Variabilities in Rainfall Onset, Cessation and Length of Rainy Season for the Various Agro-Ecological Zones of Ghana

This paper examines the onset and cessation dates of the rainy season over Ghana using rain gauge data from the Ghana Meteorological Agency (GMet) over the period of 1970–2012. The onset and cessation dates were determined from cumulative curves using the number of rainy days and rainfall amount. In addition, the inter-annual variability of the onset and cessation dates for each climatic zone was assessed using wavelet analysis. A clear distinction between the rainfall characteristics and the length of the rainy season in the various climatic zones is discussed. The forest and coastal zones in the south had their rainfall onset from the second and third dekads of March. The onset dates of the transition zone were from the second dekad of March to the third dekad of April. Late onset, which starts from the second dekad of April to the first dekad of May, was associated with the savannah zone. The rainfall cessation dates in the forest zone were in the third dekad of October to the first dekad of November, and the length of the rainy season was within 225–240 days. The cessation dates of the coastal zone were within the second and third dekad of October, and the length of rainy season was within 210–220 days. Furthermore, the transition zone had cessation dates in the second to third dekad of October, and the length Climate 2015, 3 417 of the rainy season was within 170–225 days. Lastly, the savannah zone had cessation dates within the third dekad of September to the first dekad of October, and the length of rainy season was within 140–180 days. The bias in the rainfall onset, cessation and length of the rainy season was less than 10 days across the entire country, and the root mean square error (RMSE) was in the range of 5–25 days. These findings demonstrate that the onset derived from the cumulative rainfall amount and the rainy days are in consistent agreement. The wavelet power spectrum and its significant peaks showed evidence of variability in the rainfall onset and cessation dates across the country. The coastal and forest zones showed 2–8and 2–4-year band variability in the onsets and cessations, whereas the onset and cessation variability of the transition and savannah zones were within 2–4 and 4–8 years. This result has adverse effects on rain-fed agricultural practices, disease control, water resource management, socio-economic activities and food security in Ghana.


Introduction
The schedule of agricultural activities, right from land preparation, through crop selection and planting, to the time of harvesting for a developing country like Ghana, is rainfall dependent [1][2][3][4]. The assessment and prediction of the onset and cessation dates of the rainy season is therefore crucial to the success of agricultural activities in Ghana [5,6]. The variability in the rainy season onset and cessation could pose socio-economic and developmental challenges as they threaten food security and induce poverty [2,4]. This is so because erratic and significant delays in rainfall affect the country's overall production of food and, in particular, cereals (maize, millet, soya bean and rice), which form the main staple food in the country. In addition, rainy season onset and cessation dates affect the transmission of vector-borne diseases, as the life cycle of the disease transmission vectors is sensitive to the variability and changes in temperature and rainfall. For example, the mosquito population is likely to increase rapidly during the warmer humid conditions [7].
The methods for determining rainy season onset and cessation dates fall under two main categories, namely: those taking into account atmospheric dynamics and those based on water balance, such as rainfall distribution or number of rainy days, among others [6]. Several studies (e.g., [8][9][10][11]) have employed models that take into account atmospheric dynamics to identify rainy season onset and cessation dates. Researchers, like [12,13], employed a model that used the relationship between rainfall and evapotranspiration to determine the onset and cessation of the rainy season. Others [1,6,[14][15][16][17][18][19] determined the onset and cessation of the rainy season by using rainfall data alone; this method derives onset and cessation dates by calculating the percentage of cumulative mean rainfall.
Studies conducted in semi-arid parts of West Africa indicated that there is a significant relationship between the start of rains and the length of the rainy season [20]; thus, earlier onset most often leads to longer rainy season. This is an indication that the length of the rainy season is more dependent on the rainfall onset than its cessation [21,22]. Odekunle [19,23] showed that, even at a single synoptic station, the annual variation in the onset dates could be up to 70 days (10 weeks). Therefore, defining the onset date of the rainy season can sometimes be very challenging, as the onset characteristics can vary drastically with isolated showers or heavy rainfall of varying intensity being accompanied by extended dry spells [24].
The problems associated with poor agricultural practices, food security and human health in a developing country like Ghana could be attributed to the high variability in the onset and cessation of the rainy season. The goal of this paper therefore is to determine the dates for rainfall onset and cessation, the length of the rainy season and their variability. The paper is structured as follows: a briefing of the study area and data source is given in Section 2. In Section 3, the methodology employed is given. The results and discussions are presented in Section 4, and finally, conclusions are given in Section 5.

The Study Area
The climate of Ghana is characterized by dry and wet seasons, a typical tropical monsoonal climate. Rainfall in this region is mainly associated with mesoscale convective systems and controlled by the advection of moisture from the Gulf of Guinea in the low level atmosphere [25]. This system is usually referred to as the West African Monsoon (WAM), and it is driven by the energy and temperature gradient between the Gulf of Guinea and the Sahara. The maritime tropical air mass, which originates from the Atlantic Ocean, is moisture laden and converges with the dry northeast continental tropical air mass, usually along the Inter-Tropical Discontinuity (ITD) [26,27]. Therefore, the spatial pattern of annual rainfall is closely related to the north-and south-ward migration of the ITD, resulting in changes in the rainfall regime from the south to north of the country [28][29][30]. This gives rise to two rainfall regimes: bi-modal in the south, consisting mainly of coastal and the forest zones, and uni-modal in the northern part of the country, consisting of part of the transition and savannah zones [30].

Data Source
In this study, rainfall data from 23 Ghana Meteorological Agency (GMet) stations across the four agro-ecological zones of Ghana covering the period of 1970-2012 were used. The locations of these stations across the country are shown in Figure 1.
GMet uses the World Meteorological Organisation (WMO) standards in its rainfall and other weather parameter measurements. However, these measurements may have some minor errors. These error sources include, but are not limited to, evaporation and wetting loss and wind-induced errors. This could lead to problems of data inhomogeneity [31]. Errors due to evaporation and wetting loss were less than 0.45 mm per day. All the rain gauges were wind shielded; hence, wind-induced errors were infinitesimally small. One of the biggest drawbacks in the long time series record of the meteorological dataset is the availability of data gaps. These data gaps had to be appropriately filled and quality-controlled to provide reliable, continuous and homogeneous reference time series in which divergences are only caused by weather and climate variability [32]. The missing rainfall amount in the data time series was estimated using the inverse distance weighting (IDW) method described in [33]. The rainfall data from the neighboring stations within a distance of less than 50 km were used to estimate the missing data gaps in the reference series using the relation:R and w i is given by: whereR p is the missing data in the reference series, R i is the rainfall value of the neighboring stations, N the total number of the neighboring stations, d i the distance of each station from the reference station, w i the weighting of each neighboring station and α the optimal power, which takes values between 0-1.5 and 4-5 [33]. For this study, a power of 5 was employed for the missing rainfall value estimation.

Methodology
The rainfall onset and cessation dates, as well as the length of the rainy season were determined by adopting the percentage mean cumulative rainfall and/or the rainy days method(s) described in [23,34,35] (also see the references therein). The method has the advantage of directly using the rainfall data or rainy days, rather than inferring the rainfall amount from other related parameters, making it appropriate for this study. Odekunle [19,23] and Olaniran [15] indicate that at any time and location where the method appeared to be less accurate in the determination of the onset and cessation dates of rainfall, the dry spells involved were very short (mostly less than 5 days), and the frequency of those dry spells was very low. According to [23] (also see the references therein), points of maximum curvature, corresponding to the onset and cessation of rainfall, are expected to be 7%-8% and over 90% of the annual rainfall, respectively.
The following steps were taken to determine the onset and cessation dates: Firstly, the daily mean rainfall amount or rainy days from the entire data record (1970-2012) for each meteorological station are calculated. This was followed by the computation of the mean annual rainfall or rainy days that occurred at each 5-day interval throughout the year. Next, the percentage of that mean annual rainfall and rainy days was computed, as well as their cumulative percentages. The temporal variation of the cumulative percentages for the entire year was then determined. In this method, the mean onset date of rainfall was defined as the first point of maximum positive curvature of the graph; the cessation date of rainfall was defined as the last point of maximum negative curvature and the duration of the rainy season was defined as the difference of the onset and cessation dates of rainfall (length = cessation − onset + 1). Figures 2 provides an example of the cumulative curve for two stations in different agro-ecological zones. The top panel of Figure 2 represents the forest zone, and the bottom panel represents the savannah zone.
The inter-annual variability in the rainfall onset and cessation were determined from the cumulative percentage rainfall for each year. The mode of variability in the onset and cessation was assessed using wavelet analysis described in [36]. The wavelet analysis provides a means to assess the most significant frequencies within the time series. Usually, this is done by making reference of the global power wavelet plot to locate a significant peak that crosses the 95% confidence level. Secondly, the corresponding frequency on the vertical axis of the wavelet power plot is identified as the significant band, where the most significant variability in the onset and cessation are located. A variance of the significant band of the power spectrum is indicated by scale-averaged plots. The dotted line represents the 5% significance level, which allows for identification of periods of late and early onsets and cessations. For this study, the parameters of the wavelet analysis are set as δt = 1 year, which implies s 0 = 2 years, since s = 2δt, δ j = 0.25, in order to have 4 sub-octaves per octave. j 1 is calculated as j 1 = 5/δ j for 5 powers-of-two with δ j sub-octaves each.

Tamale
Rainfall amount Rainy days Figure 2. Example of the cumulative curve used to determine rainfall onset, cessation and length of the rainy season; bi-modal rainfall regime (top) and uni-modal rainfall regime (bottom). Rainy days method (green) and the rainfall amount method (blue). Figure 3 shows the map of the rainfall onset, the cessation and the length of the rainy season using the two different methods, namely: the cumulative rainfall amount and cumulative rainy days. Figure 3a,b indicates the rainfall onset using the rainfall amount and the rainy days, respectively, whereas Figure 3c,d and Figure 3e,f show the cessation and the length of the rainy season derived from the cumulative rainfall amount and cumulative rainy days, respectively.

Onset, Cessation and Length of Rainy Season Climatology
These results show a clear distinction in the onset from the coastal zone through the forest and transition to the savannah zone, and there are slight differences in some stations for the two methods.
Stations in coastal and forest zones had earlier rainfall onset dates, which are within the second and third dekads of March; whereas the onset dates of the transition zone were variable, starting from second dekad of March in Sunyani through to second dekad of April in Kete Krachi. Late onset were observed in the savannah zone of the country; thus, the onset of this zone was found between the second dekad of April and the first dekad of May. The summary of the onset and cessation days/dates is given in Tables 1  and 2   Similarly, cessation dates observed were also different for the various agro-ecological zones. Rainfall in the savannah zone, which is uni-modal, had cessation dates much earlier. In particular, the cessation date in Navrongo was found within the third dekad of September. It was observed that the length of the rainy season in the savannah zone (Bole, Yendi, Wa and Tamale) was in the range 165-190 days. A shorter rainy season was observed in the furthest northern station of Navrongo, with a length of about 140-150 days. The cessation dates of the transition zone were within second to third dekads of October, whereas the length of the rainy season in the transition zone (Kintampo, Wenchi and Sunyani) was in the range of 200-225 days. A shorter rainy season of 170 days was observed in Kete-Krachi (see Table 2).
For stations within the forest zone, the cessation date of the rainy season were consistently observed in the third dekad of October to the first dekad of November. The length of the rainy season in the forest zone was within 225-240 days. In the coastal zone, the cessation date was within the second and third dekad of October, while the length of the rainy season was slightly shorter than the forest zone (210-220 days). The bias in the rainfall onset was consistently less than 10 days across the entire country. Similarly, the bias in the cessation and the length of the rainy season is less than 10 days in most of the stations, with the exception of Axim and Abetifi (see Figure 4). The root mean square error (RMSE) was within 5-25 days (see Figure 5). This result demonstrates that the onset derived from the cumulative rainfall amount and the rainy days are in consistent agreement.

Variability of Onset and Cessation Dates
To assess the mode of interannual variability in the rainfall onset and cessation, wavelet analysis was carried out, and the results are shown in Figures 6-13. In general, there was variability in the onset and cessation dates in all of the agro-ecological zones. Figures 6-13a show the standardized anomaly of the onset and cessation for the four agro-ecological zones. Figures 6-13b show the wavelet power spectrum of the onset and cessation dates. The global wavelet power spectrum is shown in Figures 6-13c, where the dashed line is the 5% significance level for the global wavelet spectrum. The savannah and transition zones showed onset variability with the wavelet power spectrum between the 2-4-and 4-8-year scales. This is also shown in the two significant peaks of the global wavelet power spectrum (see Figure 6c and 8c), which is an indication of a strong variability in onset between 2-4 and 4-8 years. Figure 6d    The wavelet analyses for the forest and coastal zones are shown in Figures 10 and 12a. From the wavelet power spectrum in Figures 10b and 12b, it can be seen that there is a concentration of power between the 2-4-and 4-8-year bands. These features are also highlighted by the significant peaks shown in Figures 10c and 12c, the global wavelet power spectrum. This further demonstrates that the rainfall onset in the forest zone has variability between 2-4 years and 2-8 years. Figures 10d and 12d Figure 6, but for cessation; (c) as in Figure 6; (d) the same as in Figure 6; the dashed line shows the 95% confidence level.   Figure 6; (c) as in Figure 6; (d) as in Figure 6; the dashed line shows the 95% confidence level.   Figure 6, but for cessation; (c) as in Figure 6; (d) as in Figure 6; the dashed line shows the 95% confidence level.  Figure 6; (c) the same as in Figure 6; (d) as in Figure 6; the dashed line shows the 95% confidence level.  Figure 11. Interannual variability of rainfall cessation dates for the forest zone using wavelet analysis. (a) Standardized anomaly of cessation; (b) the same as in Figure 6, but for cessation; (c) the same as in Figure 6; (d) as in Figure 6; the dashed line shows the 95% confidence level.  Figure 6; (c) the same as in Figure 6; (d) the same as in Figure 6; the dashed line shows the 95% confidence level.  Figure 13. Interannual variability of rainfall cessation dates for the coastal zone using wavelet analysis. (a) Standardized anomaly of cessation; (b) the same as in Figure 6, but for cessation; (c) the same as in Figure 6; (d) as in Figure 6; the dashed line shows the 95% confidence level.
The wavelet analyses for the cessation in the forest and coastal zones are shown in Figures 11 and 13. The wavelet power spectrum is shown in Figures 11b and 13b, and the significant peaks of the global wavelet power spectrum are shown in Figures 11c and 13c. It can be seen that there is a strong concentration of power between the 4-8 year band followed by the 2-4 year band. These results show that the rainfall cessation at the forest and coastal zones has strong variation between 2-4 and

Conclusions
The rainfall onset and cessation dates, as well as the length of the rainy season over Ghana have been assessed using rain gauge data from GMet covering the period of 1970-2012. Distinct characteristics of the onset, cessation and the length of the rainy season in the various climatic zones in the country were observed. In general, the length of the rainy season decreases from the north of the country to the south and also from the east coast to the west. The longest rainy season of about 240 ± 7 days was found in Axim, and the shortest length of the rainy season of 140 ± 5 days was seen in Navrongo.
In general, rainfall onset occurred from March-May across the country, starting from the west coast through the forest and transition zones to the savannah zone of the country. The rainfall cessation dates in the forest zone were in the third dekad of October to the first dekad of November, and the length of the rainy season was within 225-240 days. The cessation dates of the coastal zone were within second and third dekad of October, and the length of rainy season was in the range of 210-220 days. Furthermore, the cessation dates of the transition zone were in the second to third dekad of October, and the length of the rainy season was in the range of 170-225 days. Rainfall in the savannah zone had cessation dates within the third dekad of September to the first dekad of October, and the length of the rainy season was within 140-180 days. Biases in the onset, cessation and length of the rainy season were mostly less than 10 days across the entire country, and the root mean square error (RMSE) was in the range of 5-25 days. These findings demonstrate that the onset derived from the cumulative rainfall amount and the rainy days are in consistent agreement.
The wavelet analysis was carried out to assess the inter-annual variability of the onset and cessation for each agro-ecological zone. The wavelet power spectrum, its significant peaks and variances showed evidence of variability in the rainfall onset and cessation dates across the country. The coastal and forest zones showed 2-8-and 2-4-year band variability in the onsets and cessations. Similarly, the transition and the savannah zones showed 2-4-and 4-8-year onset and cessation variability.
These findings support the south to north oscillation of ITD, which modulates the pressure system of West African Monsoon [26,27,37,38]. In addition, local convective activities resulting from the nature of the vegetation and the terrain are known to play a critical role in rainfall onset and cessation [39]. For example, the Togo-Akwapim ranges, which stretch from Togo through to the central part of the forest zone of Ghana, enhance the convective activities of the forest zone, and areas in the windward side of the mountains are strongly affected. This is because, when a moisture-laden airflow encounters a terrain, the air becomes buoyantly unstable and could enhance precipitation, depending on the strength of the cross-barrier component of the upstream airflow, the degree of the thermodynamic stability of the incoming flow and the height of the terrain barrier [39]. Other atmospheric feedback mechanisms, such as the surface feedback, dynamic instability, sea surface temperature (SST) and influence of transient tropical waves [40], could account for the seasonal and the interannual variability in the rainfall onset and cessation of the region. Further study using a dynamic climate model to investigate the contribution of topography, dynamics and thermodynamics processes on the variability of onset and cessation is under consideration. These findings have implications for rain-fed agricultural practices, vector-and water-borne disease control, water resource management, other socio-economic activities and food security in Ghana.