The basin has a bi-modal rainfall pattern with the major rains occurring between May and June and the minor rains occurring around September to October (Dickson & Benneh 1995). The moist south-west monsoon is a major driver of the rainfall pattern within the basin. The dry season occurs mainly from November after the cessation of the minor rains to March (Dickson & Benneh 1995). Some of the local physical features such as basin topography (elevation ranging from −2 to 842 m), the natural meteorite lake (Lake Bosomtwe) and land cover change of the semi-deciduous forest contribute to rainfall variability in the basin. These features are linked to major driving forces such as sea surface temperature and relative humidity (Nicholson & Webster 2007; Ambrosino 2011; Goodess et al. 2011).
Dataset description
Historical daily rainfall data from seven operational climate stations (Table 1) out of the 30 stations identified in the Pra River Basin (Figure 1) with minimal missing data were acquired from the Ghana Meteorological Agency (GMet). Amelia package in R software was used to fill gaps in data for the reference period 1981–2010 to enhance the performance assessment of the models at equal conditions of station data (Arguez et al. 2012). Climate stations that had missing data were Atieku (2.5%), Konongo (0.8%), Dunkwa (1.1%), Kibi (14.2%) and Twifo Praso (4.7%). Rainfall data were subjected to quality control in RClimdex after filling the gaps to identify outliers and erroneous data such as negative rainfall values which were then removed (Aguilar et al. 2009). Table 1 presents the characteristics of the seven climate stations used in this study. The emission scenario used for analysis comprised the Representative Concentration Pathways 4.5 W/m2 mitigation scenario. The RCP4.5 represented B1 of the Fourth Assessment Report (AR4) Special Report on Emissions Scenarios (SRES) (Fenech et al. 2007; IPCC 2007; Cubasch et al. 2013). Although tropical West Africa is reported to be a hotspot of climate change under both RCP4.5 and RCP8.5 for the late 2030s to early 2040s (Diffenbaugh & Giorgi 2012; Mora et al. 2013), the goal of the United Nations Framework Convention on Climate Change through policies such as the Kyoto protocol and Paris agreement is to achieve the emission target under the RCP 4.5 scenario and even lower (Lomborg 2016; Muthee et al. 2018).
Characteristics of selected climate stations
Station ID . | Station name . | Date opened . | Latitude . | Longitude . | Altitude (m) . | Region . |
---|---|---|---|---|---|---|
18026ATI | ATIEKU | 01-09-51 | 05° 34′ N | 01° 42′ W | 106.6 | WESTERN |
17015DUN | DUNKWA ON OFFIN | 01-01-44 | 05° 58′ N | 01° 47′ W | 158.6 | CENTRAL |
23023TWI | TWIFO PRASO | 01-01-37 | 05° 36′ N | 01° 33′ W | 76 | CENTRAL |
21020KIB | KIBI | 01-01-12 | 06° 10′ N | 00° 33′ W | 274.2 | EASTERN |
21088ODA | AKIM ODA | 11-09-72 | 05° 56′ N | 00° 59′ W | 139.4 | EASTERN |
19007KON | KONONGO | 01-03-34 | 06° 37′ N | 01° 13′ W | 243.7 | ASHANTI |
17009KSI | KUMASI AIRPORT | 01-01-45 | 06° 43′ N | 01° 36′ W | 286.3 | ASHANTI |
Station ID . | Station name . | Date opened . | Latitude . | Longitude . | Altitude (m) . | Region . |
---|---|---|---|---|---|---|
18026ATI | ATIEKU | 01-09-51 | 05° 34′ N | 01° 42′ W | 106.6 | WESTERN |
17015DUN | DUNKWA ON OFFIN | 01-01-44 | 05° 58′ N | 01° 47′ W | 158.6 | CENTRAL |
23023TWI | TWIFO PRASO | 01-01-37 | 05° 36′ N | 01° 33′ W | 76 | CENTRAL |
21020KIB | KIBI | 01-01-12 | 06° 10′ N | 00° 33′ W | 274.2 | EASTERN |
21088ODA | AKIM ODA | 11-09-72 | 05° 56′ N | 00° 59′ W | 139.4 | EASTERN |
19007KON | KONONGO | 01-03-34 | 06° 37′ N | 01° 13′ W | 243.7 | ASHANTI |
17009KSI | KUMASI AIRPORT | 01-01-45 | 06° 43′ N | 01° 36′ W | 286.3 | ASHANTI |
Table 2 presents the characteristics of the GCMs and their regional climate models (RCMs) data, which were used. The model validation method of Fenech et al. (2007), which employs the region of first standard deviation (1xStdev) and second standard deviation (2xStdev), was used to choose CanESM2 and IPSL models (full name of the models described in Table 2) as best models over the Pra River Basin, out of 43 models from the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) (detailed in section ‘Description of AR5 GCMs data validation’). The RCM of the two models, downloaded from the Coordinated Regional Climate Downscaling Experiment (CORDEX) site, was the Rossby Centre regional atmospheric model (RCA4) by the Swedish Meteorological and Hydrological Institute (SMHI). From the geoportal of West African Science Service Centre on Climate Change and Adapted Land Use (WASCAL), the Weather Research and Forecasting Model (WRF) of GFDL and HadGEM2 (detailed in Table 2) were downloaded (Heinzeller et al. 2016a). High temporal (3 hrs) and spatial (12 km) resolution were some of the strengths of the WRF model, while its forecast length of 48 h and sensitivity to atmospheric conditions was classified as a weakness (Heinzeller 2019). The GFDL-ESM2M and HADGEM2-ES represent extreme wet and dry conditions, respectively, over West Africa (Heinzeller 2019). The SMHI-RCA4 CORDEX model in comparison with previous versions showed an improvement in the resolution from 50 to 44 km with a downside of overestimating precipitation based on a test result over Europe (Samuelsson et al. 2011). Statistical Downscaling Model – Decision Centric (SDSM-DC) version 5.2 with a resolution of 2 m (Wilby et al. 2002; Wilby et al. 2014) was the fifth model acquired from the SDSM website. The SDSM combines a stochastic weather generator and multiple linear regression in downscaling from the National Center for Environmental Prediction (NCEP) grids at a 2.5° × 2.5° resolution for each station. The following steps, namely screen variables selection, calibration, validation, weather generation and scenario generation, were duly followed in the SDSM-DC 5.2. The calibration and validation of SDSM were done with NCEP predictors as the daily large-scale atmospheric variables and observed records at each station as predictands. The model names GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-MR, CCCma-CanESM2 and SDSM-DC are referred to hereafter as GFDL, Hadgem, IPSL, CanESM and SDSM, respectively (Table 2). A spatial resolution of <25 km, 26–150 km and >150 km was referred to as high, mid and low resolution, respectively, in this study, especially in the presentation of results and conclusion.
Description of climate models
Acronym . | Originating group . | Country . | GCM name . | RCM . | Resolution . | Days in a year . | Reference . |
---|---|---|---|---|---|---|---|
GFDL | NOAA Geophysical Fluid Dynamics Laboratory | USA | GFDL-ESM2M | WRF | 12 km | 365 | Heinzeller et al. (2016a) |
HadGEM | Hadley Centre for Climate Prediction and Research | UK | HadGEM2-ES | WRF | 12 km | 360 | Heinzeller et al. (2016b) |
IPSL | Institut Pierre Simon Laplace | France | IPSL-CM5A-MR | SMHI-RCA4 | 44 km | 365 | Giorgi et al. (2009); Déandreis et al. (2014) |
CCCma | Canadian Centre for Climate Modeling and Analysis | Canada | CCCma-CanESM2 | SMHI-RCA4 | 44 km | 365 | Giorgi et al. (2009); Chylek et al. (2011) |
SDSM | Wilby & Dawson (2013) | UK | NCEP | SDSM-DC | 2 m | 366 | Wilby & Dawson (2013) |
Acronym . | Originating group . | Country . | GCM name . | RCM . | Resolution . | Days in a year . | Reference . |
---|---|---|---|---|---|---|---|
GFDL | NOAA Geophysical Fluid Dynamics Laboratory | USA | GFDL-ESM2M | WRF | 12 km | 365 | Heinzeller et al. (2016a) |
HadGEM | Hadley Centre for Climate Prediction and Research | UK | HadGEM2-ES | WRF | 12 km | 360 | Heinzeller et al. (2016b) |
IPSL | Institut Pierre Simon Laplace | France | IPSL-CM5A-MR | SMHI-RCA4 | 44 km | 365 | Giorgi et al. (2009); Déandreis et al. (2014) |
CCCma | Canadian Centre for Climate Modeling and Analysis | Canada | CCCma-CanESM2 | SMHI-RCA4 | 44 km | 365 | Giorgi et al. (2009); Chylek et al. (2011) |
SDSM | Wilby & Dawson (2013) | UK | NCEP | SDSM-DC | 2 m | 366 | Wilby & Dawson (2013) |
Description of AR5 GCMs data validation
Annual historical (1981–2010) and future (2006–2099) temperatures (2 m above sea level) and precipitation simulations for each station (point assessment) and also over the Pra River Basin (area assessment) for the 43 GCMs in AR5 were downloaded in the CSV format through scatterplots on the UPEI database. The UPEI database is one of the services provided by the UPEI Climate Lab to make climate baseline and future datasets available and accessible. Annual observation values for the same historical period from NCEP reanalysis datasets for each of the stations were also downloaded through the same medium. The difference between GCMs historical simulations and NCEP observed reanalysis data for temperature and precipitation were plotted as XY scatter graphs (Fenech et al. 2007). The zone of acceptability was estimated by calculating the standard deviation (1xStdev) of the XY scatter plot data and doubling it to get zone 2 (2xStdev). The zones were then drawn on the XY scatter plots to determine the type and number of models, which fell within the acceptable zone for the validation of the GCMs results for the study area. Graphs showing changes in each future period [2020s (2011–2040), 2050s (2041–2070) and 2080s (2071–2099)] were produced from the maximum, minimum, mean (ensemble) and validated (mean of models that fell within the acceptable zone) of the annual simulations of the GCMs. The two most recurring validation GCMs at the seven climate stations and in the area analysis over the basin available on CORDEX Africa-domain were CCCma-CanESM2 and IPSL-CM5A-MR (Table 2).
Data analysis
Calibration of SDSM
The analysis showed a very low partial correlation level between the station rainfall data and predictors (Table 3). The best predictors selected for rainfall calibration were direct shortwave radiation, surface lifted index, vorticity near the surface, 850 hPa and 500 hPa, surface divergence, precipitable water, total precipitation and relative humidity at 850 hPa and 500 hPa and near-surface relative humidity. Each station was calibrated with a minimum of four of the listed predictors with significance at a 95% confidence level. The partial correlation of rainfall as a conditional model and predictors was very low. It implied that rainfall was very difficult to predict in this zone. Rainfall as a conditional process has an intermediary process between the local weather and regional forcings that have a direct link with wet and/or dry day occurrence. The process further depends on parameters like atmospheric pressure and humidity (Wilby & Dawson 2004; Gulacha & Mulungu 2016). Therefore, rainfall downscaling has been found to be problematic and difficult compared with temperature (Hassan & Harun 2011). The results are similar to the findings of Gulacha & Mulungu (2016). The cross-validation was done at twofold, that is, the data were divided into two equal 15 years' intervals for calibration and the remaining 15 years for validation. The mean of proportion accuracy (prop correct) for validation of the model was in the range of 61–71% across the seven stations. Future climate scenarios for SDSM were generated using the mean validated projected rainfall change of the 2020s and 2050s from the 43 GCMs of the IPCC fifth assessment report (AR5) at each station (Table 4). The GCMs projections for each station were determined using the climate database of the University of Prince Edward Island (UPEI 2017; Bessah et al. 2018). The event threshold for rainfall modelling in SDSM was set at 0.85 (Garbutt et al. 1981) and the mean addition treatments of rainfall rate were employed in generating the future scenarios for 2020–2049.
The partial correlations between the selected predictors and the observed station rainfall of SDSM model calibration for the base period 1981–2010
Predictor . | Description . | Synoptic stations . | ||||||
---|---|---|---|---|---|---|---|---|
Atieku . | Dunkwa on Ofin . | Twifo Praso . | Akim Oda . | Kibi . | Kumasi . | Konongo . | ||
dswr | Direct shortwave radiation | −0.04 | ||||||
lftx | Surface lifted index | −0.09 | −0.08 | −0.08 | −0.10 | −0.08 | −0.09 | |
p_z | Vorticity near the surface | −0.06 | ||||||
p_zh | Surface divergence | 0.06 | ||||||
p5_z | Vorticity at 500 hPa | 0.04 | 0.09 | |||||
P8_z | Vorticity at 850 hPa | 0.04 | ||||||
pr_wtr | Precipitable water | 0.09 | 0.04 | 0.06 | 0.08 | 0.08 | 0.04 | 0.09 |
prec | Precipitation total | 0.08 | 0.06 | 0.05 | 0.04 | 0.06 | 0.06 | 0.06 |
r850 | Relative humidity at 850 hPa height | 0.07 | −0.06 | −0.04 | −0.08 | −0.06 | −0.07 | −0.08 |
r500 | Relative humidity at 500 hPa height | 0.06 | ||||||
rhum | Near-surface relative humidity | 0.05 |
Predictor . | Description . | Synoptic stations . | ||||||
---|---|---|---|---|---|---|---|---|
Atieku . | Dunkwa on Ofin . | Twifo Praso . | Akim Oda . | Kibi . | Kumasi . | Konongo . | ||
dswr | Direct shortwave radiation | −0.04 | ||||||
lftx | Surface lifted index | −0.09 | −0.08 | −0.08 | −0.10 | −0.08 | −0.09 | |
p_z | Vorticity near the surface | −0.06 | ||||||
p_zh | Surface divergence | 0.06 | ||||||
p5_z | Vorticity at 500 hPa | 0.04 | 0.09 | |||||
P8_z | Vorticity at 850 hPa | 0.04 | ||||||
pr_wtr | Precipitable water | 0.09 | 0.04 | 0.06 | 0.08 | 0.08 | 0.04 | 0.09 |
prec | Precipitation total | 0.08 | 0.06 | 0.05 | 0.04 | 0.06 | 0.06 | 0.06 |
r850 | Relative humidity at 850 hPa height | 0.07 | −0.06 | −0.04 | −0.08 | −0.06 | −0.07 | −0.08 |
r500 | Relative humidity at 500 hPa height | 0.06 | ||||||
rhum | Near-surface relative humidity | 0.05 |
AR5 Ensemble and validated precipitation rate projections of the selected stations in the Pra River Basin
Climate stations . | Ensemble (Mean) . | Validated Meana . | ||||||
---|---|---|---|---|---|---|---|---|
Baseline (mm/d) . | 2020s (%) . | 2050s (%) . | 2080s (%) . | Baseline (mm/d) . | 2020s (%) . | 2050s (%) . | 2080s (%) . | |
Akim Oda | 3.60 | 1.64 | 2.63 | 4.40 | 3.73 | 2.01 | 1.96 | 0.47 |
Kumasi | 3.92 | 0.63 | −0.04 | 0.40 | 4.07 | −0.85 | −3.32 | −4.92 |
Atieku | 3.45 | 1.00 | 1.30 | 3.16 | 4.48 | 1.38 | 0.97 | 0.86 |
Dunkwa on Offin | 3.65 | 1.21 | 1.95 | 3.60 | 4.27 | 0.32 | −0.61 | −1.57 |
Twifo Praso | 3.40 | 0.78 | 1.36 | 2.86 | 4.17 | 1.94 | 1.68 | 1.41 |
Kibi | 3.92 | 1.00 | 0.83 | 1.43 | 3.87 | −0.09 | −1.57 | −2.51 |
Konongo | 3.89 | 0.91 | 0.76 | 1.53 | 4.43 | −1.21 | −3.47 | −5.11 |
Mean | 3.69 | 1.02 | 1.26 | 2.48 | 4.15 | 0.5 | −0.62 | −1.62 |
Climate stations . | Ensemble (Mean) . | Validated Meana . | ||||||
---|---|---|---|---|---|---|---|---|
Baseline (mm/d) . | 2020s (%) . | 2050s (%) . | 2080s (%) . | Baseline (mm/d) . | 2020s (%) . | 2050s (%) . | 2080s (%) . | |
Akim Oda | 3.60 | 1.64 | 2.63 | 4.40 | 3.73 | 2.01 | 1.96 | 0.47 |
Kumasi | 3.92 | 0.63 | −0.04 | 0.40 | 4.07 | −0.85 | −3.32 | −4.92 |
Atieku | 3.45 | 1.00 | 1.30 | 3.16 | 4.48 | 1.38 | 0.97 | 0.86 |
Dunkwa on Offin | 3.65 | 1.21 | 1.95 | 3.60 | 4.27 | 0.32 | −0.61 | −1.57 |
Twifo Praso | 3.40 | 0.78 | 1.36 | 2.86 | 4.17 | 1.94 | 1.68 | 1.41 |
Kibi | 3.92 | 1.00 | 0.83 | 1.43 | 3.87 | −0.09 | −1.57 | −2.51 |
Konongo | 3.89 | 0.91 | 0.76 | 1.53 | 4.43 | −1.21 | −3.47 | −5.11 |
Mean | 3.69 | 1.02 | 1.26 | 2.48 | 4.15 | 0.5 | −0.62 | −1.62 |
aValidated mean is the average of the models that were within an acceptable range of performance for the study area.
RESULTS AND DISCUSSION
AR5 projection of rainfall over Pra River Basin
The Pra River Basin lies between latitude 4.94°N and 7.20°N and longitude 0.95°W and 2.65°W on the UPEI database, which was used for the extraction of the total rainfall data (Bessah et al. 2018). A mean rainfall rate of 3.67 mm/day was recorded for the ensemble of 43 GCMs in the fifth Assessment Report (AR5) of IPCC with a range of 1.81–7.09 mm/day for 1981–2010. The mean rainfall rate of the climate stations within the basin was 3.69 mm/day with a difference of 0.02 over the area (Table 4). Rainfall in the basin was projected to a respective increase from the base period (1981–2010) by 0.81%, 0.60% and 1.62% for the 2020s (2011–2040), 2050s (2041–2070) and 2080s (2071–2099) (Figure 2). The projected mean of the seven stations confirmed results determined for the basin area in Figure 2. However, validated results from the selected stations projected rainfall decrease of 0.62% and 1.62% for the 2050s and 2080s, respectively (Table 4). The contradictions in the results of individual stations and the basin as a whole were due to the spatial variation of rainfall within the basin.
Change in total rainfall projected by 43 GCMs over the Pra River Basin.
Temporal climate variability during the observed period (1981–2010)
The mean annual rainfall in the basin for the observed period was about 1446 (±226) mm with the lowest at Kumasi (1,314.66 ± 216.90) and highest at Atieku (1,553.41 ± 249.46) (Figure 3). The mean rainfall trend increased slightly at a slope of 2.78. There were seven years of drier than normal and wetter than normal in the basin from 1981 to 2010 with variability indicated by the SAI as high as −1.59 and +1.20 in 1983 and 2007, respectively (Figure 3).
Annual historical rainfall amount (mm) (a) and SAI (b) in the Pra River Basin. The linear regression model was generated by fitting a linear trend line automatically to the mean time-series graph of the seven climate stations in Microsoft Excel.
Annual historical rainfall amount (mm) (a) and SAI (b) in the Pra River Basin. The linear regression model was generated by fitting a linear trend line automatically to the mean time-series graph of the seven climate stations in Microsoft Excel.
Evaluating the performance of models
Evaluating the performance of the models with the three time-series metrics showed that SDSM performed the best amongst the five models in simulating historical rainfall in the basin (Table 5). Acceptable models should have NSE values ranging between 0.0 and 1.0, lower RMSE and R2 above 0.5 (Moriasi et al. 2007). Bias-corrections were applied to two stations for CanESM and to five stations for IPSL before the models fell within an acceptable range (Table 5). The WRF models (Hadgem and GFDL) at 12 km resolution needed no bias-correction to perform well. Thus, higher resolution models (SDSM and WRF) showed better agreement with the observed rainfall over the basin compared with the SMHI-RCA4 (CanESM and IPSL). Therefore, the efficiency of the models in descending order of monthly means assessment was SDSM, Hadgem, GFDL, CanESM and IPSL. The performance evaluation was done on a monthly basis since it best depicted the characteristics of change in the rainfall pattern (Gulacha & Mulungu 2016). The high variations in the performance of the models emphasised the uncertainties in climate models (Karambiri et al. 2011; Paeth et al. 2011). This could be as a result of their computational process, dynamical structure or greenhouse gas emission scenarios (Covey et al. 2003; Semenov & Stratonovitch 2010). It has also been found that it is more challenging to downscale precipitation compared with the temperature in Africa and this might contribute to the varying results of models on precipitation in Ghana (Hassan & Harun 2011; Obuobie et al. 2012; Nutsukpo et al. 2013).
Performance of the models with the observed data
Models . | Ateiku . | Dunkwa . | Twifo Praso . | Kibi . | Akim Oda . | Konongo . | Kumasi . |
---|---|---|---|---|---|---|---|
Nash–Sutcliffe efficiency (NSE) | |||||||
CanESM | 0.53 | 0.72 | 0.38 | 0.45a | 0.63a | 0.21 | 0.58 |
GFDL | 0.37 | 0.30 | 0.22 | 0.15 | 0.01 | 0.50 | 0.02 |
Hadgem | 0.53 | 0.75 | 0.51 | 0.418 | 0.71 | 0.70 | 0.50 |
IPSL | 0.25a | 0.06 | 0.35a | 0.43a | 0.63a | 0.85b | 0.02 |
SDSM | 0.90 | 0.89 | 0.82 | 0.90 | 0.83 | 0.94 | 0.84 |
Root-mean-square error (RMSE) | |||||||
CanESM | 0.97 | 0.96 | 1.21 | 1.29a | 1.01a | 1.48 | 1.10 |
GFDL | 1.28 | 1.37 | 1.28 | 1.57 | 1.42 | 1.14 | 1.45 |
Hadgem | 1.09 | 0.83 | 1.04 | 1.11 | 0.83 | 0.93 | 1.06 |
IPSL | 1.47a | 1.51 | 1.37a | 1.31a | 1.03a | 0.64b | 1.53 |
SDSM | 0.53 | 0.58 | 0.71 | 0.54 | 0.67 | 0.43 | 0.69 |
Coefficient of determination (R2) | |||||||
CanESM | 0.72 | 0.85 | 0.66 | 0.98a | 0.99a | 0.82 | 0.86 |
GFDL | 0.59 | 0.57 | 0.55 | 0.58 | 0.46 | 0.69 | 0.59 |
Hadgem | 0.62 | 0.76 | 0.62 | 0.67 | 0.75 | 0.72 | 0.76 |
IPSL | 0.78a | 0.67 | 0.96a | 0.96a | 0.99a | 0.92b | 0.75 |
SDSM | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 |
Models . | Ateiku . | Dunkwa . | Twifo Praso . | Kibi . | Akim Oda . | Konongo . | Kumasi . |
---|---|---|---|---|---|---|---|
Nash–Sutcliffe efficiency (NSE) | |||||||
CanESM | 0.53 | 0.72 | 0.38 | 0.45a | 0.63a | 0.21 | 0.58 |
GFDL | 0.37 | 0.30 | 0.22 | 0.15 | 0.01 | 0.50 | 0.02 |
Hadgem | 0.53 | 0.75 | 0.51 | 0.418 | 0.71 | 0.70 | 0.50 |
IPSL | 0.25a | 0.06 | 0.35a | 0.43a | 0.63a | 0.85b | 0.02 |
SDSM | 0.90 | 0.89 | 0.82 | 0.90 | 0.83 | 0.94 | 0.84 |
Root-mean-square error (RMSE) | |||||||
CanESM | 0.97 | 0.96 | 1.21 | 1.29a | 1.01a | 1.48 | 1.10 |
GFDL | 1.28 | 1.37 | 1.28 | 1.57 | 1.42 | 1.14 | 1.45 |
Hadgem | 1.09 | 0.83 | 1.04 | 1.11 | 0.83 | 0.93 | 1.06 |
IPSL | 1.47a | 1.51 | 1.37a | 1.31a | 1.03a | 0.64b | 1.53 |
SDSM | 0.53 | 0.58 | 0.71 | 0.54 | 0.67 | 0.43 | 0.69 |
Coefficient of determination (R2) | |||||||
CanESM | 0.72 | 0.85 | 0.66 | 0.98a | 0.99a | 0.82 | 0.86 |
GFDL | 0.59 | 0.57 | 0.55 | 0.58 | 0.46 | 0.69 | 0.59 |
Hadgem | 0.62 | 0.76 | 0.62 | 0.67 | 0.75 | 0.72 | 0.76 |
IPSL | 0.78a | 0.67 | 0.96a | 0.96a | 0.99a | 0.92b | 0.75 |
SDSM | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 |
aLinear scaling + double quantile mapping bias-corrected.
bLinear scaling bias-corrected.
Projected rainfall variability and change
Figure 4 shows the projected ensemble monthly mean rainfall compared with the observed records (1981–2010) and the SAI of the future period (2020–2049). In the future period from the ensemble of the five models, the mean daily rainfall amount decreased over the basin in January, February, March, June, October, November and December at 32.14, 32.25, 8.42, 7.67, 5.72, 3.16 and 19.62%, respectively, while the remaining months experienced increases in daily rainfall amounts (Table 6). All the models succeeded in depicting the bi-modal distribution of rainfall in the basin except for GFDL, which showed a mono-modal rainfall pattern from April to October (Figure 4). The bi-modal distribution coincided with the onset of rainfall in February–March to end in August and the minor season beginning in September to end in November–December. This was similar to the results obtained by Laprise et al. (2013), where the ERA-driven fifth-generation Canadian Regional Climate Model was successful in reproducing the minor raining season of the Guinea Coast region in September. The increase in rainfall amount in July and August might make it seem as if there was no break from the major season into the minor season in the future period.
Rainfall percentage (%) change over the Pra River Basin (2020–2049)
Month . | CanESM . | IPSL . | GFDL . | Hadgem . | SDSM . | Ensemble . |
---|---|---|---|---|---|---|
Jan | −85.16 | −48.77 | −69.13 | 34.80 | 7.56 | −32.14 |
Feb | −67.49 | −35.05 | −63.05 | −3.12 | 7.48 | −32.25 |
Mar | −10.10 | −22.14 | −29.48 | 6.53 | 13.08 | −8.42 |
Apr | 8.55 | −21.13 | 7.94 | 52.13 | 11.88 | 11.87 |
May | 12.58 | −5.91 | −3.59 | 27.35 | 11.09 | 8.30 |
Jun | −12.93 | −11.29 | −21.07 | −8.17 | 15.12 | −7.67 |
Jul | 3.87 | 10.65 | 48.20 | −18.23 | 19.72 | 12.84 |
Aug | −3.63 | 46.82 | 132.99 | 21.46 | 24.49 | 44.43 |
Sep | 3.43 | −6.64 | 33.31 | 5.26 | 15.78 | 10.23 |
Oct | −36.53 | −29.57 | −6.26 | 31.61 | 12.18 | −5.72 |
Nov | −65.91 | −34.80 | −25.35 | 100.81 | 9.46 | −3.16 |
Dec | −63.08 | −55.47 | −64.27 | 71.49 | 13.25 | −19.62 |
Mean | −26.37 | −17.78 | −4.98 | 26.83 | 13.43 | −1.77 |
Month . | CanESM . | IPSL . | GFDL . | Hadgem . | SDSM . | Ensemble . |
---|---|---|---|---|---|---|
Jan | −85.16 | −48.77 | −69.13 | 34.80 | 7.56 | −32.14 |
Feb | −67.49 | −35.05 | −63.05 | −3.12 | 7.48 | −32.25 |
Mar | −10.10 | −22.14 | −29.48 | 6.53 | 13.08 | −8.42 |
Apr | 8.55 | −21.13 | 7.94 | 52.13 | 11.88 | 11.87 |
May | 12.58 | −5.91 | −3.59 | 27.35 | 11.09 | 8.30 |
Jun | −12.93 | −11.29 | −21.07 | −8.17 | 15.12 | −7.67 |
Jul | 3.87 | 10.65 | 48.20 | −18.23 | 19.72 | 12.84 |
Aug | −3.63 | 46.82 | 132.99 | 21.46 | 24.49 | 44.43 |
Sep | 3.43 | −6.64 | 33.31 | 5.26 | 15.78 | 10.23 |
Oct | −36.53 | −29.57 | −6.26 | 31.61 | 12.18 | −5.72 |
Nov | −65.91 | −34.80 | −25.35 | 100.81 | 9.46 | −3.16 |
Dec | −63.08 | −55.47 | −64.27 | 71.49 | 13.25 | −19.62 |
Mean | −26.37 | −17.78 | −4.98 | 26.83 | 13.43 | −1.77 |
Projected annual rainfall (mm) amount (a) and SAI (b) by the assessed climate models.
Projected annual rainfall (mm) amount (a) and SAI (b) by the assessed climate models.
The SAI in Figure 4 shows that there will be drier than normal years (2021, 2023 and 2031) than wetter than normal years (2042 and 2049) during 2020–2049. However, 2020 (−0.38) and 2036 (−0.38) are closer to drier than normal years and should be considered in planning against drought with the three projected years with SAI < −0.5. Hadgem and CanESM recorded the highest negative (−2.28) and positive (2.61) SAI in the year 2020 and 2033, respectively. The CanESM and IPSL projections (Table 6) with a respective 26% and 18% decrease in rainfall are similar to the findings of Obuobie et al. (2012), where an ECHAM4/CSIRO joint model showed rainfall decreases of 17.8% by 2020 and 25.9% by 2050. This could be attributed to the similarities in their spatial resolution (44–55 km). However, SDSM (best model) and Hadgem (second best model) projected rainfall increase of 13.4% and 26.8%, respectively, with GFDL (third best model) projecting a decrease of 4.9%. The projected increase of rainfall by the two best models in this study is in agreement with the findings of Donat et al. (2014) who project that the world's dry places like the tropics will experience more extreme precipitation, which could result in rainfall increase. The dry seasons of the bi-modal rainfall patterns in the basin were during July–August and November–March (Dickson & Benneh 1995). The ensemble results show that the future rainfall pattern in the basin could favour the short dry season between July and August because rainfall is projected to increase by 12.8% and 44.4%, respectively (Table 6). The advantages of these findings are that crops requiring a constant supply of water for maturing may receive enough water from April to September (HarvestChoice 2010; Guan et al. 2015). Consequently, the harvest period of cash crops like cocoa may increase, thereby increasing farmers' yields (HarvestChoice 2010; Preethi & Revadekar 2013). Disadvantages of the projected increase in monthly rainfall (Table 6) include the rise in vector-borne diseases such as malaria and dengue fever as breeding grounds increase (Githeko et al. 2000; Thomson et al. 2018) and the possibility of floods (Colwell et al. 2008). However, the major dry season (November–March) may be elongated by the future rainfall decrease from October to March with increased dry spells between December and January at a mean rate of 28% (Figure 4, Table 6). The disadvantages will range from rise in the occurrence of bush fire (Drever et al. 2009; Xystrakis et al. 2014), increased herdsman–farmer conflicts as they compete for farmland with foliage for their flocks (Kima et al. 2016; Bessah et al. 2019) and food insecurity (Blanc 2012; Guan et al. 2015; Fishman 2016).
The spatial distribution of the projected rate of change (%) in the mean annual rainfall amount from 2020 to 2049 varied amongst the models as shown in Figure 5. The SDSM and Hadgem had similar spatial rainfall distributions of increasing amounts southward and at the east end of the basin. The SDSM and Hadgem projected a rate of change between 8–19% and 0–29%, respectively. The GFDL also projected an increasing trend in the southern part of the basin, which has forest cover and coastal savannah zones of about 3–5% and in the north (2–17%). The rate of change by GFDL from the centre to the east of the basin was uniform (Figure 5). The CanESM and IPSL models projected that rainfall would increase spatially northward in the basin, that is, increasing rainfall in the semi-deciduous zones, while the forest and coastal savannah zones have a decreasing amount of rainfall between 1% and 35%. This might result in an increased runoff in the deciduous zones leading to flooding. The ensemble captured spatial increasing rate of change in the mean annual rainfall amount from the east to the west in the range of 4–8%. The ensemble suggests a decrease in rainfall at the east end of the basin. Adaptation measures need to be assessed to prevent shocks from the projected changes if it happens.
The projected rate of change (%) in the mean annual rainfall by models (2020–2049).
The projected rate of change (%) in the mean annual rainfall by models (2020–2049).
Variations in onset, cessation and duration of rainfall
Figure 6 presents the rainfall onset, cessation dates and the LRS projected by the five models and the reference records of the observed period (1981–2010). During the observed period, early onset rainfall was on 4th February 2004 and the late-onset was on 6th April 1983. Early and late cessation dates were on 16th October 1987 and 15th December 1990, respectively, with an average LRS at 255 days. The rainy season is defined as the period where the total monthly rainfall was equal to or above 50.8 mm (Matthew et al. 2017). There was a drought in the year 1983 in Ghana, as rainfall ceased on 26th October, giving the lowest span of rainy days (218 days) during the 30-year observed period (1981–2010). The rainfall onset had a decreasing trend, that is, onset became earlier over the observed period and increasingly had a later cessation date. This implied that the LRS increased across the years over the basin in the observed period, starting with 243 days in 1981 and increasing to 289 days in 2010 (Figure 6). The method does not account for the break in the bi-modal rainfall pattern in the semi-deciduous forest zone, which occurs in July and August before the minor rains start in September (Dickson & Benneh 1995).
Observed and projected rainfall onset, rainfall cessation and length of raining season over the Pra River Basin.
Observed and projected rainfall onset, rainfall cessation and length of raining season over the Pra River Basin.
The SDSM and Hadgem models projected an average LRS of 267 and 278 days, respectively, in the period 2020–2049. There is a projection of increased rainy days by 12 and 23 days from the observed period, whereas GFDL, CanESM and IPSL projected a decrease in rainy days by 29, 40 and 33 days, respectively. Rainfall onset, as projected by SDSM, from 2020 to 2049 will be as early as 5th February in 2045 and lightly later on 13th March in 2039. Cessation is on 22nd October 2034 and 17th December 2026 for early and late cessation, respectively (Figure 6). Hadgem, on the other hand, projected an early onset on 31st January 2027 and the late-onset on 28th March 2021, while the early and late rainfall cessation was on 4th November 2028 and 20th December 2038, respectively (Figure 6). According to the best performing models from this study, it implies that rainfall onset will be delayed by one month from the observed period into the future period, while rainfall cessation will be maintained in the same months with additional rainy days before ending.
The GFDL projected early and late-onset of rainfall in February and April and rainfall cessation to be October and November with 225 days as the rainy season. The GFDL projected increasing trends for both onset and cessation of rainfall, which implies that rainfall will start late and end early in 2020–2049. There was a slight difference between the projection of the GFDL, CanESM and IPSL models. The CanESM and IPSL (SMHI-RCA4) models showed decreasing trends in rainfall onset and cessation. It implies that rainfall onset is expected to be late and cessation to be early across future periods. The average LRS for CanESM and IPSL were 215 and 222 days, respectively. Early and late-onset for SMHI-RCA4 models were in February and April, respectively, while early cessation was in September and October for CanESM and IPLS, respectively. Late rainfall cessation for SMHI-RCA4 models was projected to be in November. The GFDL and Hadgem models, which have the same spatial resolution running on the WRF model, performed differently in this study. This might be due to the varying boundary conditions under which the RCMs were set in the GCMs (Nikiema et al. 2016).
Table 7 presents the onset date, cessation date and LRS at the seven stations considered in the study. The observed period was between 1981 and 2010, whereas the projected years were between 2020 and 2049. Two synoptic stations, namely Akim Oda and Kumasi, were used as references, based on findings from other studies. At the Akim Oda station, early and late-onset of rainfall was on 23rd January and 4th April, respectively, while early and late cessation of rainfall was on 3rd October and 25th December, respectively. The mean onset and cessation dates for the observed period were on 27th February and 18th November, respectively. The average LRS was 264 (±30) days. Mensah et al. (2016) used the fuzzy logic approach in Instat software to determine the onset date, cessation date and length of raining season for Akim Oda for the period 1998–2012. The mean onset and cessation dates were on 15th March and 10th November, respectively, and the duration was 240 days (Mensah et al. 2016). Amekudzi et al. (2015) used the percentage mean cumulative rainfall method over the period 1970–2012 and to determine the onset dates, cessation dates and length of rain to be 11th March, 6th/11th November (rainfall amount/rainy days) and 245 days, respectively. Comparatively, the study's modified Walter–Olaniran method (Matthew et al. 2017) results were 7 and 19 days over that of cumulative curves method of Amekudzi et al. (2015) for cessation date and LRS, respectively, and 8 days for cessation date and 24 days for LRS above the results of Mensah et al. (2016). However, the mean onset date was earlier in this study at 12 and 16 days before the determined onset by Amekudzi et al. (2015) and Mensah et al. (2016), respectively.
Onset, cessation and LRS of the observed (1981–2010) and model projections (2020–2049) at the selected stations in the Pra River Basin
Data . | Onset of rainfall . | Cessation of rainfall . | LRS (days) . | ||||||
---|---|---|---|---|---|---|---|---|---|
Early . | Late . | Mean . | SD (days) . | Early . | Late . | Mean . | SD (days) . | ||
Akim Oda Station | |||||||||
Observed | 23 Jan | 04 Apr | 27 Feb | 20 | 03 Oct | 25 Dec | 18 Nov | 23 | 264 ( ± 30) |
SDSM | 21 Jan | 16 Mar | 20 Feb | 14 | 03 Oct | 31 Dec | 24 Nov | 26 | 277 ( ± 30) |
Hadgem | 22 Jan | 01 Apr | 23 Feb | 18 | 01 Nov | 31 Dec | 30 Nov | 23 | 280 ( ± 29) |
GFDL | 11 Feb | 01 May | 16 Mar | 20 | 01 Oct | 20 Dec | 27 Oct | 23 | 225 ( ± 30) |
CanESM | 3 Feb | 10 May | 19 Mar | 24 | 01 Sept | 31 Dec | 26 Oct | 33 | 221 ( ± 46) |
IPSL | 1 Feb | 03 Jun | 31 Mar | 35 | 01 Oct | 29 Dec | 10 Nov | 27 | 224 ( ± 48) |
Ateiku Station | |||||||||
Observed | 11 Jan | 14 Apr | 21 Feb | 20 | 15 Sept | 29 Dec | 16 Nov | 28 | 268 ( ± 34) |
SDSM | 05 Feb | 01 Apr | 25 Feb | 13 | 01 Oct | 29 Dec | 18 Nov | 23 | 266 ( ± 26) |
Hadgem | 13 Jan | 04 Apr | 09 Feb | 22 | 01 Nov | 30 Dec | 09 Dec | 17 | 303 ( ± 29) |
GFDL | 19 Jan | 10 Apr | 08 Mar | 22 | 01 Oct | 30 Dec | 10 Nov | 28 | 248 ( ± 38) |
CanESM | 02 Feb | 05 Apr | 10 Mar | 12 | 04 Sept | 22 Nov | 12 Oct | 19 | 215 ( ± 22) |
IPSL | 18 Jan | 17 Apr | 06 Mar | 21 | 01 Sept | 23 Dec | 06 Nov | 28 | 246 ( ± 35) |
Dunkwa on Ofin Station | |||||||||
Observed | 21 Jan | 05 Apr | 26 Feb | 16 | 01 Oct | 24 Dec | 09 Nov | 28 | 256 ( ± 32) |
SDSM | 23 Jan | 05 Apr | 26 Feb | 17 | 01 Oct | 28 Dec | 22 Nov | 26 | 268 ( ± 31) |
Hadgem | 22 Jan | 02 Apr | 22 Feb | 18 | 02 Nov | 31 Dec | 04 Dec | 23 | 285 ( ± 26) |
GFDL | 19 Jan | 10 Apr | 15 Mar | 21 | 01 Oct | 28 Dec | 01 Nov | 27 | 231 ( ± 31) |
CanESM | 26 Feb | 04 Apr | 13 Mar | 10 | 03 Sept | 22 Nov | 14 Oct | 16 | 215 ( ± 21) |
IPSL | 17 Feb | 10 Apr | 18 Mar | 13 | 01 Oct | 30 Nov | 21 Oct | 22 | 217 ( ± 29) |
Kibi Station | |||||||||
Observed | 11 Jan | 20 Apr | 21 Feb | 20 | 01 Oct | 30 Dec | 17 Nov | 29 | 269 ( ± 35) |
SDSM | 15 Jan | 05 Mar | 03 Feb | 13 | 01 Oct | 27 Dec | 29 Nov | 30 | 299 ( ± 35) |
Hadgem | 02 Feb | 03 Apr | 27 Feb | 20 | 01 Nov | 23 Dec | 28 Nov | 21 | 274 ( ± 31) |
GFDL | 14 Feb | 23 Apr | 20 Mar | 18 | 24 Aug | 30 Dec | 26 Oct | 30 | 220 ( ± 27) |
CanESM | 15 Jan | 03 May | 09 Mar | 18 | 15 Sept | 10 Dec | 15 Oct | 24 | 220 ( ± 31) |
IPSL | 05 Jan | 01 Jun | 16 Mar | 31 | 03 Sept | 29 Dec | 29 Oct | 28 | 226 ( ± 40) |
Konongo Station | |||||||||
Observed | 28 Jan | 02 May | 02 Mar | 21 | 04 Sept | 23 Dec | 22 Oct | 25 | 235 ( ± 32) |
SDSM | 02 Feb | 06 Apr | 22 Feb | 21 | 01 Oct | 28 Dec | 25 Oct | 29 | 241 ( ± 39) |
Hadgem | 07 Feb | 07 Apr | 12 Mar | 17 | 01 Oct | 31 Dec | 12 Nov | 23 | 246 ( ± 25) |
GFDL | 05 Mar | 24 Apr | 25 Mar | 14 | 01 Oct | 29 Nov | 19 Oct | 20 | 208 ( ± 25) |
CanESM | 27 Feb | 03 Apr | 14 Mar | 9 | 04 Sept | 15 Nov | 11 Oct | 13 | 211 ( ± 15) |
IPSL | 12 Jan | 13 Apr | 09 Mar | 23 | 04 Oct | 31 Dec | 28 Oct | 26 | 233 ( ± 36) |
Kumasi Station | |||||||||
Observed | 14 Jan | 02 Apr | 27 Feb | 17 | 01 Oct | 21 Dec | 22 Oct | 26 | 237 ( ± 32) |
SDSM | 01 Feb | 10 Mar | 27 Feb | 15 | 01 Oct | 24 Dec | 21 Oct | 24 | 236 ( ± 29) |
Hadgem | 04 Feb | 05 Apr | 09 Mar | 17 | 01 Oct | 31 Dec | 21 Nov | 24 | 257 ( ± 31) |
GFDL | 26 Jan | 20 Apr | 02 May | 19 | 01 Oct | 23 Dec | 20 Oct | 22 | 211 ( ± 32) |
CanESM | 25 Feb | 04 Apr | 13 Mar | 09 | 02 Sept | 15 Nov | 11 Oct | 14 | 212 ( ± 15) |
IPSL | 22 Feb | 09 Apr | 19 Mar | 12 | 01 Oct | 21 Nov | 12 Oct | 14 | 207 ( ± 21) |
Twifo Praso Station | |||||||||
Observed | 17 Jan | 08 Apr | 28 Feb | 25 | 01 Oct | 29 Dec | 12 Nov | 25 | 257 ( ± 35) |
SDSM | 23 Jan | 01 Apr | 18 Feb | 17 | 01 Nov | 30 Dec | 28 Nov | 25 | 283 ( ± 28) |
Hadgem | 14 Jan | 04 Apr | 12 Feb | 21 | 01 Nov | 26 Dec | 09 Dec | 17 | 300 ( ± 27) |
GFDL | 08 Feb | 05 Apr | 08 Mar | 20 | 01 Oct | 22 Dec | 03 Nov | 25 | 240 ( ± 37) |
CanESM | 21 Feb | 05 Apr | 12 Mar | 10 | 04 Sept | 26 Nov | 12 Oct | 19 | 213 ( ± 22) |
IPSL | 11 Mar | 14 May | 01 Apr | 15 | 01 Sept | 28 Nov | 21 Oct | 25 | 203 ( ± 27) |
Data . | Onset of rainfall . | Cessation of rainfall . | LRS (days) . | ||||||
---|---|---|---|---|---|---|---|---|---|
Early . | Late . | Mean . | SD (days) . | Early . | Late . | Mean . | SD (days) . | ||
Akim Oda Station | |||||||||
Observed | 23 Jan | 04 Apr | 27 Feb | 20 | 03 Oct | 25 Dec | 18 Nov | 23 | 264 ( ± 30) |
SDSM | 21 Jan | 16 Mar | 20 Feb | 14 | 03 Oct | 31 Dec | 24 Nov | 26 | 277 ( ± 30) |
Hadgem | 22 Jan | 01 Apr | 23 Feb | 18 | 01 Nov | 31 Dec | 30 Nov | 23 | 280 ( ± 29) |
GFDL | 11 Feb | 01 May | 16 Mar | 20 | 01 Oct | 20 Dec | 27 Oct | 23 | 225 ( ± 30) |
CanESM | 3 Feb | 10 May | 19 Mar | 24 | 01 Sept | 31 Dec | 26 Oct | 33 | 221 ( ± 46) |
IPSL | 1 Feb | 03 Jun | 31 Mar | 35 | 01 Oct | 29 Dec | 10 Nov | 27 | 224 ( ± 48) |
Ateiku Station | |||||||||
Observed | 11 Jan | 14 Apr | 21 Feb | 20 | 15 Sept | 29 Dec | 16 Nov | 28 | 268 ( ± 34) |
SDSM | 05 Feb | 01 Apr | 25 Feb | 13 | 01 Oct | 29 Dec | 18 Nov | 23 | 266 ( ± 26) |
Hadgem | 13 Jan | 04 Apr | 09 Feb | 22 | 01 Nov | 30 Dec | 09 Dec | 17 | 303 ( ± 29) |
GFDL | 19 Jan | 10 Apr | 08 Mar | 22 | 01 Oct | 30 Dec | 10 Nov | 28 | 248 ( ± 38) |
CanESM | 02 Feb | 05 Apr | 10 Mar | 12 | 04 Sept | 22 Nov | 12 Oct | 19 | 215 ( ± 22) |
IPSL | 18 Jan | 17 Apr | 06 Mar | 21 | 01 Sept | 23 Dec | 06 Nov | 28 | 246 ( ± 35) |
Dunkwa on Ofin Station | |||||||||
Observed | 21 Jan | 05 Apr | 26 Feb | 16 | 01 Oct | 24 Dec | 09 Nov | 28 | 256 ( ± 32) |
SDSM | 23 Jan | 05 Apr | 26 Feb | 17 | 01 Oct | 28 Dec | 22 Nov | 26 | 268 ( ± 31) |
Hadgem | 22 Jan | 02 Apr | 22 Feb | 18 | 02 Nov | 31 Dec | 04 Dec | 23 | 285 ( ± 26) |
GFDL | 19 Jan | 10 Apr | 15 Mar | 21 | 01 Oct | 28 Dec | 01 Nov | 27 | 231 ( ± 31) |
CanESM | 26 Feb | 04 Apr | 13 Mar | 10 | 03 Sept | 22 Nov | 14 Oct | 16 | 215 ( ± 21) |
IPSL | 17 Feb | 10 Apr | 18 Mar | 13 | 01 Oct | 30 Nov | 21 Oct | 22 | 217 ( ± 29) |
Kibi Station | |||||||||
Observed | 11 Jan | 20 Apr | 21 Feb | 20 | 01 Oct | 30 Dec | 17 Nov | 29 | 269 ( ± 35) |
SDSM | 15 Jan | 05 Mar | 03 Feb | 13 | 01 Oct | 27 Dec | 29 Nov | 30 | 299 ( ± 35) |
Hadgem | 02 Feb | 03 Apr | 27 Feb | 20 | 01 Nov | 23 Dec | 28 Nov | 21 | 274 ( ± 31) |
GFDL | 14 Feb | 23 Apr | 20 Mar | 18 | 24 Aug | 30 Dec | 26 Oct | 30 | 220 ( ± 27) |
CanESM | 15 Jan | 03 May | 09 Mar | 18 | 15 Sept | 10 Dec | 15 Oct | 24 | 220 ( ± 31) |
IPSL | 05 Jan | 01 Jun | 16 Mar | 31 | 03 Sept | 29 Dec | 29 Oct | 28 | 226 ( ± 40) |
Konongo Station | |||||||||
Observed | 28 Jan | 02 May | 02 Mar | 21 | 04 Sept | 23 Dec | 22 Oct | 25 | 235 ( ± 32) |
SDSM | 02 Feb | 06 Apr | 22 Feb | 21 | 01 Oct | 28 Dec | 25 Oct | 29 | 241 ( ± 39) |
Hadgem | 07 Feb | 07 Apr | 12 Mar | 17 | 01 Oct | 31 Dec | 12 Nov | 23 | 246 ( ± 25) |
GFDL | 05 Mar | 24 Apr | 25 Mar | 14 | 01 Oct | 29 Nov | 19 Oct | 20 | 208 ( ± 25) |
CanESM | 27 Feb | 03 Apr | 14 Mar | 9 | 04 Sept | 15 Nov | 11 Oct | 13 | 211 ( ± 15) |
IPSL | 12 Jan | 13 Apr | 09 Mar | 23 | 04 Oct | 31 Dec | 28 Oct | 26 | 233 ( ± 36) |
Kumasi Station | |||||||||
Observed | 14 Jan | 02 Apr | 27 Feb | 17 | 01 Oct | 21 Dec | 22 Oct | 26 | 237 ( ± 32) |
SDSM | 01 Feb | 10 Mar | 27 Feb | 15 | 01 Oct | 24 Dec | 21 Oct | 24 | 236 ( ± 29) |
Hadgem | 04 Feb | 05 Apr | 09 Mar | 17 | 01 Oct | 31 Dec | 21 Nov | 24 | 257 ( ± 31) |
GFDL | 26 Jan | 20 Apr | 02 May | 19 | 01 Oct | 23 Dec | 20 Oct | 22 | 211 ( ± 32) |
CanESM | 25 Feb | 04 Apr | 13 Mar | 09 | 02 Sept | 15 Nov | 11 Oct | 14 | 212 ( ± 15) |
IPSL | 22 Feb | 09 Apr | 19 Mar | 12 | 01 Oct | 21 Nov | 12 Oct | 14 | 207 ( ± 21) |
Twifo Praso Station | |||||||||
Observed | 17 Jan | 08 Apr | 28 Feb | 25 | 01 Oct | 29 Dec | 12 Nov | 25 | 257 ( ± 35) |
SDSM | 23 Jan | 01 Apr | 18 Feb | 17 | 01 Nov | 30 Dec | 28 Nov | 25 | 283 ( ± 28) |
Hadgem | 14 Jan | 04 Apr | 12 Feb | 21 | 01 Nov | 26 Dec | 09 Dec | 17 | 300 ( ± 27) |
GFDL | 08 Feb | 05 Apr | 08 Mar | 20 | 01 Oct | 22 Dec | 03 Nov | 25 | 240 ( ± 37) |
CanESM | 21 Feb | 05 Apr | 12 Mar | 10 | 04 Sept | 26 Nov | 12 Oct | 19 | 213 ( ± 22) |
IPSL | 11 Mar | 14 May | 01 Apr | 15 | 01 Sept | 28 Nov | 21 Oct | 25 | 203 ( ± 27) |
LRS – length of raining season.
At the Kumasi station, early and late-onset of rainfall was on 14th January and 2nd April, respectively, while early and late cessation of rainfall was on 1st October and 21st December, respectively. The mean onset and cessation dates for the observed period were on 27th February and 22nd October, respectively. The average LRS was 237 (±32) days (Table 7). Amekudzi et al. (2015) determined the onset, cessation and LRS for the Kumasi station to be 11th March, 22nd/27th October (rainfall amount/rainy days) and 300 days, respectively. Cessation dates of this study were the same when using the method of rainfall amount to determine the cessation date of Amekudzi et al. (2015), although the period of analysis was 30 and 43 years, respectively. Onset was earlier in this study when compared with that of Amekudzi et al. (2015). Mensah et al. (2016) also determined 21st March, 20th November and 244 days as the onset, cessation and LRS for the Kumasi station between 1998 and 2012. The difference in onset and cessation dates of the two methods compared with the method of this study was less than 15 days. A study with the same period is hereby recommended to test the modified method of this study with other available methods to determine its efficiency. The mean observed onset at all stations in this study was in February, except the Konongo station, which was in March (Table 7). November was the month for the mean observed cessation date at all stations, except at Kumasi and Konongo, which was in October.
On average at all stations, SDSM and Hadgem projected the mean onset date earlier than what was determined for the observed period or the same month, while GFDL, CanESM and IPSL projected the late-onset of rainfall in relation to the observed period. However, at Kibi, Konongo and Kumasi, the future onset date of Hadgem was about 5 days later than the observed period. The trend was the same with projected cessation dates by all models. On average SDSM and Hadgem projected the late cessation dates compared with the observed period, whereas GFDL, CanESM and IPSL projected an early cessation.
The projected LRS increased with SDSM and Hadgem, whereas GFDL, CanESM and IPSL projected a decrease in the length of the rainy days. The projected increased length of the rainy days will facilitate crop growth since growth depends more on the number of rainy days in the season than the amount (Lebel & Le Barbe 1997; Vischel & Lebel 2007). In addition, the hydrological cycle will be positively affected by the projection of SDSM and Hadgem (Modarres 2010). However, the decreased LRS projected by GFDL, CanESM and IPSL might result in prolonged drought and stunted crop growth.
CONCLUSIONS
The study established that the performance of the models based on the agreement of their historical simulations with station records using time-series metrics analyses in descending order was SDSM, Hadgem, GFDL, CanESM and IPSL. Based on the results of the models with the best ability (SDSM and Hadgem), the rainfall amount is projected to increase in the basin. This is contrary to findings of early research in the basin using low to medium resolution climate models (CanESM and IPSL). However, the ensemble projection of below 2% decrease in the rainfall amount fits into the findings of AR4 climate change scenarios over the West African sub-region and the IPCC Fifth Assessment report 43 GCMs projections over the Pra River Basin. The ensemble monthly projections indicate that the minor dry season (July–August) could have enough rain to change the rainfall pattern to follow a mono-modal type from April to September and a long dry spell from October to March. There could be a severe water scarcity according to the projection of the ensemble. The SAI results of the likelihood of drier than normal years further confirm the findings on a monthly variation in rainfall, which might result in drought. The modified approach in determining the rainfall onset, cessation and duration was closely related to other existing methods and will, therefore, need further modification to capture the bi-modal trend in rainfall in the basin. According to SDSM and Hadgem models, the rainfall onset date and cessation date are projected to be earlier and later, which consequently increases the rainy days without capturing the short dry season before the minor rainy season starts. The increase in the LRS might lead to more flooding or short dry spells within a season, depending on the distribution and intensity of rainfall. The study has shown that high-resolution models are better equipped to simulate rainfall at the basin or point scale. Thus, high-resolution models should be combined with other mid-resolution models in projecting rainfall to determine the climate change impact, vulnerability, adaptation planning and policy formulations. Alternative sources of water should be sought to supplement rainfed crop production considering the dry spell projection of the ensemble. This will have an impact on food security since the basin contributes highly to the production of cash crops like cocoa and tuber crops with valuable economic gains to the nation.
ACKNOWLEDGEMENTS
This study formed part of a PhD research under the Pan African University Institute of Life and Earth Sciences at the University of Ibadan (Nigeria), funded by the African Union Commission. The authors acknowledge the Ghana Meteorological Agency for the rainfall data. The authors thank the University of Prince Edward Island (UPEI) for granting access to their climate database for 43 GCMs analysis. The authors are also grateful to CORDEX Africa-domain for accessing SMHI-RCA4 models at climate4impact.eu and West Africa Science Service Center on Climate Change and Adapted Land Use (WASCAL) Geoportal, for WRF model downloads. Finally, the authors are grateful to co-public.lboro.ac.uk for giving free access to the SDSM model and NCEP data.