Impact of Large-Scale Climate Indices on Meteorological Drought of Coastal Ghana

The devastating effects of drought on agriculture, water resources, and other socioeconomic activities have severe consequences on food security and water resource management. Understanding the mechanism that drives drought and predicting its variability is important for enhancing early warning and disaster risk management. In this study, meteorological droughts over six coastal synoptic stations were investigated using three-month Standardized Precipitation Index (SPI). The dry seasons of November-December-January (NDJ), December-January-February (DJF), and January-February-March (JFM) were the focal seasons for the study. Trends of dry seasons SPIs were evaluated using seasonal Mann–Kendall test. The relationship between drought SPI and ocean-atmosphere climate indices and their predictive ability were assessed using Pearson correlation and Akaike Information Criterion (AIC) stepwise regression method to select best climate indices at lagged timestep that fit the SPI. The SPI exhibited moderate to severe drought during the dry seasons. Accra exhibited a significant increasing SPI trend in JFM, NDJ, and DJF seasons. Besides, Saltpond during DJF, Tema, and Axim in NDJ season showed significant increasing trend of SPI. In recent years, SPIs in dry seasons are increasing, an indication of weak drought intensity, and the catchment areas are becoming wetter in the traditional dry seasons. Direct (inverse) relationship was established between dry seasons SPIs and Atlantic (equatorial Pacific) ocean's climate indices. The significant climate indices modulating drought SPIs at different time lags are a combination of either Nino 3.4, Nino 4, Nino 3, Nino 1 + 2, TNA, TSA, AMM, or AMO for a given station. The AIC stepwise regression model explained up to 48% of the variance in the drought SPI and indicates Nino 3.4, Nino 4, Nino 3, Nino 1 + 2, TNA, TSA, AMM, and AMO have great potential for seasonal drought prediction over Coastal Ghana.


Introduction
Drought is a climate phenomenon on land where a given location experiences below normal precipitation. It can happen on a different timescale. e impacts of drought are visible in areas of agriculture, energy (hydroelectric production), and water resource management for both domestic and industrial use. To properly understand drought, four categories of drought exist depending on the physical impact on the environment [1]. ey are agricultural, meteorological, hydrological, and socioeconomic drought. Agricultural drought occurs when there is lack of needed soil moisture for plant growth; meteorological drought reflects on lack of precipitation from the atmosphere. Socioeconomic drought deals with a lack of water supply for the society, whereas hydrological drought focuses on deficiency in the amount of surface and groundwater [2]. Based on the type of drought and its complexity, several analysis techniques have been developed to analyze them. ese include Palmer Drought Severity Index (PDSI), Standardized Precipitation-Evapotranspiration Index (SPEI), Standardized Precipitation Index (SPI), and Normalized Difference Vegetation Index (NDVI) [3].
In sub-Saharan Africa, agriculture is predominantly rain-fed. e year-to-year variation of rainfall affects food security. Drought is a threat to agriculture in sub-Saharan Africa [4]. Studies have shown the devastating impact of drought on West Africa most especially over the Sahel region [5,6]. e Sahel region was plagued with prolonged drought periods for over a decade from 1970 [7,8].
Several factors facilitate and influence drought occurrence. ey can be triggered by anthropogenic or natural processes. Ocean-atmosphere mechanism is a natural process that contributes to drought occurrence on spatiotemporal timescales [9]. Variability of drought due to ocean-atmospheric mechanisms is triggered by changes in anomalous sea surface temperatures (SSTs) and sea-level pressure (SLP) of remote ocean-atmospheric teleconnection. e ocean's unique features of high heat capacity and memory are major drivers of global and African climate systems [10]. e role of SSTs and associated remotely forced phenomena on West Africa, Sahel, East Africa, and Southern Africa's climate has been investigated [11,12]. Key teleconnection factors linked to the rainfall over Africa are North Atlantic Oscillation (NAO), El Niño Southern Oscillation (ENSO), Atlantic Meridional Mode (AMM), and Atlantic Multidecadal Oscillation (AMO), among others. On drought, ENSO is known to have played a significant role in contributing to drought occurrence at different timescales in the US and Columbia [13][14][15]. In West Africa, the prolonged drought situation in the Sahel was connected to the influence of large ocean-atmosphere climate indices like ENSO, AMO, and Tropical Atlantic Oceanic indices [16].
Although studies done on drought in West Africa have primarily focused on Sahel, on a country scale, little of such studies have been done over Ghana. Ghana, a country within the West Africa subregion, had its share of the drought menace in the 1980s but was not on a scale as the Sahel [17]. Climatologically, the country has a dry and rainy season, bimodal rainfall season in the south, and unimodal season in the north of the country [18,19]. Precipitation over Ghana is largely driven by the position of the Intertropical Discontinuity (ITD) which determines the seasonal cycle of rainfall. e ITD forms when two air masses of moist southwest trade winds meet with northeast trade winds. e dry air mass emanates from the Sahara Desert, and the moist air comes from the South Atlantic Ocean.
Ocean-atmospheric drivers of drought over Ghana are not clearly known. is hinders the effort of meteorologists or climate scientists in making projections on sub-seasonal to seasonal timescale. To address this gap, this study seeks to investigate meteorological drought and its trends, ascertain the impact of remote global ocean climate indices on drought, and understand its predictability. e fundamental understanding of mechanisms driving drought are critical to develop an early warning system for climate disaster risk management.
In this study, the statistical analysis techniques were adopted. ey include computing three (3) months' Standardized Precipitation Index (SPI) of rainfall, seasonal Mann-Kendall trend test, correlation analysis, and Akaike Information Criterion (AIC) [20] stepwise regression between the SPI and the climate indices. is paper is arranged as follows: Section 2 gives a brief description of the study area, Section 3 highlights the data and methods, Section 4 presents the results and discussions, and Section 5 is the conclusion.

Study Area
Ghana is geographically located between latitude 4.5°N and 11.5°N and longitude 3.5°W and 1.5°E (Figure 1). e country has both dry and wet seasons. e latitudinal oscillation of the ITD mainly influences the climate of Ghana.  Coastal rainfall season starts from March through to July and subsides in August ( Figure 2). e month of August is usually termed as the little dry spell. e minor rainy season commences in September and ends in November. e dry season commences in late November to February ( Figure 2). During the rainy season, most coastal stations experience more intense rainfall in June with maximum rainfall amount of 929.20 mm, 598.30 mm, 506.50 mm, 471.80 mm, 450 mm, and 420 mm, respectively for Axim, Ada, Takoradi, Saltpond, Tema, and Accra. Stations in the West Coast of Ghana (Axim, Takoradi, and Saltpond) have higher rainfall as compared to stations of the East Coast (Accra and Tema), although the characteristics of rainfall in Ada are like West Coast stations. e dry season occurs from November to March.

Rainfall Data.
Monthly daily rainfall data from six meteorological coastal synoptic stations of Ghana Meteorological Agency (GMet) were sourced for this study. e datasets were subjected to a quality control test to determine the percentage missing values and any outliers within the data. From the data quality analysis, the overall percentage missing values in the datasets were 2% or less for some stations. Duration of data used for the analysis spans from 1980 to 2014 (35 years).

Climate Indices.
Climate indices are diagnostic quantities used to understand the current and future state of a climate system. Each of these indices was created to monitor an aspect of the climate system around the world. e indices are computed from data records of oceanic and atmospheric processes. Several of such indices have been developed, ten climate indices were selected to understand its linkages to meteorological drought over Coastal Ghana. e oceanic indices are based on SST anomalies over a given location and, in some instances, differences between SLP. e climate indices used for this study are Atlantic multidecadal oscillation (AMO), Niño 1 + 2, Niño 3, Niño 3.4, and Niño 4, Atlantic Meridional Mode (AMM), Tropical North Atlantic (TNA), Tropical South Atlantic (TSA), North Atlantic Oscillation (NAO), and Southern Oscillation Index (SOI). AMO is derived by averaging SST anomaly over the North Atlantic Ocean within latitude 0-80°N. e tropical Pacific indices are Niño 1 + 2 of SST averages over latitude 0-10°S and 90°W-80°W, Niño 3 index (average SSTfrom 5°N to 5°S and 150°W-90°W), Niño 3.4 (5°N-5°S and 170-120°W), and Niño 4 (5°N-5°S and 160°E-150°W). AMM is crossequatorial tropical Atlantic SST gradient within latitude 21°S-32°N and longitude 74°W-15°E [21], TNA (5.5°N-23.5°N and 15°W-57.5°W), and TSA (Eq-20°S and 10°E-30°W). NAO is based on sea-level pressure difference between Azores High and the Subpolar Low; SOI is also SLP differences between Tahiti and Darwin, Australia. All the climate indices were sourced from the National Oceanic and Atmospheric Administration's (NOAA) Earth Systems Research Laboratory (ESRL) accessible at https://psl.noaa. gov/data/climateindices/list/. e data selected span from 1980 to 2014.

Methods.
e methods adopted for the study were Standardized Precipitation Index (SPI), Seasonal Mann-Kendall test, correlation, and AIC stepwise regression.

Standardized Precipitation Index (SPI).
e Standardized Precipitation Index [22] was developed to monitor precipitation anomalies such as droughts. SPI is used widely to monitor identified periods and duration of meteorological drought occurrence. e computation of SPI is done by using the long-term precipitation data and fitting it to gamma distribution and transforming to normal distribution. e gamma is expressed as the probability density function [23,24]. R function spi from SPEI package was used for the computation [25].
It is mathematically expressed as

Advances in Meteorology
where a and β represent the shape and scale parameters, x is the precipitation amount, and Γ(a) is the gamma function.
e estimation of Maximum Likelihood of a and β is given as  β � x/a, where A � (x) − (In(x))/n, n � number of observations. In SPI computation, precipitation data can be aggregated for 1, 3, 6, 12, and 24 months. e study uses three (3) months' SPI for analysis because of its usefulness in the primary agricultural region [26] that highlights the available moisture conditions at a given period.

3.3.2.
Seasonal Mann-Kendall Trend Test. e Mann-Kendall (MK) test determines the monotonic trend in the time series of the SPI indices [27,28]. e original MK test can be expressed mathematically as Note that whenever S > 0, then later observations in the time series tend to be larger than those that appeared earlier in the time series, while the reverse is true if S < 0. e variance of S is given by where σ 2 is the variance and t j are the sets of data points in j th term. e statistic S is approximately normally distributed with S is also related to Kendall's τ in the expression. τ � S/D, tau (τ) indicates the magnitude of the trend and D is given as e 95% confidence level was used to determine the significance of the trend analysis. If z is significant, the observed trend is significant and vice versa. Sen's nonparametric method [29] was also used to estimate the magnitude of trends. e null hypothesis of the original MK test was based on independent and randomly distributed data with no seasonality and serial correlation in the data. Once seasonality or serial correlation is present, using original MK may lead to false identification of trends [30]. In this study, the JFM, NDJ, and DJF SPI data are seasonal but the autocorrelation in time series [31] was tested which proved insignificant. Hence, the seasonal MK test proposed by [32] to eliminate the effect of seasonality was used for the trend analysis. e seasonal Mann-Kendall trend test is expressed as e variance of S ′ is given as where σ 2 g � Var(S g ); σ gh � Cov(S g , S h ).

Correlation.
To establish the relationship between the global climate indices and the SPI, the Pearson correlation was computed at a significance level of 5%. e multiple correlation between dry seasons SPIs, both concurrent and lagged seasonal climate indices, was done using rcorr() function [33] from

Stepwise Regression.
Stepwise regression fits the best regression models to the observed SPI by choosing the best models of predictive variables (climate indices) based on specific criteria. e stepwise regression was based on AIC criteria [20]. First, multiple regression (MR) analysis was done to filter statistically significant climate indices that fit the station's SPI as predictant [34,35]. e regression model can be represented mathematically as where . . , β i = coefficients of predictors, and ε = error margin.
Once the multiple regression is done, the AIC criteria for selection of best fit model are done. e stepAIC() function from the MASS R package [36] based on both forward and backward steps was used for the analysis. e results with minimum AIC values had the best goodness of fit and, hence, maintained to fit the model.

Drought Characterization in Coastal Ghana.
e meteorological drought over Coastal Ghana was characterized based on the SPI classification shown in Table 1 Drought conditions as shown in the SPI results exhibited interannual variations with moderate to severe drought conditions occurring nearly every half a decade for all stations (Figure 3). e SPI further suggests, drought in recent years does not last long for most stations.

Dry Seasons SPIs.
e study focuses on the dry seasonal composites of JFM, NDJ, and DJF to understand the trend and pattern of dry seasons SPIs. e dry seasons in Ghana are mostly characterized by dry dusty northeast trade winds emanating from the Sahara Desert to the Gulf of Guinea which makes the atmosphere dry and results in meteorological drought.

JFM.
In JFM, the season is characterized by drought based on the SPI classification. During this season, all coastal meteorological stations exhibit mild to moderate drought occurrence (Figure 4(a)). e pattern of SPI suggests from 2010 SPI increased for stations like Accra, Tema, and Ada. Accra indicates an increasing SPI during the period of study. Accra shows a significant increasing trend for the period of study. Stations like Saltpond and Tema depict positive tau and Sen's slope; nevertheless, the overall trend from 1984 to 2014 was not significant ( Table 2). ere is an indication that the intensity of drought over these stations is becoming weaker over the years; hence, coastal Ghana is becoming wetter during this season.

DJF.
During boreal winter (DJF), SPI showed mild to moderate drought conditions for all stations (Figure 4(c)) apart from Axim which had some severe drought conditions from 1983 to 1987 and 2009. e pattern of SPI for Accra and Saltpond had a significant increasing trend with tau of 0.28 and 0.24 and Sen's slope of 0.016 and 0.011, respectively (Table 3). Tema and Ada showed a rising SPI from 2005 to 2014 but the overall trend was not significant. e drought SPI demonstrated interannual variability.

NDJ. NDJ SPI demonstrates interannual variability
with Accra, Axim, and Saltpond having significant increasing monotonic trend (Table 4). e overall trend of stations like Tema and Takoradi was not significant but SPIs of recent years are gradually increasing. ere is an indication that Accra, Saltpond, Tema, and Takoradi stations are having more wetter conditions in recent years (Figure 4(b)).
Ada is becoming dryer in recent years with SPI decreasing. In general, meteorological stations in the West Coast of Ghana including Axim, Saltpond, and Takoradi showed signals of weaker drought SPI (see Figure 4(b)). During the NDJ season, the West Coast is becoming moist and drought conditions are becoming weaker.

Impact of Oceanic Indices on Drought Variability.
e relationship between the drought SPI and multiple climate indices was established using multiple correlation analysis and AIC stepwise regression. e correlation analysis was done for both concurrent and lagged seasons of the SPI. e AIC penalizes the predictors to select the best fit model of observed SPI. Climate indices that satisfy this AIC criteria were retained and used for the predictive model. Dry seasons SPIs were the focus of the AIC stepwise regression by using the first and second lagged timestep of each SPI season's corresponding climate indices.  e stepwise AIC regression points to different combinations of climate indices from the Atlantic and equatorial Pacific oceans as having the combined effect of modulating and explaining some percentage variance in the drought SPI. A comparison of the output of the model and the observed SPI is shown in Figures 5-7. Majority of the fitted SPI models for the various stations were significant with p-value less than 0.05. e predictive potential of these indices was determined by focusing on the first and second time lagged of NDJ, DJF,     Figures 5-7. e output of the stepwise regression analysis between lagged OND climate indices and drought SPI of NDJ shows combination of Nino 3, Nino 3.4, Nino 4, AMM, and NAO are the predictors that best fit the NDJ SPI of Accra with multiple R-square of 0.36 ( Figure 5). NDJ SPI of Tema and Ada were significantly controlled by a combination of Nino 1 + 2, Nino 4, AMM, AMO, TNA, TSA, and SOI as shown in Table 5 explaining 41% and 42% of total variance for Tema and Ada, respectively. Nino 3.4 and TSA were the only predictors for Saltpond. e stepwise regression models of Accra, Tema, Ada, and Saltpond were significant (Table 5). e second lag season to NDJ, i.e., SON, indicates the model fit of Accra, Saltpond, Axim, and Tema were significant with multiple R-square ranging from 0.20 to 0.36 with unique combinations of either AMO, Nino3.4, Nino 1 + 2, Nino 4, AMO, AMM, TSA, or TSA.

Relationship between Drought SPI and Climate
AIC stepwise regression of boreal winter (DJF) seasonal SPI and NDJ climate indices exhibits that Nino 3, Nino 3.4, Nino 4, AMM, TNA, TSA, and NAO were the best predictors identified as forming the predictors to fit DJF SPI. e models fit explained between 21% and 48% of the variability in DJF season SPI (Table 6 and Figure 6). e model fit for all stations was significant except Tema. At the second lag season of OND, Accra was the only station with significant model fit that could explain 36%       of DJF SPI variance. e climate indices that constitute the total model were like the identified climate indices of NDJ lagged season (see Figure 6 and Table 6). e stepwise regression between JFM seasonal SPI and DJF climate indices also indicates Nino 1 + 2, Nino 3, Nino 4, and Nino 3.4 are major drivers of JFM SPI (Table 7). e DJF model fit of Accra, Saltpond, and Tema was significant, explaining 23 to 31% of total variance in JFM SPI. Axim did not show any of the climate indices having any connections with the SPI. e DJF lagged seasonal indices showed AMM, AMO, TSA, TNA, Nino 1 + 2, Nino 3, Nino 4, and Nino 3.4 indices as prominent   Based on the results of both correlation and AIC stepwise regression, the prominent indices linked to drought SPI variability over Coastal Ghana are AMM, AMO, NAO, TNA, TSA, Nino 1 + 2, Nino 3, Nino 3.4, and Nino 4. e physical processes associated with climate indices are linked to changes in the orientation of Intertropical Convergence Zone (ITCZ) and the walker circulations. e differential heating over North and South Atlantic Ocean shifts ITCZ meridionally [16,41] and displaces the anomalous positive rainfall toward the position of the ITCZ. AMM in its high peak enhances latent heat and evaporation over the Atlantic Ocean causing variations in trade winds circulation. AMM also enhances moisture convergence over the equatorial Atlantic leaving the Gulf of Guinea dry. When AMM is in the cold phase, the ITCZ moves from the   equatorial belt northward which favours precipitation at the Gulf of Guinea [41]. e results suggest a direct relationship between AMM and dry seasons' SPI implies that high (low) AMM leads to low (high) SPI over Coastal Ghana. AMO, a leading mode of North Atlantic SST variability on multi-decadal timescale, is known to have played an important role in multi-decadal rainfall variability over the Sahel [16,42]. e warm phase of AMO is associated with the northward displacement of the ITCZ which pushes the precipitation belt northward. e warm phase of AMO weakens the northerly component of trade winds and enhances the trade winds from the south driving rain belt to the Sahel thereby depriving the Gulf of Guinea the need for precipitation. However, the cold phase of AMO displaces ITCZ southward, enhancing precipitation over Gulf of Guinea and leaving the Sahel dry [43]. e positive relationship between SPI and AMO demonstrates that AMO signals have direct remote impact on SPI over Coastal Ghana.
ENSO indices (Nino 1 + 2, Nino 3, Nino 4, and Nino 3.4) have a negative impact on dry season SPI; therefore, when ENSO is in its high phase, the seasonal SPI will be in the low phase and vice versa. During El Niño (La Niña) years, there is coupling between the ocean and the atmosphere. e ocean-atmospheric coupling during El Niño (La Niña) weakens (strengthens) the TEJ and the low-level African Easterly Jet that drives convective systems over the tropical Africa through a process called atmospheric bridge where the walker circulation would be perturbed to either suppress or enhance rainfall [44]. In an El Nino event, above normal atmospheric subsidence of the walker circulation over tropical Africa results in reduction of rainfall. e La Niña conditions enhanced rainfall in most parts of tropical Africa [45,46]. e inverse relationship established suggests when ENSO is in the high phase (El Niño) and low phase of ENSO (La Niña), the seasonal SPI over Coastal Ghana will be in the low phase and high phase, respectively. e study suggest the combined effect of AMM, AMO, NAO, TNA, TSA, Nino 1 + 2, Nino 3, Nino 3.4, and Nino 4 climate indices has great potential for predicting drought SPI over Coastal Ghana.

Conclusion
e study investigated the meteorological drought over coastal Ghana. e SPI characterizes the drought period and intensity from 1980 to 2014. e relationship between the SPI and set of climate indices was established using correlation and AIC stepwise regression methods for fitting predictive models. On an annual timescale, historical meteorological drought periods between 1983 to 1984, 1988, 1993, 1997 to 1998, 2000 to 2001, 2009, and 2012-2013 were identified. e dry seasons of JFM, NDJ, and DJF were the focal seasons of the study. e drought intensity over Coastal Ghana during these seasons can be classified as moderate to severe drought for most stations. e durations of drought over Coastal Ghana in recent times are short with weak intensities and some stations showing significant increase in the SPIs for these seasons. An indication of the wetting tendency (decrease in drought conditions) over Coastal Ghana in recent years. e temporal patterns of drought SPIs exhibited interannual variability.
e Atlantic Ocean's AMM, AMO, TSA, and TSA demonstrated positive correlation with dry seasons SPIs whereas the equatorial Pacific Nino 1 + 2, Nino 3, Nino 3.4, and Nino 4 showed negative association with the dry season SPIs. In general, the dominant climate indices impacting drought SPIs vary with respect to a given meteorological station. However, the prominent climate indices identified for most meteorological stations are AMM, AMO, NAO, TNA, TSA, Nino 1 + 2, Nino 3, Nino 3.4, and Nino 4. e model fit for the SPIs had multiple R-square ranging from 0.1 to 0.48 for some stations. e result suggests the combination of the identified climate indices can explain significant percentage variance in the dry seasons SPIs. e impact of AMM, AMO, NAO, TNA, TSA, Nino 1 + 2, Nino 3, Nino 3.4, and Nino 4 on dry seasons SPIs has been established and is useful for developing a robust predictive tool for drought in Coastal Ghana for disaster risk management.
Data Availability e rainfall data used in this study were obtained from the Ghana Meteorological Agency, and climate indices were downloaded from the website of the National Oceanic and Atmospheric Administration's (NOAA) Earth Systems Research Laboratory (ESRL).

Conflicts of Interest
e authors declare that they have no conflicts of interest.