Analysis of Rainfall Trends and Its Relationship with SST Signals in the Lake Tana Basin , Ethiopia

,e impacts of climate change and climate variability on human life have led the scientific community to monitor the behavior of weather and climate variables at different spatial and temporal scales.,is paper explores seasonal and annual trends of rainfall in the Lake Tana basin (LTB) and their teleconnections with global sea surface temperatures (SSTs) over the period between 1979 and 2015. ,e nonparametric Mann–Kendall test and Sen’s slope estimate are applied to the rainfall data collected from the National Meteorology Agency (NMA) of Ethiopia for detecting and estimating rainfall trends. Additionally, Pearson’s correlation coefficient method is used to determine the effect of SST variations on rainfall. ,e assessment of rainfall trends indicates that the amount of annual rainfall in the Lake Tana basin is increasing, but the rate of increase is not statistically significant. Seasonal analysis reveals that the smallest amount of rainfall occurs in the Bega season, and this season is getting drier with time. However, the analysis indicates that the other two seasons (Belg and Kiremt) are becoming wetter.,e rainfall in Kiremt is increasing significantly (significant at the p � 0.05 level) in Debre Tabor station with a rate of 10.20mm/year. Besides, 78.1% of the total annual rainfall in the basin occurs during this rainy (Kiremt) season, whereas Bega and Belg contribute some 9.4% and 12.5%, respectively. Furthermore, the correlation analysis of rainfall and SSTs indicates that rainfall of the LTB is highly affected by the variations of SSTs.


Introduction
e impacts of climate change and climate variability on human life have led the scientific community to monitor the behavior of weather and climate variables.Rainfall as one of the most important of these variables has a direct and indirect impact on the natural environment and human life.A large spatial and temporal variability of rainfall leads to an increased incidence of extreme events such as floods and droughts.It is also recognized that rainfall is one of the key climatic variables that affect both the spatial and temporal patterns on water availability [1].Moreover, erratic rainfall can trigger various disasters, for example, floods, landslides, water logging, erosion, and salinity intrusion [2].Every minor change in the rainfall intensity or amount imposes a severe challenge on the rural people since their main livelihood depends on agriculture which mostly relies on a short rainy season.
Various studies have indicated changes in the spatial and temporal variability and trends of rainfall pattern due to climate change and climate variability at different spatial (e.g., regional and national) and temporal (e.g., annual, seasonal, and monthly) scales in Ethiopia.For example, a study by Worku [3] indicates that flooding is a familiar event in the upper Blue Nile basin and did cause a lot of destruction in the past years.Bewket and Conway [4] have discussed about the decline of annual rainfall in the northwestern parts of Ethiopia, while there was no clear trend of annual rainfall in other parts during their observation time.Osman and Sauerborn [5] also determined that summer rainfall, locally called Kiremt rainfall, in the central highlands of Ethiopia declined in the second half of the 20th century.erefore, monitoring and quantifying the background trends of rainfall in the country as a whole and its sections is of vital importance.
Hydrologic time series almost always exhibit seasonality due to the periodicity nature of the weather.In Ethiopia, this arises greatly from seasonal variations in the precipitation volume, as well as in the rate of evapotranspiration. is variation can also be associated with the sea surface temperature (SST) dynamics.Previous studies mention that correct representation of ocean-atmosphere interaction is important for simulating certain aspects of climate such as the organization and propagation of intraseasonal variability [6,7] and precipitation variability and its relationship with underlying sea surface temperatures [8][9][10].Trends of surface temperature and rainfall over Ethiopia may be affected by global ocean-atmosphere coupling [11].However, the analysis of rainfall trends is important in studying the impacts of climate change and variability on water resource planning and management [12]; it has been recognized that global or continental scale observations of historical climate are less useful for local or regional scale planning [13][14][15].Despite many papers have been documented on rainfall trends and variability, their analysis is limited to the main rainy season (June to September) and large spatial scales in Ethiopia; there are also very limited works done on the other seasons and around the LTB, and most researches lack to relate the variability or trends to large-scale drivers like SST. e aim of this study is therefore to analyze the spatial and temporal trends of rainfall in the LTB and their association with the global SST variations.e study uses the Mann-Kendall trend analysis, Sen's slope estimator, and correlation techniques.Although the Mann-Kendall test is suitable for cases where the trend may be assumed to be monotonic and thus no seasonal or other cycle is present in the data, the Sen's method uses a linear model to estimate the slope (magnitude) of the trend and the correlation coe cient is used to evaluate the relationship between rainfall and SST variations.

e Study Area.
e LTB (Figure 1) is geographically located in northwestern Ethiopia at latitude 10.95 °and 12.78 °N and longitude 36.89°and 38.25 °E, with a drainage area of about 15,000 km 2 [16].Lake Tana, the largest lake in Ethiopia and the third largest in the Nile basin, is located in this basin.is lake is the largest freshwater and oligotrophic high-altitude lake in the world [17].It is shared by the four administrative zones called Awi, North Gondar, South Gondar, and West Gojjam.e climate of the area is largely controlled by the movement of the intertropical convergence zone (ITCZ) and tropical highland monsoon, which results in a single rainy Kiremt season between June and September.Rainfall (70-90% of total annual rainfall) occurs in this season.e mean annual rainfall is about 1358.42 mm, and the mean annual temperature is about 21 °C.

Data.
is study uses rainfall data of ve meteorological stations on a monthly basis for the 1979-2015 timescale, collected from the National Meteorology Agency of Ethiopia (NMA), Bahir Dar branch o ce.Speci c locations and the period of the rainfall data record of the meteorological   Advances in Meteorology stations are described in Figure 1 and Table 1, respectively.And the mean monthly rainfall of each station is illustrated in Figure 2. Station selection has been done based on quality, long-range data, and representation of various climatic zones in the study area.Additional criteria are, as suggested by Alexandersson in 1986 [18], the study of historical climate variability, and the change should utilize reliable data that are free of arti cial trends or changes.Artifacts of the measurement caused by changes in the observation practice, equipment, site exposure, and location can led to misleading results when used in trend analyses [19].Hence, the homogeneity test of the time-series rainfall data following some of the most reliable procedures recently proposed in the technical and scienti c literature [20] has been applied in this study.e homogeneity test is an important issue to detect the variability of the data with time.We also utilize the SST data from the four observations of the National Oceanic and Atmospheric Administration (NOAA) COBE-SST version to relate the variations in the trend of rainfall with the global seasonal and annual SST dynamics [21].Currently, the Earth System Research Laboratory, Physical Sciences Division (PSD), makes available these reanalysis data sets to the public in the standard netCDF le format at the following website: http://www.esrl.noaa.gov/psd/.
In Ethiopia, there are three seasons with four months each, classi ed based on climatological means of rainfall and temperature.ese seasons are locally known as Bega (October, November, December, and January), Belg (February, March, April, and May), and Kiremt (June, July, August, and September) [22,23].In this study, we divided the time series of monthly rainfall data into three seasons and annual time-series categories as suggested by [22,23].

Descriptive Statistics.
In this study, we use these descriptive statistics to display information about the distribution of the rainfall data.
ese approaches are more precise and objective.
e statistical descriptions of the annual rainfall of each meteorological station and seasonal rainfall of the Lake Tana basin are stated in Tables 2 and 3, respectively.

Mann-Kendall Trend Test.
e Mann-Kendall (MK) test [24,25], a nonparametric test method, has been widely used to detect whether trends exist in meteorological or hydrologic time series [26].e Mann-Kendall trend test is also less sensitive to outliers, and it is the most robust and suitable for detecting trends in rainfall [27].
e nonparametric model Mann-Kendall test has been applied to the seasonal and the annual rainfall series to investigate the rainfall trends.For making statistical decision, the test statistics are evaluated at the 5% (p 0.05) level of signi cance.
e Mann-Kendall test statistic is denoted by S and is computed using each pair of the observed values x i and x j of the random variable under consideration.Each pair is then inspected to nd out whether x i > x j or x i < x j .e Mann-Kendall statistic S of the series x is given by where sgn is the signum function and x i and x j are the annual values in the years i and j, i > j, respectively, and

Advances in Meteorology
Under the null hypothesis of no trend, the statistic S in (1) follows approximately normal distribution with mean zero and variance where m is the number of tied groups and t k is the number of data points in the group k.
When the sample size n ≥ 10, as used in this study, the test statistic Z is calculated from [25] e null hypothesis that there is no trend is rejected when the computed Z value is greater than Z α/2 in absolute value, where Z α/2 is the smallest Z which has the probability less than α/2 to appear in case of no trend.Accordingly, a positive value of Z indicates an upward trend and a negative value of Z indicates a downward trend.Finally, we explored the trend through time-series tables and plots of the rainfall data.

Sen's Slope
Estimator.Sen's slope estimation [28] is another nonparametric method for trend analysis of the precipitation data set.It is used to detect the magnitude of the trend.e Sen's slope estimation is considered more robust than the least-squares method due to its relative insensitivity to extreme values and better performance even for normally distributed data [29].In general, the slope Q between any two values of a time series x can be estimated from where x j and x k are the data values for j and k times of a period (j > k). e slope is estimated for each observation.For a time series x having n observations, there are a possible N � n(n − 1)/2 values of T i that can be calculated.
According to Sen's method, the overall estimator of the slope is the median of these N values of T i .e overall slope estimator Q i is thus for N odd observations, When significant trends in the data were detected, 95% confidence intervals were calculated using nonparametric techniques as described by Salmi et al. [30].e positive or negative slope Q i is obtained as an upward (increasing) or downward (decreasing) trend.
e Mann-Kendall and Sen's tests were done by using MAKESENS, a Microsoft Excel template, which was developed by the Finnish Meteorological Department for detecting and estimating trends in the time series of the annual values of atmospheric and precipitation concentrations.MAKESENS is a widely used software for detecting and researching rainfall trends [31][32][33][34].Detail information about MAKESENS can be obtained from the article written by Määttä et al. [35], and the website of the Finnish Meteorological Institute can be used to get more information on the MAKESENS application for trend calculation.

Correlation Coefficient.
In this study, we use Pearson's correlation coefficient (r) as defined by [36], with the significance assessed using the rainfall pattern of the LTB and the likelihood global sea surface temperature variations.A two-tailed, 5-percent significance level is adopted unless indicated otherwise.It should be mentioned that this significance test assumes normally distributed populations for both variables.e Pearson correlation coefficient r in this research is defined as follows: Suppose that there are two variables X and Y, each having n values X 1 , X 2 , . . ., X n and Y 1 , Y 2 , . . ., Y n , respectively.Let the mean of X be X and the mean of Y be Y. en, Pearson's r is given by where the summation proceeds across all n possible values of X and Y in this sample.By design, r is described as −1 ≤ r ≤ 1.A correlation coefficient of approximately −1 indicates that as values of one variable increase, there is a perfectly predictable decrease in values of the other variables (in our case, SST and rainfall), and a correlation coefficient of approximately 1 indicates  e lowest amount of rainfall in this season occurs in Bahir Dar, and the highest amount of rainfall occurs in Dangila.After Belg, the rainfall increases sharply.

Results and Discussion
e highest amount of rainfall occurs during Kiremt, and the amount of rainfall varies from 800 to 1200 mm.During the season, the maximum amount of rainfall occurs in Dangila (mean rainfall 1260.7 mm) where its domination extends to the other seasons, and Bahir Dar and Debre Tabor receive a little more than 1100 mm of rainfall.Adet and Gondar receive the lowest amount of rainfall which is around 800 mm.After Kiremt, rainfall decreases sharply to its lowest amount.
e trend statistics for the seasonal trend of rainfall (Table 4) show that there is an increasing trend during Kiremt in all meteorological stations except Adet.Kiremt rainfall is increasing in Bahir Dar, Dangila, Debre Tabor, and Gondar, and the rate (Sen's slope) of increase is 2.14, 1.53, 10.20, and 3.18 mm, respectively.e rate of decrease in Kiremt rainfall in Adet is 4.34 mm. e increasing trend of Kiremt rainfall is statistically signi cant in Debra Tabor (bold values in Table 4).During Belg, rainfall shows an increasing trend in Dangila and Debre Tabor and a decreasing trend in Adet, Bahir Dar, and Gondar.On the contrary, Bega rainfall decreases in all stations except in Dangila which shows an increasing trend.e decreasing trend of Bega rainfall is statistically insigni cant in Adet, Bahir Dar, Dangila, and Gondar; the rate of decrease is 0.43, 0.43, 0.08, and 0.20 mm, respectively.Generally, it can be noted that the trends of season rainfall for all stations during the study period  have been insigni cant except for Kiremt in Debre Tabor.

Annual Rainfall.
Trend analysis was also performed on an annual scale to examine whether there are trends in the data at this scale.e annual rainfall of the meteorological stations in the LTB varies by the location of the stations.Tables 2 and 5 show descriptive statistics and trend results of the annual rainfall for each station in the LTB, respectively.
e annual rainfall during 1979-2015 varies from 843.8 to 2008.60 mm recorded at Gondar and Dangila, respectively (Table 2).Dangila which is situated in the southeastern part of the basin has received around 21% higher annual rainfall than Gondar.During the analysis of the descriptive statistics, the standard deviation of the annual rainfall in all stations shows high values than the average values.is relation between the standard deviation and the average values indicates that the deviation from the normal distribution is considerable.is fact is especially

Advances in Meteorology
supported by the coe cients of variation of the annual rainfall which are over one or close to one.ese show that, in mountainous areas, the rainfall values are much di erent from the average.Furthermore, it can be noted that the fact that the CV values (Table 2) are so high is an indication of the rainfall being highly variable and not dependable or reliable or can be attributed to the length of the data set being used or the quality of data.On the contrary, the lower the coe cient of variation of the rainfall amount in any year, the lower the variability and the greater the dependability.Coe cient of variation values less than 0.5 show lower variability from the mean.e amount of annual rainfall shows high variability from station to station and time to time.e results of the Mann-Kendall trend test show an increasing trend in annual rainfall in four (Bahir Dar, Dangila, Debre Tabor, and Gondar) out of the ve meteorological stations and a decreasing trend of annual rainfall in Adet.
e annual rainfall in Debre Tabor is increasing signi cantly.And Sen's slope estimate rate of increase in annual rainfall in Bahir Dar, Dangila, Debre Tabor, and Gondar is 2.20, 3.42, 6.58, and 2.88 mm per year, respectively, whereas the rate of decrease in annual rainfall in Adet is −4.48 mm/year (Table 5).In general, rainfall in Adet shows distinct characteristics than that in the other stations.
is could be associated with the geographical location of Adet, which is somewhat away from the other stations in the LTB (Figure 1).e increase in annual rainfall of the four stations is due to the rainfall received in Kiremt (main rainy seasons).

Regional Trend Analysis.
In addition to the trend analysis of rainfall at the station level, we examined the regional-level trends by averaging the rainfall of all stations in the study area.Although the climate change a ects resources in a cumulative manner, it is useful to determine whether certain seasons are more susceptible.Table 3 presents the descriptive statistics of the seasonal rainfall in the Lake Tana basin for the period of 1979-2015 and

6
Advances in Meteorology seasonal contribution to the annual rainfall.
e mean seasonal rainfall is 124.18 mm, 164.45 mm, and 1063.57mm for Bega, Belg, and Kiremt, respectively.Similarly, the mean annual rainfall is 1352.20.Inspection of seasonal rainfall shows that 78.1% of the total annual rainfall in the basin is received during the Kiremt season.e Belg season receives the second highest amount of rainfall next to Kiremt.12.5% of the total annual rainfall occurs during the Belg season.e lowest amount of rainfall occurs during the Bega season, that is, 9.4% of the total annual rainfall.
Previous studies con rm the similar statistical values of the seasonal rainfall around the LTB.For example, Setegn et al. [37] have indicated that 70-90% of total rainfall in the LTB occurs during the Kiremt season between June and September.
e availability of high rainfall during the Kiremt season is part of a larger East African monsoon season spurred on by the shifting of the intertropical convergence zone (ITCZ) northward [38,39].Consequently, Bega is the driest season in the LTB and Kiremt is the wettest.
e standard deviation and coe cient of variance are high in the seasonal rainfall relative to the average values (Table 3).e coe cient of variance for Bega and Belg is 44%, indicating that the Bega and Belg rainfall in the LTB is highly variable.However, the Kiremt and annual rainfall coe cient of variation values are 11% and 9%, indicating less variability.Cherkos [40] summarizes that the coe cient of variation values <20% indicate less variability, between 20% and 30% indicate moderate variability, and >30% indicate high variability.e regional rainfall trend (Figure 4) reveals an increasing trend in both seasonal and annual timescales.But the rate of increase is still insigni cant as it is indicated by Sen's slopes.e rate of increase for Bega, Belg, and Kiremt rainfall was 0.055, 0.39, and 0.56 mm.Similarly, annual rainfall was increased by 0.36 mm during 1979-2015.Rainfalls in the study period have been rather highly variable (Figure 5).Previous studies near the Lake Tana basin and other parts of Ethiopia have also found insigni cant trends at di erent spatial and temporal scales.For example, from the trend analysis of 53 years' daily precipitation data in Debre Markos, Shang et al. [41] found that there is no increasing trend in the extreme precipitation in Debre Markos.Similarly, Wing et al. [42] studied the trends and spatial distribution of the annual and seasonal rainfall in di erent parts of Ethiopia using data from 134 stations of 13 watersheds during 1960 to 2002 and showed no signi cant changes in annual watershed rainfall for any of the watersheds examined, except that a signi cant decline in Kiremt rainfall was recorded in watersheds located in the southwestern and central parts of Ethiopia.A recent study of the trend by Addisu et al. [43] at the signi cance level of α 0.05 using NCEP data around the Tana basin found statistically signi cant annual rainfall irregularities.Although the Kiremt rainfall shows an increasing rate of 0.56, still it is not signi cant unless it is in uenced by the signi cant trend of seasonal and annual rainfall in Debre Tabor station.
Figure 5 illustrates tting of the linear regression trend and 5-year moving average line to the seasonal and annual   5).On the contrary, the linear regression trend con rms that there was no signi cance in the seasonal and annual rainfall trends during 1979-2015.It is worth witnessing that the Mann-Kendall trend test for the entire period was not statically signi cant.

Rainfall-SST Relationship.
Previous studies have demonstrated that signi cant relationships exist on seasonal and intraseasonal timescales between precipitation and SST, shortwave ux, latent heat ux, and zonal momentum ux [8].In this study, we focus on the relationship between seasonal and annual rainfall and SST by calculating correlations coe cients.Accordingly, the correlation coe cients show a positive relationship of rainfall with the SST variations particularly in the tropical region (Figure 6).Some parts of eastern Indian Ocean and Paci c Ocean also show negative correlation coe cients.is is associated with the variability in opposite modes of temperatures in the oceans such as El Niño-Southern Oscillation (ENSO) of the Paci c Ocean and Indian Ocean Dipole (IOD) of the Indian Ocean.
ese opposite SST anomalies provide varying rainfall as high and low to Ethiopia and the LTB in particular.A study in the upper Blue Nile basin by Seleshi and Camberlin [44] revealed that boreal summer rains correlation with ENSO has shown that warm ENSO periods (El Niño years) are typically associated with lower precipitation and drought years, while cold periods (La Niña years) are associated with higher precipitation quantities.e average uctuation of the oceanic SST particularly the ENSO is between 3 and 7 years; the combined linear regression and 5-year moving  average plot (Figure 5) confirms the fluctuations of rainfall between this period.Furthermore, the year 2015 (which was El Niño year) was considered as one of the warmest years in record and a dry year to Ethiopia; this is due to the failure of the Kiremt and annual rainfalls associated with the positive ENSO (El Niño) events.

Conclusions
e present study assessed the seasonal and annual trends of rainfall in the Lake Tana basin and their relationship with SST anomalies using time-series data from 1979 to 2015.Understanding of trends of rainfall would provide useful information for the planning, development, and management of water in any area or region.e hydrological response of any region depends on several climatic variables, particularly rainfall.
e seasonal and annual trend of rainfall was analyzed at station and basin levels.e Mann-Kendall trend analysis confirms that there was no significant trend observed in all seasons.In addition, the linear regression trend analysis confirms the insignificant trend analyzed by the Mann-Kendall test.However, the seasonal and annual deviation and 5-year moving plots confirm that rainfall in the LTB was highly variable during the entire study period of 1979-2015.Generally, rainfall in Bega is decreasing in four out of five stations.is season receives the lowest amount of rainfall, that is, 128 mm per year.Hence, Bega is becoming drier.Although the annual rainfall in the Lake Tana basin is increasing, the rate of increase expressed by Sen's slope is still not statistically significant.
e seasonal contribution to the total annual rainfall was 78.1% during the rainy (Kiremt) season and 12.5% and 9.4% during Belg and Bega seasons, respectively.Furthermore, the rainfall has been related to the SST drivers to examine whether the strong seasonal variability and annual variability are associated.From the rainfall-SST relationships, we conclude that the rainfall of the Lake Tana basin is affected by the SST variations.
Further studies are recommended to be conducted to investigate other rainfall characteristics at high (daily) temporal resolution such as extreme rainfall, dry days, rain days, and other climate change parameters for this region to confirm whether a significant trend exists, performing the proposed homogenization procedures and data quality analysis.And also local factors like topography and water bodies must be considered in addition to the global factors like SST.

Figure 1 :
Figure 1: Study area and location of the meteorological stations (red points) used in the study.

Figure 2 :
Figure 2: Mean monthly amount of rainfall for the selected stations averaged over the period 1979-2015.

3. 1 .
Station Basis Trend Analysis 3.1.1.Seasonal Rainfall.e mean seasonal amounts of rainfall for the ve stations in the LTB have been analyzed and summarized in Figure 3. e lowest amount of rainfall occurs during the Bega season, and the highest amount of rainfall occurs during the Kiremt season for all stations.e mean seasonal rainfall during Bega varies from 100.7 to 156.8 mm.Rainfall gradually increases from the Bega season to the Belg season every year.During the Belg season, the mean rainfall varies from 115.8 to 215.5 mm.

Figure 3 :
Figure 3: Mean seasonal amount of rainfall for each station from 1979 to 2015.

Figure 5 :
Figure 5: Mean seasonal and annual rainfall and equivalent linear regression trend line and 5-year moving average line plotted for the di erent aggregated station data sets.(a) Bega.(b) Belg.(c) Kiremt.(d) Annual.

Figure 6 :
Figure 6: Rainfall and SST (averaged over 50 S-50 N, 200-200 E) correlations: (a) correlation coe cient of the Bega (ONDJ) rainfall trend and SST raw; (b) correlation coe cient of the Belg (FMAM) rainfall trend and SST raw; (c) correlation coe cient of the Kiremt (JJAS) rainfall trend and SST raw; (d) correlation coe cient of the annual rainfall trend and SST raw.

Table 1 :
Geographical description of the meteorological stations and rainfall data record period.

Table 2 :
Descriptive statistics of the annual rainfall for each station during 1979-2015.
SD standard deviation; CV coe cient of variance; SE standard error.

Table 3 :
Descriptive statistics of the seasonal and annual rainfall in the LTB.

Table 4 :
Seasonal rainfall trend and Sen's slope estimate for each station in the LTB, at a signi cance level of α 0.05.

Table 5 :
Mann-Kendall and Sen's test results of annual rainfall trends in the ve stations at a signi cance level of α 0.05.