Tettey et al.

Earth Perspectives (2017) 4:6

DOI 10.1186/s40322-017-0042-6

RESEARCH ARTICLE Open Access

Markov chain analysis of the rainfall

patterns of five geographical locations in
the south eastern coast of Ghana
Meshach Tettey1,2, Francis T. Oduro1, David Adedia3 and Daniel A. Abaye3*

Abstract
This study develops an objective rainfall pattern assessment through Markov chain analysis using daily rainfall data
from 1980 to 2010, a period of 30 years, for five cities or towns along the south eastern coastal belt of Ghana; Cape
Coast, Accra, Akuse, Akatsi and Keta. Transition matrices were computed for each town and each month using the
conditional probability of rain or no rain on a particular day given that it rained or did not rain on the previous day.
The steady state transition matrices and the steady state probability vectors were also computed for each town and
each month. It was found that, the rainy or dry season pattern observed using the monthly steady state rainfall vectors
tended to reflect the monthly rainfall time series trajectory. Overall, the probability of rain on any day was low to
average: Keta 0.227, Akuse 0.382, Accra 0.467, Cape Coast, 0.50 and Akatsi 0.50. In particular, for Accra, the rainy
season was observed to be in the months of May to June and September to October. We also determined that
the probability of rainfall generally tended to increase from east to west along the south eastern coast of Ghana.
Keywords: Markov chain analysis, Transition state matrix, Rainfall prediction, Probability of rain or dry, South Eastern
Coast of Ghana

Background or changes in rainfall patterns. Most reports were either

Global warming has been confirmed to be affecting anecdotal or in computer models, rather than from obser-
regional and continental rainfall patterns of the world, vation. One problem for researchers has been lack of accur-
with the recorded evidence showing that there is an in- ate, long-term rainfall data from around the world that
crease in precipitation in Canada, northern Europe and would enable them to distinguish between regional or
northern Russia, while vast areas of sub-Saharan Africa, cyclical shifts in rainfall patterns.
southern India and South-East Asia are becoming dryer In their seminal work, von Storch and Zwiers (1999)
(Groisman et al. 2005; Kharin et al. 2005; Westral et al. analysed two data-sets of global rainfall patterns cover-
2013; Fletcher et al. 2015). Changes in precipitation pat- ing the years 1925 to 1999. A comparison was made be-
terns have significant effects on ecosystems, agriculture tween their data and that of 14 powerful computer
and human regions that are sensitive to changes in precipi- models that simulate the world’s climate systems and
tation, such as the Sahel (Groisman et al. 2005; Dore, 2005; found a remarkably close fit. Their work concluded that,
IPCC 2014). Precipitation patterns are certain to be affected over the 75-year period under study, global warming
by global warming, because air and sea temperatures and ‘contributed significantly’ to increases in precipitation in
sea-level atmospheric pressure, which are the underlying the northern hemisphere’s mid-latitudes, a region between
forces behind these patterns, are already changing. How- 40° and 70° north. In contrast, the northern hemisphere’s
ever, until the beginning of the twenty-first Century, there tropics and subtropics, a region spanning from the Equator
was little observable evidence to support this interference to 30° north, became drier. These trends have since been
confirmed by many other reports (e.g. Groisman et al.
* Correspondence: [email protected] 2005; Westra et al. 2013; IPCC, 2014). Nevertheless, tor-
3
School of Basic and Biomedical Sciences, University of Health and Allied rential rains have continued to deluge locations in these
Sciences, PMB, 31 Ho, VR, Ghana drier zones including Accra, Ghana’s capital creating floods
Full list of author information is available at the end of the article

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made.
Tettey et al. Earth Perspectives (2017) 4:6 Page 2 of 11

in various parts that have caused great havoc to property hydrology that finds its greatest applications in the
and loss of lives, on 22 June 2010 (Coffie, 2010). In the first design and operations of water resources, engineering
two decades of the twenty-first century, such scenes have works as well as agricultural systems (Srikanthan and
been repeated across all continents. McMahon, 2005). Quantification of rainfall is generally
Ghana is one of the countries in the world whose done by using pluvial maps and Intensity-Duration-
economy is highly dependent on rain-fed agriculture. Frequency (IDF) curves (Raiford et al. 2007). A more
Ghana is experiencing increasing temperatures and recent interest in rainfall modeling is the perspective of
unpredictable rainfall patterns (EPA, Ghana 2000; using model parameters to characterize changes in the
Nelson and Agbey, 2005). This has led to the National precipitation patterns because of the greenhouse effect
Meteorological Services Agency of Ghana (NMSAG and climate change. Available models have been linked
http://www.meteo.gov.gh/website/#) actively participat- to the temporal and spatial scale required for the ana-
ing in the National Early Warning System aimed at miti- lysis. In the empirical case, there is no attempt to
gating the effects of natural disasters, including drought incorporate physical modeling of the atmosphere but to
and flooding. But the role of the Agency will be more the empirical stochastic models to the available data.
beneficial if the information it provides is updated regu- The probability estimation for rainfall states from
larly by monitoring, recording and analysis of the accur- available time series helps to obtain predictions for
ate and reliable data and their publications to the rainfall statistical parameters such as the averages, stand-
relevant national bodies. ard deviations and the first order autocorrelation coeffi-
Ghana is located within the tropical region. It is there- cient. The transition probability estimations between the
fore, influenced by weather systems of various scales, states of successive time instances are necessary for
from mesoscale, such as thunderstorms, to large-scale El model construction. Also, theoretical Weibull, Gamma,
Nino Southern Oscillation (ENSO) related phenomena Extreme Value Distribution functions are used most
(e.g. Cai et al. 2015). The major rain-bearing system for often in practice and for predicting the magnitude of
the main rainy season (June to September) is the Inter rainfall (Villarini et al. 2010; Papalexiou et al. 2013). For
Tropical Convergence Zone (ITCZ). On the other hand, accounting dependency in any time series, often a first
the eastward moving mid-latitude troughs will facilitate order Markov Chain is used for modeling. For instance,
the interaction between the mid-latitude cold air and the large variety of weather events modeling and simulation
tropical warm air so that unstable conditions will be cre- (Gringorten, 1996) and long time series of weather data
ated for the moisture that comes into Ghana from the generations (Racsko et al. 1991) were studied through
Atlantic ocean during the small rainy season (February Markov Chain. For rainfall data, many authors have
to May; Climate Change Knowledge Portal). demonstrated that Markov Chain model can be used to
Recent advances in the application of statistical synthesize rainfall time series (Haan et al. 1976; Wilks
methods have dramatically improved the range of tech- and Wilby, 1999; Gregory et al. 1993; Robertson et al.
niques available for analyzing data that are not from nor- 2004; Paulo and Pereira, 2007).
mal distribution. One of these applications, Markov The study on the sequence of daily rainfall occur-
chain analysis, which is used in this study, parallel those rence was initiated by Gabriel and Neumann (1962).
used in the analysis of variance (ANOVA) and regression They demonstrated that the daily rainfall occurrence
for normally distributed data. This development is of for the Tel Aviv, Israel data was successfully fitted
considerable importance, since daily rainfalls are clearly with the first-order Markov chain model. Much later,
not normally distributed (Stern and Coe, 1984). The other studies also reported that the first order of the
amount of rain, intensity, duration and frequency inevit- Markov chain model was found to fit the observed
ably impact on all activities that determine crop produc- data in Italy successfully (Kottegoda et al. 2004). The
tion and also influence the socio-economic texture of model is based on the assumption that there is a de-
the communities and countries world-wide. A mean pendency of the daily rainfall occurrence to that of
temperature change of 1 °C near the earth’s surface leads the previous day. One of the most attractive features
to large changes in rainfall patterns (Hansen et al. 2010). of the Markov chain model involved providing the
With the development of industrialization and the rapid ease in identifying the seasonality in daily rainfall oc-
growth of population, the management of water re- currence and in most cases, the Markov chain of the
sources is becoming more important not only in Ghana first-order model can describe the daily rainfall occur-
but throughout the world. rence (Stern and Coe 1984); however, there are cases
The analysis of precipitation patterns behavior particu- where this model failed to fit the observed data. As
larly in terms of amount of rainfall, frequency, intensity, an alternative, the use of the Markov chain model of
duration and distribution, is beneficial for managing the higher orders often improved these inadequacies
consumption of water. Rain plays a major role in (Wilks, 1999; Hayhoe, 2000).
Tettey et al. Earth Perspectives (2017) 4:6 Page 3 of 11

Aims A ‘discrete-time’ random process means a system

The recent change in climatic patterns is of major con- which is in a certain state at each ‘step’, with the state
cern and has led to disasters in various parts of the changing randomly between steps. The steps are often
world including Ghana. Hence, in this report, we set out thought of as time, but they can equally well refer to
to model rainfall patterns using rainfall data in order to physical distance or any other discrete measurement;
determine the rainy or dry seasons in the selected area formally, the steps are just the integers or natural num-
under investigation. The sub-objectives of the report are bers, and the random process is a mapping of these two
three-fold; 1, to modeling the rainfall pattern in south states, that is, discrete-time with discrete-state-space and
eastern coast of Ghana using Markov chain analysis, 2, discrete-time with a continuous state-space. Since the
to determine the positions of rainy or dry seasons within system changes randomly, it is generally impossible to
the year in five selected meteorological stations and, 3, predict the exact state of the system in the future. How-
to make recommendations for agricultural and commer- ever, the statistical properties of the system’s future can
cial applications in south eastern coast of Ghana. The be predicted. In many applications it is these statistical
information on the wet and dry patterns has vital im- properties that are important. The changes of state of
portance to all allied fields like agriculture, industry and the system are called transitions, and the probabilities
insurance etc. Once the rainfall process is adequately associated with various state-changes are called transi-
and appropriately modeled, the model can then be used tion probabilities. The set of all states and transition
in agricultural planning, aid in predicting periods of probabilities completely characterizes a Markov chain.
drought, soil erosion and flooding. The model can also By convention, we assume all possible states and transi-
be extended to assess the impact of climate change on tions have been included in the definition of the pro-
relatively small geographical areas, rainfall runoff mod- cesses, so there is always a next-state and the process
eling, crop growth studies and other important fields. A goes on forever.
literature search showed that a number of global pre- Markov chain has been widely applied in the disci-
cipitation datasets which also cover Ghana exist. For plines of natural science, engineering, economics and
example, the University of California Irvine Center for management (Liu et al. 2009). The Markov chain
Hydrometeorology and Remote Sensing (CHRS) using modeling approach is useful in understanding the sto-
remotely sensed data (https://www.ncdc.noaa.gov/cdr/ chastic characteristics of droughts and rainfall
atmospheric/precipitation-persiann-cdr). Their data through the analysis of probabilities for each severity
covered the years 1983 to near-present. However, no class, times for reaching the non-drought class from
publication of rainfall patterns in Ghana using Markov any drought severity state, and residence times in
chain analysis has been published from within Ghana. each drought class (Paulo and Pereira, 2007). The
approach can be satisfactorily used as a predictive
Methodology tool for forecasting transitions among drought
Markov Chain Model severity classes up to 3 months ahead (Paulo and
A Markov chain or Markov process, named after Russian Pereira, 2007). An early warning system for drought
mathematician, Andrey Markov (Shannon, 1948), is a management was developed using the Markov chain,
mathematical system that undergoes transitions from one in two climatic areas of Virginia, USA (Lohani and
state to another (from a finite or countable number of Loganathan, 1997; Lohani et al. 1998). Liu et al.
possible states) in a chain like manner. It is a random (2009) demonstrated two advantages of the Markov
process endowed with the Markov property. The Markov chain technique for forecasting drought and rainfall
property states that the conditional probability distribu- conditions. They were: 1, the predictive performance
tion for a system at the next step given its current state increased greatly as the severity of drought increased,
depends only on the current state of the system, and not and 2, the predictive performance was always satisfac-
additionally on the state of the system at previous steps. tory for drought state transitions, and the prediction
Since the system changes randomly, it is generally impos- performance was acceptable for the successive and
sible to predict the exact state of the system in the future. smooth states. The analysis of daily rainfall data
Usually, a Markov chain would be defined for a discrete shows that Markov Chain approach provides one
set of times (i.e. a discrete-time Markov chain) although alternative to modeling future variation in rainfall.
some authors use the same terminology where ‘time’ can These variations may either be in the form of too
take continuous values. The use of the term in Markov much water, which will lead to flooding or too little
chain Monte Carlo methodology covers cases where the water, which will lead to drought. Markov modeling
process is in discrete-time (discrete algorithm steps) with is one of the tools that can be utilized to assist
a continuous state-space. The following concentrates on planners in assessing the rainfall patterns; frequency,
the discrete-time with discrete-state-space case. duration, intensity and distribution.
Tettey et al. Earth Perspectives (2017) 4:6 Page 4 of 11

Markov chain modeling regardless of where it begins. Such π is called the equi-
The Markov chain model is explained as follows: librium distribution of the chain. The long-term transi-
The probability of going from state i to state j in n tion matrix, the value of pn approaches a fixed square
time steps is given as matrix as n increases.
ðn Þ
Pij ¼ PrðX n ¼ jjX 0 ¼ iÞ Applying the Markov Chain Modeling to the rainfall data
A Markov chain is a chain of events for which the prob-
and the single-step transition is
abilities of outcomes or states depend on what has hap-
Pij ¼ PrðX 1 ¼ jjX 0 ¼ iÞ: pened previously, that is, the probabilities of it being
rainy or dry on a particular day depend on whether it
For a time-homogeneous Markov chain: was rainy or dry on the previous day. The assumption is
ðn Þ that, the states ‘rain’ and ‘dry’ do not overlap and cover
Pij ¼ PrðX kþn ¼ jjX k ¼ iÞ all probabilities. A state matrix or state vector is a row
and matrix which shows the probability of each state e.g.
take X = (0.15 0.85) to mean that the probability of it
Pij ¼ PrðX kþ1 ¼ jjX k ¼ iÞ: being rainy on a certain day is 0.15 and probability of it
being dry on that day is therefore, 0.85. The sum of the
The n-step transition probabilities satisfy the
elements of a state matrix is 1.
Chapman-Kolmogorov equation (Papoulis, 1984), that
The historical daily rainfall (time-sequenced) data
for any k such that 0 < k < n,
used in this study were obtained from the Meteorology
ðn Þ
X ðk Þ ðn−k Þ Department, Accra, Ghana, and covered the years
Pij ¼ Pir Prj
r∈S
1980–2010, a period of 30 years, in respect of Cape
Coast, Accra, Akuse, Akatsi and Keta meteorological
Where, S is the state space of the Markov chain. stations located in the south eastern coast of Ghana
The marginal distribution Pr(Xn = x) is the distribution (Fig. 1). The distance from Cape Coast (most westerly)
over states at time n. The initial distribution is Pr(X0 = x). to Keta (most easterly) is approximately 270 km.
The evolution of the process through one time step is The data set was cleaned, that is, missing values were
described by treated using average values of the daily rainfall for the
X particular month for all the years. For instance, if the
Pr ¼ ðX n ¼ jÞ ¼ Prj PrðX n−1 ¼ r Þ
rainfall value for 1 June 2009 was missing, then all the
X ðnÞ r∈S
rainfall values of 1 June of all the other years from 1980
¼ P rj PrðX 0 ¼ r Þ:
to 2010 were computed and the mean used for the value
r∈S
for 1 June.
The (i,j)th element of the matrix product pn − 1 . p = pn , The probability for rain or drought was determined for
which confirms p(n) = pn. The result p(n) = p(0)pn is each town with their transition matrices. The Markov
obtained by noting that chain analysis was carried to find the stable future prob-
X abilities and therefore achieving the steady states.
p ð X n ¼ jÞ ¼ pðX n ¼ jjX 0 ¼ iÞpðX 0 ¼ iÞ The rainfall results (Fig. 2 to e) are the daily mean
i∈S
(and ±standard error) amount of rain (mm) on each day,
From this theory, the n-step transitions probabilities over the 30 year period. The software packages, Microsoft
can be easily obtained by Simple matrix multiplication, Excel 2010 (Microsoft, Redmond, WA, USA) and Matlab
for larger state space efficient of pn are needed. (The MathWorks Inc., Natick, MA, USA) were used to
analyze the data.
Steady-state analysis and limiting distributions
If the Markov chain is a time-homogeneous Markov Results and Discussion
chain, so that the process is described by a single, time- A preliminary look at the rainfall data collected over
independent matrix pij, then the vector is called a sta- the 30-year period (1980–2010) in the five geographical
tionary distribution (or invariant measure) if its entries locations, Cape Coast, Accra, Akuse, Akatsi and Keta in
πj are non-negative and sum to 1 and if it satisfies south eastern coast of Ghana indicates that the rainfall
X pattern is not evenly distributed. It is however, central-
πj ¼ π i P ij : ized from the months of May to September in all the
i∈S
five cities or towns without any certainty. From each
Note that there is no assumption on the starting distri- year, the maximum amount of rainfall was derived on
bution; the chain converges to the stationary distribution daily basis for the five towns, over the 30 year period
Tettey et al. Earth Perspectives (2017) 4:6 Page 5 of 11

Fig. 1 Map of the south eastern coastal area of Ghana, showing the locations of Cape Coast, Accra, Akuse, Akatsi and Keta

and the daily means (and ±standard error) were then The wettest months are May to September, with June
determined. 14th (14.9 mm) to 18th (17.2 mm) recording the high-
est mean daily rainfalls. There are progressively drier
Preliminary Discussions on each town days from August to December, with a few rainy days
A preliminary discussion on patterns of the mean rain- in October.
fall for each location, beginning from Cape Coast, the Akuse experiences dry spells during the months of
most westerly city, proceeding eastward to Accra, Akuse, January and February (Fig. 2c). Akuse has two low peak
Akatsi and then Keta, is described subsequently. The rainy periods; from March, increasing to June and a
towns Akuse and Akatsi are not coastal towns. decrease in late July and early August. A second spell
Cape-Coast is located 130 km west of the capital occurs in September to October. The highest daily mean
Accra. Rainy days are concentrated in the months from amount of rainfall occurs in May (10th, 10.5 mm).
late April to June (Fig. 2). The city experiences a mild November and December recorded slightly more rain
drought from December to March. During April to than January and February.
September and close to November, the mean rainfall Akatsi is located 130 km east of Accra. The rains set
pattern appears to be average. The highest mean daily in during the early part of February, rising to maximum
rainfall over the 30 years occurs in May with a peak on values in May and June, then decreasing in early July
26th May (16.9 mm). The rainfall distribution approaches (Fig. 2d). There is a second spell of a rainy period from
a normal graph. September to early December. The later part of August
For Accra, the periods from January to April are dry experiences a little drought. The rainfall pattern around
with heavy rains during a few specific days in January Akatsi shows two normal distributions; April to June
and March (Fig. 2b). The mean daily rain increases and October to December. The highest daily mean rain-
from March, reaching a maximum mean value in June. fall occurs on 29th May (14.7 mm).
Tettey et al. Earth Perspectives (2017) 4:6 Page 6 of 11

Fig. 2 a-e The daily mean (±standard errors) amount of rain (mm) from January to December for the years 1980 to 2010, a period of 30 years, for the
locations, Cape Coast (Fig. 2), Accra (Fig. 2b), Akuse (Fig. 2c), Akatsi (Fig. 2d) and Keta (Fig. 2e)

Keta is the most south easterly located town. The except for a few days in September and October, Keta
rainfall pattern of Keta is very unstable (Fig. 2e). It rains experiences low rainfall. The highest mean daily rainfall
during the period March to July. January is a period of occurs on May 25th (20.1 mm) and to 7th July.
drought, with little rain or no rain occurring form 1st (18.4 mm). A combined rainfall graph for all five towns
January to 3rd February. From July 14th to December, of the 30 year period is shown in Fig. 3.
Tettey et al. Earth Perspectives (2017) 4:6 Page 7 of 11

Fig. 3 A combined rainfall graph for all five towns over the 30 year period under investigation

Markov Chain Analysis and drought indication is 50%. Looking at data from
Markov Chain analyses for rainfall pattern of Cape-Coast January to December; it is clear that the rainfall in-
The mean daily rainfall pattern was calculated for each creases gradually from May with the probability of 0.372
day in each month. For each day a rainy day was picked until August with the probability of 0.336 and began to
to determine whether the next day is dry or rainy. The decrease from September gradually until December. The
sum of the number of such days (rain or dry) was calcu- maximum rainfall was recorded in the months of May
lated and divided by the total number of days in the to August. In general, Cape-Coast has a greater prob-
month. Similar computations were done for dry days. ability of rain than the other locations.
Given that it is rainy, the probability that it would be
rainy a day later is given by 0.06, and given that it is dry, Markov Chain analysis for rainfall pattern in Accra
the probability that it would be rainy a day later is given For Accra, given that it is rainy, the probability that it
by 0.16. Therefore, would rain a day later is given by 0.10, and given that it
  is dry, the probability that it would be rainy a day later is
0:06 0:94 given by 0.13. Therefore, the transition matrix is,
P¼ ð1Þ
0:16 0:84
 
0:10 0:90
The steady state transition matrix, the value of pn ap- P¼ ð2Þ
0:13 0:87
proaches a fixed square matrix as n increases. This is
represented in the Table 1 (panel, Cape Coast).
Each row (0.145 0.855), from p3, represents the long The steady state transition matrix, the value of pn
term probabilities of each state. Whatever the weather is approaches a fixed square matrix as n increases. This is
on a given day, in the long term the probability that it indicated in Table 1 (panel Accra). The transition matrix
will be rainy on any day is 0.145 and the probability that approaches the steady state matrix (0.126 0.874) regard-
it will be dry on a day is 0.855. The state matrix ap- less of the value of the initial state matrix (Table 1). The
proaches the steady state matrix (0.145 0.855) regardless steady state matrix (0.126 0.874) shows that in the fu-
of the value of the initial state matrix. The steady state ture, the rainfall pattern of Accra will follow the pattern
matrix (0.145 0.855) shows that in the future, the rainfall of the given probability. Thus in future, the probability
pattern of Cape Coast will follow the pattern of the that it will rain on a particular day in Accra is 0.126 and
given probability. Thus in future, the probability that it the probability that it will be dry on a particular in Accra
will rain on a particular day in Cape Coast is 0.145 and is 0.874.
the probability that it will be dry on a particular in Cape The conditional probability of either rainy or drought
Coast is 0.855. These calculations are summarized in in a particular day for the months for Accra shows the
Table 2. The calculations of the steady state matrices for rainfall pattern and it reveals that the rains are averagely
the four other towns are shown in Additional file 1: distributed among the entire zone of Accra from April
(Appendices A to D). to October. The data reveals that the highest rainfall is
The steady state probability rainfall for Cape Coast in- recorded in the months May, June and October with
dicated that, that the likelihood of rainfall is more likely probabilities of 0.467, 0.333 and 0.324, respectively. The
in July with a probability of 0.5 as compared with other rainfall reduces in December considering the probability
months. This indicates that the probability of both rain value of rain in December of 0.031.
Tettey et al. Earth Perspectives (2017) 4:6 Page 8 of 11

 
Table 1 The steady state transition matrix for each town, the value 0:13 0:87
of pn as n increases P¼ ð3Þ
0:06 0:94
Cape Coast
p p2 p3 The steady state transition matrix, the value of pn
0.06 0.94 0.154 0.846 0.1446 0.8554 approaches a fixed square matrix as n increases. This
0.16 0.84 0.144 0.856 0.1456 0.8544 expressed in Table 1 (panel Akuse).
Each row (0.065 0.935, Table 1) represents the long term
p4 p5 p6
probabilities of each state. On any given day, in the long
0.1455 0.8545 0.1454 0.8546 0.1455 0.8545
term the probability that it will be rainy on any day is
0.1455 0.8545 0.1455 0.8545 0.1455 0.8545 0.065 and the probability that it will be dry is 0.935. The
p7
p 8
state matrix approaches a fixed matrix (0.065 0.935), the
0.1455 0.8545 0.1455 0.8545 steady state or stable matrix. The steady state matrix
0.1455 0.8545 0.1455 0.8545 (0.065 0.935) shows that in the future, the rainfall pattern
of Akuse will follow the pattern of the given probability.
Accra
Thus in future, the probability that it will rain on a par-
p p2 p3
ticular day in Akuse is 0.065 and the probability that it will
0.10 0.90 0.127 0.873 0.1262 0.8738 be dry on a particular in Akuse is 0.935.
0.13 0.87 0.126 0.874 0.1262 0.8738 Basically, the rainfall pattern of Akuse may be described
p4 p5 p6 as average. Following the conditional probability of either
0.13 0.87 0.0691 0.9309 0.06484 0.9352 rainy or drought in a particular day for the various months
in Akuse; in the long run, the rainfall is lower in January
0.06 0.94 0.0642 0.9358 0.0645 0.9355
to early April. The rainfall then increases from May to-
Akuse
wards the month October, with a small dry trough in early
p p2 p3 August. The rains began to reduce again drastically from
0.13 0.87 0.0691 0.9309 0.06484 0.9352 November to December. The conditional probability indi-
0.06 0.94 0.0642 0.9358 0.0645 0.9355 cates that the highest rainfall would occur in May with
p4
p 5
p6 probability of 0.382; that is, if it will rain on any particular
day in May.
0.0645 0.9355 0.0645 0.9355 0.0645 0.9355
0.0645 0.9355 0.0645 0.9355 0.0645 0.9355
Markov Chain analyses for rainfall pattern of Akatsi
Akatsi Given that it is rainy, the probability that it would be
p p2 p3 rainy a day later in Akatsi is given by .10, and given that
0.10 0.90 0.064 0.936 0.0626 0.9374 it is dry, the probability that it would be rainy a day later
0.06 0.94 0.0624 0.9376 0.0625 0.9375 is given by .06. Therefore, p is given by,
p4 p5 p6  
0:10 0:90
0.0625 0.9375 0.0625 0.9375 0.0625 0.9375 P¼ ð4Þ
0:06 0:94
0.0625 0.9375 0.0625 0.9375 0.0625 0.9375
Keta The steady state transition matrix, the value of pn
p p2 p3 approaches a fixed square matrix as n increases. This is
expressed in Table 1 (panel Akatsi). The state matrix
0.03 0.97 0.0591 0.9409 0.0582 0.9418
approaches the steady state matrix (0.0625 0.9375)
0.06 0.94 0.0582 0.9418 0.0583 0.9417
regardless of the value of the initial state matrix.
p4
p 5
p6 The steady state matrix (0.0625 0.9375) shows that in
0.0583 0.9417 0.0583 0.9417 the future, the rainfall pattern of Akatsi will follow the
0.0583 0.9417 0.0583 0.9417 pattern of the given probability. Thus, in future, the prob-
ability that it will rain on a particular day in Akatsi is
0.0625 and the probability that it will be dry on a particu-
lar in Akatsi is 0.9375. The observed data revealed that
Markov Chain analyses for rainfall pattern in Akuse the rains were low, below average with risk of in the
For Akuse, given that it is rainy, the probability that it months of January to February. Gradually, the rains in-
would be rainy a day later is given by 0.13, and given crease from March to July with a decrease in the daily
that it is dry, the probability that it would be rainy a day mean amount of rain in the month of August. September
later is given by 0.06. This yields the transition matrix; to October, see the increase in the mean daily rainfall.
Tettey et al. Earth Perspectives (2017) 4:6 Page 9 of 11

Table 2 Rainfall pattern of Cape Coast

Month Initial matrix (p) Long term probability matrix Probability of either
‘rain’ or ‘dry’ on a day
   
January 0:06 0:94 0:145 0:854 0.145 0.854
p8 ¼
0:16 0:84 0:145 0:854
   
February 0:14 0:86 0:075 0:923 0.075 0.923
p6 ¼
0:07 0:93 0:075 0:923
   
March 0:23 0:77 0:072 0:928 0.072 0.928
p9 ¼
0:06 0:94 0:072 0:928
   
April 0:29 0:71 0:280 0:720 0.280 0.720
p6 ¼
0:42 0:58 0:280 0:720
   
May 0:23 0:77 0:372 0:628 0.372 0.628
p ¼
8
0:30 0:70 0:372 0:628
   
June 0:60 0:40 0:403 0:597 0.403 0.597
p ¼
13
0:27 0:43 0:403 0:597
   
July 0:45 0:55 0:50 0:50 0.50 0.50
p7 ¼
0:55 0:45 0:50 0:50
   
August 0:23 0:77 0:336 0:664 0.336 0.664
p8 ¼
0:39 0:61 0:336 0:664
   
September 0:33 0:67 0:130 0:870 0.130 0.870
p ¼
10
0:10 0:90 0:130 0:870
   
October 0:32 0:68 0:382 0:618 0.382 0.610
p ¼
7
0:42 0:58 0:382 0:618
   
November 0:23 0:77 0:182 0:820 0.182 0.820
p6 ¼
0:17 0:83 0:182 0:820
   
December 0:03 0:97 0:118 0:882 0.118 0.882
p7 ¼
0:13 0:87 0:118 0:882

According to the conditional probability of either rainy or The observed data indicates that, the rainfall pattern
drought in a particular day for the various month in of Keta is not favourable; there is a higher probability of
Akatsi, in June, it shows clearly that the probability of drought. Looking at the conditional probability values of
rains occurring has equal chance as that of drought or no either rainy or drought in a particular day for the various
rain (Probability of 50%). On average, Akatsi seems to month in this town, it shows clearly that rainfall may
have a higher probability of it being rainy on any day than occur a little in the month of April to May with even a
the four other towns. downward trend. August and September also have a
little rain where the rest of the month is likely not to
Markov Chain analyses for rainfall pattern of Keta experience any rain.
For Keta, given that it is rainy, the probability that it The Markov chain analysis for the various towns
would be rainy a day later is given by .03, and given that showed that, the probability of rain in each town varies
it is dry, the probability that it would be rainy a day later day by day. Also the long term probability differs from
is given by .06. This is given as, town to town. Observations made on the long term
probability values for each town showed that all the five
  locations tended to be dry. That is, all locations experi-
0:03 0:97
P¼ ð5Þ ence more dry days than rainy days. This is because in
0:06 0:94
each case, the long term probability of rain falling on
any day is generally low (Table 3). We also observed that
The steady state transition matrix, the value of pn from Keta (most easterly), through Accra to Cape Coast
approaches a fixed square matrix as n increases, and it is (most westerly), the maximum rainfall probability gener-
expressed in Table 1 (panel Keta). The state matrix ally increases from east to west.
approaches the steady state matrix (0.058 0.942) and Until recently, rainfall pattern assessment has been a
shows that in the future, the rainfall pattern of Keta will challenging task among rainfall researchers and profes-
follow the pattern of the given probability. Thus in fu- sionals. There are many rainfall tools that have been
ture, the probability that it will rain on a particular day developed around the world and are commonly used to
in Keta is 0.058 and the probability that it will be dry on quantify rainfall conditions. It was found that in most
a particular in Keta is 0.942. cases, Markov chain analysis developed for a specific
Tettey et al. Earth Perspectives (2017) 4:6 Page 10 of 11

Table 3 Maximum rainfall and Maximum probability rainfall

Town Month Maximum Daily Mean Rainfall (mm) Month Maximum
Probability
Cape Coast May 16.9 May & July 0.372 & 0.50
Accra June 17.2 May 0.467
Akuse May 12.5 May 0.382
Akatsi June 14.7 June 0.50
Keta May–July 20.1 & 18.4 May-Sept. 0.215 & 0.277

region, could not be directly applied to other regions their activities accordingly; preparing the land and plant-
due to inherent complexity of rainfall phenomena, different ing of seed crops, and offer advice to their clients in
hydro-climatic conditions and catchment characteristics. terms of investments. The study also demonstrated the
Markov chain analysis has been employed to assess differ- practical application of the Markov chain analysis to
ent climatic regions around the world (Robertson et al. meteorological data, the first of its kind in Ghana.
2004; Paulo and Pereira, 2007). Data bases including the
University of California Irvine (CHRS) indicates a trend to- Additional file
wards drier weather in Ghana (https://www.ncdc.noaa.gov/
cdo-web/datasets). However, no such work has been con- Additional file 1: Supplementary material: Appendix A, B, C, D, E.
(DOCX 135 kb)
ducted with Markov chain to show rainfall patterns and
published from within Ghana.
In this investigation, an employment of Markov chain Acknowledgements
The authors thank the Ministry of Agriculture and Meteorology Department,
analysis for modeling historical daily rainfall within the Ghana for allowing access to the rainfall data.
south eastern coast of Ghana has been done. Daily mean
rainfall values recorded in south eastern coast during Authors’ contributions
MT and FTO developed the idea. Data was collected and analysed by MT,
1980–2010 were used in this study to investigate how FTO, DA and DAA. Manuscript writing was led by DAA and DA. All four
well the Markov chain method was capable of defining authors approved the final manuscript.
rainfall conditions. The study showed that Markov chain
Competing interests
analysis is a good method in predicting patterns of rain- The authors declare that they have no competing interests.
fall and detecting similarities in historical rainfall. The
summary of the highest recorded rainfall from the time
Publisher’s Note
series data and the highest expected rainfall probability Springer Nature remains neutral with regard to jurisdictional claims in published
for the month is indicated in the Table 3. maps and institutional affiliations.

Author details
Conclusions 1
Department of Mathematics, Kwame Nkrumah University of Science and
This report aimed to classify and assessed annual rainfall Technology, Kumasi, Ghana. 2Current address: Department of Mathematics, E. P.
College of Education, PO Box AM 12, Amedzofe, VR, Ghana. 3School of Basic and
over the south eastern coast of Ghana into rainy and
Biomedical Sciences, University of Health and Allied Sciences, PMB, 31 Ho, VR,
drier groups. It was determined that a Markov chain Ghana.
analysis classifies this pattern. The steady state transition
Received: 27 April 2017 Accepted: 28 July 2017
matrices and the steady state probability vectors were
computed for each town and each month over a 30-year
period (1980–2010). It was found that, the rainy or dry References
season patterns observed using the monthly steady state Cai W, Wang G, Santoso A, McPhaden MJ, Wu L, Jin F-F, Timmermann A, Collins M,
Vecchi G, Lengaigne M, England MH, Dommenget D, Takahashi K, Guilyardi E
rainfall vectors tended to coincide with the monthly (2015) Increased frequency of extreme La Niña events under greenhouse
rainfall time series trajectory. That is, the highest mean warming. Nat Clim Chang 5:849–859. doi:10.1038/nclimate2492
daily rainfall occurring from the time series data corre- Climate Change Knowledge portal: http://sdwebx.worldbank.org/climateportal/
index.cfm?page=country_historical_ climat e&ThisCCode=GHA
sponds with the computed maximum probability from Coffie ED (2010) The Rains Are Here Again: Whose Responsibility Is It to Protect Us?
Markov chain analysis. For Accra, for example, the rainy Ghanaweb.com. http://www.ghanaweb.com/GhanaHomePage/features/ The-
season was observed to be in the month of May to June Rains-Are-Here-Again-Whose-Responsibility-Is-It-To-Protect-Us-184733
Dore MHI (2005) Climate change and changes in global precipitation patterns:
and September to October. The probability of rainfall What do we know? Environ Int 31(8):1167–1181
generally tended to increase from east to west along the Environmental Protection Agency, Ghana (2000) Climate Change Impacts, Vulnerability
south eastern coast of Ghana. This work could be help- and Adaptation Assessments in Ghana, 162 pp 4 pdf. 1st National Communication
of Ghana to the United Nation Framework Convention on Climate Change.
ful to agro industries, agricultural insurance practitioners Environmental Protection Agency, UNFCCC, Cambridge. http://unfccc.int/resource/
and other business organizations, to enable them plan docs/natc/ghanc3.pdf
Tettey et al. Earth Perspectives (2017) 4:6 Page 11 of 11

Fletcher CG, Matthews L, Andrey J, Saunders A (2015) Projected Changes in Mid- Wilks DS (1999) Inter annual variability and extreme-value characteristics of several
Twenty-First-Century Extreme Maximum Pavement Temperature in Canada. J Appl stochastic daily precipitation models. Agric For Meteorol 93(3):153–169
Meteor Clim 55:961–974 Wilks DS, Wilby RL (1999) The weather generation game: a review of stochastic
Gabriel KR, Neumann J (1962) A Markov chain model for daily rainfall occurrence weather models. Prog Phys Geogr 23(3):329–357
at Tel Aviv. Q J R Meteorol Soc 88:90–95 von Storch H, Zwiers FW (1999) Statistical Analysis in Climatology. Cambridge
Gregory JM, Wigley TML, Jones PD (1993) Application of Markov models to area- University Press, Cambridge, pp 197–213 https://pdfs.semanticscholar.org
average daily precipitation series and inter-annual variability in seasonal totals. /d457/3b4307d80e5b9f1845770c226759fb331023.pdf
Clim Dyn 8(6):299–310
Gringorten II (1996) A stochastic model of the frequency and duration of weather
events. J Clim Appl Meteorol 5:606–624
Groisman PY, Knight RW, Easterling DR, Karl TR, Hegerl GC, Razuvaev VN (2005)
Trends in intense precipitation in the climate record. J Clim 18:1326–1350
Haan CT, Allen DM, Street JO (1976) A Markov chain model of daily rainfall. Water
Resour Res 12(3):443–449
Hansen J, Ruedy R, Sato M, Lo K (2010) Global Surface Temperature Change. Rev
Geophys 48(4) RG4004)):1–29
Hayhoe HN (2000) Improvements of stochastic weather data generators for diverse
climates. Clim Res 14:75–87
IPCC 2014. Settele J, Scholes R, Betts R, Bunn S, Leadley P, Nepstad D, Overpeck TK,
Taboada MA. Terrestrial and Inland Water Systems. In: Climate Change 2014:
Impacts, Adaptation and Vulnerability. Part A: Global and Sectoral Aspects.
Contribution of Working Group II to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. https://www.ipcc.ch/pdf/
assessment-report/ar5/wg2/WGIIAR5-Chap4_FINAL.pdf
Kharin VV, Zwiers FW, Zhang X (2005) Inter-comparison of near-surface temperature
and precipitation extremes in AMIP-2 simulations, re-analyses, and observations.
J Clim 18:5201–5223. doi:10.1175/JCLI3597
Kottegoda NT, Natale L, Raiteri E (2004) Some considerations of periodicity and
persistence in daily rainfalls. J Hydrol 296:23–37
Liu X, Ren L, Yuan F, Yang B, Nanjing PR (2009) Meteorological drought Forecasting
using Markov Chain model. In: 11th International Conference on Environmental
Science and Information Application Technology, Chania, Crete, Greece, pp 23–26
http://toc.proceedings.com/18229webtoc.pdf
Lohani VK, Loganathan GV (1997) An early warning system for drought management
using the Palmer drought index. J Am Water Resour Assoc 33:1375–1386
Lohani VK, Loganathan GV, Mostaghimi S (1998) Long-term analysis and
short-term forecasting of dry spells by Palmer Drought Severity Index.
Nord Hydrol 29(1):21–40
Meteorological Services Agency of Ghana. http://www.meteo.gov.gh/website/#
Nelson W, Agbey SND (2005) Climate change and poverty linkages. In: Allotey J,
Mensah L (eds) Ghana climate change impacts, vulnerability and adaptation
assessments, under the Netherlands climate assistance programme. Accra,
Ghana: Environmental Protection Agency, Chapter 5, pp 169–206
Papalexiou SM, Koutsoyiannis D, Makropoulos C (2013) How extreme is extreme? An
assessment of daily rainfall distribution tails. Hydrol Earth Syst Sci 17(2):851–862
Papoulis A (1984) Probability, Random Variables, and Stochastic Processes, 2nd edn.
McGraw-Hill, New York
Paulo AA, Pereira LS (2007) Prediction of SPI drought class transitions using Markov
chains. Water Resour Manag 21(10):1813–1827
Racsko PL, Szeidl L, Semenov M (1991) A series approach to local stochastic weather
models. Ecol Model 57:27–41
Raiford JP, Aziz NM, Khan AA, Powell DN (2007) Rainfall Depth- Duration- Frequency
Relationships for South Carolina, North Carolina, and Georgia. Am J Environ Sci
3(2):78–84
Robertson AW, Kirshner S, Smyth P (2004) Downscaling of daily rainfall occurrence
over northeast Brazil using a hidden Markov model. J Clim 17(22):4407–4424
Shannon CE (July–October 1948). A Mathematical Theory of Communication
(PDF). Bell Syst Tech J 27 (3): 379–423. doi:10.1002/j.1538-7305.1948.tb01338.x
Srikanthan R, McMahon TA (2005) Automatic Evaluation of Stochastically Generated
Rainfall Data, Conference Paper. Aust J Water Resour 8(2):195–201
Stern RD, Coe R (1984) A model fitting analysis of daily rainfall data. J R Stat Soc A
147(1):1–34
University of California Irvine Center for Hydrometeorology and Remote Sensing
(CHRS) (https://www.ncdc.noaa.gov/cdr/atmospheric/precipitation-persiann-cdr)
Villarini G, Smith JA, Napolitano F (2010) Nonstationary modeling of a long record of
rainfall and temperature over Rome. Adv Water Resour 33(10):1256–1267
Westral S, Alexander LV, Zwiers FW (2013). Global Increasing Trends in Annual
Maximum Daily Precipitation. J Climate, 26: 3904–3918. American
Meteorological Society http://dx.doi.org/10.1175/JCLI-D-12-00502.1