Minor Seanonal Rainfall Variability over Southern Ghana

2 Lancaster University, Lancaster, England; f.otu-larbi@lancaster.ac.uk * Correspondence: braimahmm@gmail.com Abstract: Rainfall variability has resulted in extreme events like devastating floods and droughts which is the main cause of human vulnerability to precipitation in West Africa. Attempts have been made by previous studies to understand rainfall variability over Ghana but these have mostly focused on the major rainy season of AprilJuly, leaving a gap in our understanding of the variability in the September-November season which is a very important aspect of the Ghanaian climate system. The current study seeks to close this knowledge gap by employing statistical tools to quantify variabilities in rainfall amounts, rain days and extreme precipitation indices in the minor rainfall season over Ghana. We find extremely high variability in rainfall with Coefficient of variation (CV) between 25.3% and 70.8%, and moderate to high variability in rain days (CV=14.0% 48.8%). Rainfall amount was found to be higher over the middle sector (262.7 mm – 400.2 mm) but lowest over the east coast (125.2 mm – 181.8 mm). Analysis of the second rainfall season using Mankandell Test presents a nonsignificant trend of rainfall amount, and extreme indices (R10, R20, R95p and R99p) for many places in southern Ghana. Rainfall Anomaly Indices show that the middle sector recorded above normal precipitation which is the opposite for areas in the transition zone. The result of this work provides a good understanding of rainfall in the minor rainfall season and may be used for planning purposes.


Introduction
Rainfall variability over West Africa is essential for various activities such as forecasting for early warning systems, hydroelectric power generation, agriculture and food security. Literature has shown that the past five decades provides evidence of increasing rainfall extremes and frequency over west Africa [1][2][3]. The increase in the rainfall extremes raises much concern about whether high impact weather events will worsen in the near future. However, the scarcity of water resources in some parts of the West African sub-region has made it necessary for a good understanding of the rainfall regimes and variability. Rainfall variability has resulted in extreme events like devastating floods and droughts in the history of West Africa. For instance, the Sahel region of West Africa experienced severe drought and famine during the early 1970s as a result of rainfall irregularities [4]. Floods are the main environmental challenge affecting people in West Africa. Example of such disastrous cases of flood due to extreme precipitation over West Africa includes 39 deaths in Abidjan (Côte d'Ivoire) in 2014, damages due to runoff of the Bagre Dam in Burkina Faso in 1994 and 2009 [5], death of 154 people in Accra (Ghana) in 2015 [6] and and two-stage rainfall-triggered landslide which claimed 1100 lives and affected 5000 people in Freetown peninsula, Sierra Leone in 2017 [7] . In Ghana, rainfall-related floods affected 3.81 million people from 1968 to 2011 [8,9]. Many studies have contributed to the understanding of rainfall regimes over west Africa and more specifically the southern portions of the sub-region [1,3,[10][11][12][13]. Ta et al. (2016) [1] analyzed rainfall trends and variability and found out that the penetration of monsoon flows into the land influences extreme rainfall over West Africa. Many of the previous studies give a general trend of rainfall over the African continent and therefore necessary to zoom into Ghana and conduct an analysis of the rainfall seasons [13,14]. In Ghana, major studies have been done on annual and seasonal scale rainfall trends and variability. For example, Baidu et al. (2017) [10] analyzed rainfall

Description of Data
In situ rainfall observations from 1981 to 2018 for seventeen stations (Figure 1) over southern Ghana were obtained from the Ghana Meteorological Agency (GMet). Rainfall data used in this work has been verified at the quality control unit of GMet. For this analysis, we group stations into four (4) sections based on their geographical locations as shown in Table 1. Data gaps in Akatsi were filled with data from a rainfall station which is situated about 10km away from Akatsi synoptic station.

Rainfall Amount, Rain Days and Extreme Precipitation
Rainfall totals and total rain days for September to November in each year were calculated for each station. For this analysis, we consider a rainy day as an event of rainfall amount greater or equal to 0.1 mm [15,26]. Descriptive statistical parameters which include the maximum, minimum, mean, skewness, kurtosis, and coefficient of variation generated for each station were calculated. Coefficient of variation (CV) explains the degree of rainfall amount and frequency variability in the minor season and its stated as less (CV < 20%), moderate (20% < CV < 30%), high (CV > 30%), very high (CV > 40%), and extremely high (CV > 70%) [27,28]. We use four indices (R10, R20, R95p, and R99p) from the World Meteorological Organization's list of Experts Team on Climate Change Detection Indices (ETCCDI) for our analysis [29][30][31][32]. Heavy Precipitation Days (R10)precipitation greater or equal to 10 mm, and Very Heavy Precipitation Days (R20) -number of days with precipitation greater or equal to 20 mm, were calculated for each year in each station. Very Wet Days (R95p), -number of days with precipitation greater than the 95 th percentile and Extreme Wet Days (R99p) -number of days where precipitation is greater than the 99 th percentile were also calculated for each station.

Trend Analysis
Mann Kendall Trend Test (M-K test) is a non-parametric test which has been widely used for analysis of rainfall trends in different parts of the world, and we use this test to find trends of rainfall, rain days and extreme indices for the minor season [33 -39]. In the M-K test the null hypothesis assumes there is no trend and it's tested against the alternative hypothesis (there is trend). The Mann Kendell Statistics S, the Variance of S V(S) and the standard test statistics Z are stated mathematically by where and are the time series of observations in order of chronology, n is the length of time series, is the number of ties for the pth value of observation, q is the number of tied values.
High positive values of Mann Kendell Statistics S, Variance of S V(S) and standard test statistics Z indicates an increasing trend whilst high negative values show decreasing trends. Sen's Slope Estimator Technique is a non parametric method of estimating the magnitude of a trend based on the method of least squares [40 -41] and we apply this method to find magnitudes of trends in the minor season.
Where is the median of the values of and its symbolized as the Sen's Slope Estimator. Positive values of Qi show an increasing trend, and negative Qi values represent decreasing trends. In this study python programming language package (pyMannKendall) was used for the M-K test at a confidence limit of 95% which returns the trend, h (True if the trend is present and False if there is no trend), p-value (p), normalized test statistics (z), Kendall Tau (Tau), Mann-Kendall's Score (s), variance S (var_s) and the Sen's slope (S Slope) [42]. Mann-Kendall Trend test (M-K test) were carried out on total rainfall amount, rain days, R10, R20, R95p and R99p. An equation of a linear regression is given mathematically by y = + (7) where is the dependent variable, is the independent variable, is the intercept and is the slope of the line. Positive values of m indicate increase in trend, and negative m values indicate decreasing trend [39]. For this analysis we plot rainfall amounts, and rain days against the years and a slope is generated which tell us whether there are decreasing or increasing trends.

Standardized Anomaly Index Estimation Model
Standardized Anomaly Index (SAI), a commonly used index for regional climate studies was used here to calculate minor seasonal rainfall anomalies for 2014 -2018, using the average rainfall from 1981 to 2010 as the long-term mean (equation 8). SAI is stated mathematically as = σ (8) i= 1, 2,3, 4, 5,…..n Where is the SAI, r is the mean of observations, is the long-term mean, and σ is the standard deviation of the observation. Positive values of is considered above long-term mean, and negative values of depicts below the long-term mean [39]. SAI = 0 means normal rainfall, SAI > 0 means above normal and SAI < 0 below normal rainfall, SAI >= 2 is an extreme wetness, and SAI< = -2 is an extreme dryness

Results and Discussion
In section 3. 1 We present the results of descriptive statistics of rainfall and rain days. Section 3.2 presents results of regression analysis which include linear regression equations. Mann Kendall test results are presented in Section 3.3 and 3.4. Section 3.5 presents standardized anomaly index results for the minor season. For this work, rainfall stations may be presented either as individual stations or as part of a group of stations in a given section.  , and the second-highest maximum of 58 rainy days with a coefficient of variation of 14.2%, a signal of very low standard deviation with respect to the mean. Rainfall stations over the east coast (5.5 °N 0.5 °W -5.5 °N 1 °E) have comparatively lower rainfall amounts and corresponding rainy days. Citing Tema as a typical example in the east coast, the station has the lowest mean rainfall amount of 125.2mm and lowest maximum rainfall of 373.5 mm with a slightly skewed (skewness of 1.095) data, and insignificant outliers (kurtosis of 1.5). Tema doubles as having a symmetrical number of rain days (skewness= -0.1), with the lowest number of mean rainy days (15.6 days), the lowest maximum of 26 rain days with no outliers (kurtosis of -0.6), and having 37% variability. From our analysis, the west coast has higher rainfall amount and rain days as compared with the eastern side of the coast with Axim known to experience the highest rainfall in Ghana per climatology being a typical example. Rainfall amounts show coefficient of variation (CV) values between 25.3% and 70.8% (moderate and extremely high variability). The degree of variability in the total number of rainy days is between 14.0% and 48.8% which is moderate to high variability (Table 3). Kurtosis values of rainfall amount fall within the normal range, an indication of no or insignificant outliers in the data [43][44][45]. The number of rain days also have kurtosis values which fall within the normal value range for all stations except Wenchi. Wenchi, a station located over the transition zone, has mean rain days of 37.6 which is among the stations with the highest mean rain days. However, Wenchi has Kurtosis value which is greater than 7.0 and skewness of -2.0 which makes us suggest the presence of few outliers and non-symmetrical respectively for the dataset. Moreover, the rainfall amount and number of rainy days for many rainfall stations are fairly symmetrical (skewness between -0.5 and 0.5).

Descriptive Statistics of Rainfall Amount and Rain Days
The absence of significant outliers portrays the level of certainty in the data set which was used in this work. For a clearer overview of rainfall amount and frequency during the minor season, we can say that rainfall amounts in the minor season are highest over the middle sector and lowest over the east coast. Perhaps, green vegetation and the topography of the middle sector can be a contributing factor that influences higher rainfall amounts and rain days.

Regression Analysis of Minor Seasonal Total Rainfall Amount and Rain Days
The results of regression analysis for both rainfall amount and number of rain days were analysed with a time series plot generated for each synoptic station.   Ho is the only station which shows a slight decreasing trend of rain days as explained by the regression model. It is interesting to know that Ho in the middle sector, is located in the Volta region of Ghana where previous study [11] found oscillatory trends rainfall at annual scales. The regression model explains variability of up to 27% in the number of rain days with the middle sector and the western coast exhibiting greater variability ( Figure S1, S2, S3 and S4). Result of regression in this work this work coincides clearly with results of most recent study [19] within the study area which found an increasing trend of annual rainfall totals for Accra and Kumasi In the Mann Kendall trend test, if p<0.05, the trend is significant but if p>0.05, the trend is considered insignificant or simply no trend [46][47]. In general, all the stations showed a positive Sen's Slope (between S Slope= 0.40 and 4.41) for rainfall indicating at least an increasing trend (Table 4) With these results, we can say that there is an increasing trend of rainfall amounts over southern Ghana, but this increase is not significant at a 95% confidence limit over most places. Krachi have a significant increase in rain days and non-significant trends in rainfall amounts. However, Accra, Saltpond, Takoradi, Abetifi, Ho, Akuse, Akatsi, Sunyani, and Wenchi show non-significant trends for both rainfall amounts and rain days. The significant increase in rainfall frequency and non-significant increase in rainfall amounts from 1981 to 2018 over many stations in southern Ghana could be a result of lesser rainfall amounts which are recorded at the beginning of the rainfall season as reported by [12] 3  Akim Oda is the only station having an increasing trend of R10 (Table 6). Abetifi, Akim Oda,Sefwi Bekwai, and Kumasi located over the middle sector (5.5 °N -7.0 °N) show significant trends of R 20 ( Table 7). The other rainfall stations have no significant trends for R10 and R20 since their p-values are greater than 0.05 (p > 0.05). Concerning R95p and from Table 8, we can see that Akatsi, located over the middle sector has increasing trends ((p<0.05) of very wet days. All the 17 stations studied have no significant trends (p>0.05) of extreme wet days (R99p) during the 37-year period (Table S1). R95p and R99p show Sen's Slopes values of zero (0) or p-value greater than 0.05 for rainfall stations which is an indication of no trends. From the above analysis, 4 stations in the middle sector have increasing trends for R20 and this compliment the trends of rain days which was revealed earlier in this work. It is worth knowing that all these stations in the middle sector are located in forest areas of Ghana, and it is therefore not surprising to find out increasing trends. Therefore, at a confidence level of 95%, we can explain that some areas in the middle sector have experienced significant increase of extreme rain days. It is worth reminding that Kumasi and Akim Oda are the two stations which showed significant increase in rainfall amount and at the same time showing increasing trends of R20. It is therefore not surprising that some areas like Kumasi, Koforidua etc. experienced recent cases of extreme weather events such as floods and hail stones [48][49][50] It can be observed that stations in the transition zone (7.0 °N -8.0 °N ) is captured with non-increasing trends of R10, R20, R95p and R99p which slightly differs from the findings of [30] which found decreasing trends of wet indices over areas between on the volta lake (7.0 °N -9.5 °N ) on annual scales.

Standardized Anomaly Index for Minor Seasonal Rainfall Amount
The general rainfall standardized anomaly indices with more emphasis on each sector was described in section 2.2. To further understand the recent variabilities, a spatial map of the minor season rainfall standardized anomaly indices for 2014 -2018 were generated. This 5-year period was used to show whether rainfall recorded in recent years are above or below the long term mean In this analysis, the mean standardized index is zero which is the benchmark for comparison. Indices greater than zero are wet and those less than zero are dry. Generally, 2015 stands out as the dry year while 2016 stands out as the wet year. In 2015, Ghana experienced a general decline in rainfall with many areas having SAI less than zero. Notwithstanding the extreme dryness in 2015, few places in the west coast and middle sector (6.5 ° N 2 ° W -6 ° N 0 °E) were wet with SAI greater than 1.0. In addition to a few areas, 2016 is a wet year over most places with SAI between 1.0 and 3.0 which are extreme high rainfall anomalies. In 2018, the coastal and middle sectors experienced greater improvements in annual rainfall with standardized index between 0.5 and 2.5. Many places located in the transition zone (7.0 °N -8.0 °N) experienced rainfall deficits in 2015. Previous studies [30] reported that annual rainfall extreme indices over areas along the Volta Lake (From 7.0 N Northwards) have decreasing trends of wet indices and the anomaly plots in this work for the transition zone is a clear confirmation. Ho, Akuse, Akatsi, and Kete Krachi are in the Volta region, (0.0 °E -0.5 °E) and their oscillatory anomalies in the minor seasonal rainfall confirm the findings of [11] which reveals that rainfall trends within all zones of the volta region are oscillatory. Again, throughout this work, we observe the uniqueness of Akim Oda in terms of higher rainfall amounts, higher rain days, increasing R20 and the presence of positive standardized rainfall anomalies.

Conclusion
In this study we perform analysis of the minor rainfall season with the aim of finding recent trends and extremes in rainfall. We perform descriptive statistics, and find trends of rainfall amount, rain days, and some extreme indicator Even though variation in rainfall amount and rain days were found to be high, trends of rainfall amounts, and extreme indicators were generally non-significant for the 37 year-period. The results of descriptive statistics of the area show that rainfall amounts are highest over the middle sector and lowest over the eastern coast. Moderate to an extremely high degree of variation in rainfall amount (between CV=25.3% and 70.8%), and moderate to high variation (between CV=14.0% and 48.8%) in rain days were shown by the coefficient of variation. Information on skewness and kurtosis of rainfall amount and its frequency in the minor season illustrate the existence of a fairly symmetrical data set with the presence of no significant outliers. Trends of total rainfall amounts and rain days in the minor season from 1981 to 2018 were studied using linear regression and Mann-Kendall's trend test. Apart from Ho which has shown decreasing trends of rain days, linear regression equations show positive slopes for both total rainfall amount and number of rainy days (most of the slopes were very small). Variability in total rainfall amounts between 0.2% and 15.6% for rainfall amount, and between 0.3% and 21.9% were explained by the linear regression analysis (R-square values). Like the regression analysis, positive Sens slopes were generated for rainfall amounts (S Slope= 0.40 -4.41) and rain days (S slope= 0.000-0.333) in all the 17 stations studied and this is a sign of at least an increasing trend. Even though positive slopes were found in the regression analysis, Mann Kendall test which was conducted at a 95% confidence limit indicated non-significant trends (p > 0.05) of total rainfall amounts for 15 out of 17 stations. In terms of rainfall frequency, 7 out of 17 stations generated a statistically significant increase in rain days (p < 0.05). Man Kendall test results for Kumasi and Akim Oda in the middle produce significant increasing trends (p < 0.05) for rainfall both rainfall amount and rain days. The significant increase in rain days (7 stations) in the minor season over southern Ghana is in line with a previous study conducted over southern West Africa [3] that reported increasing trends of rainfall frequency. Man-kendell tests conducted on wet indices show significant increasing trends of R20 for 4 stations situated over the middle sector. Generally, non-significant trends of R95p and R99p, and R10 were found. The transition zone however, show no trends for all the indices (R10, R20, R95 and R99). Rainfall standardized anomalies which were computed and plotted at spatial and station levels, show that 2015 is a dry year and 2016 is a wet year. The middle sector experienced lots of positive rainfall anomalies over the 5-year period. Anomaly indices generated over the transition zone were comparatively below normal. From this analysis, it can be deduced that at a significant level of 95%, trends of rainfall amounts, and extreme rain days in the minor season were found to be non-significant for the second rainfall season in many places of southern Ghana. However, some stations over the middle sector experienced significant increasing trends of rain days and R20 during the 37 -year period. If the current trend of increasing rainfall continues in the coming years for the middle sector, the minor rainfall season may experience more frequent rainfall.