Variability and Trend Analysis of Temperature, Rainfall and Characteristics of Crop Growth Season in Eastern Zone of Tigray Region, Northern Ethiopia

To favour farmers and adjusting their farming practices, long term weather analyses is essential to determine future directions and making adjustments required to existing systems. The main purpose of this study was thus to analyze the variability and trends of climatic variables (temperature and rainfall) and characteristics of crop growth season in Eastern zone of Tigray region for the period of 1980 – 2009. Detail investigations were carried out using parametric (Linear regression) and non- parametric tests (Mankendall and Sen’s slope estimator). Moreo ver, homogeneity test was applied using a method developed by Van Belle and Hughes for the general trend analysis. Furthermore, the trend of rainfall end to characterize crop growth season using R-Instat and XLSTAT software. It was found that the general trend of monthly rainfall experienced an overall significant increasing trend. The seasonal rainfall experienced significantly increasing trend during the summer rainy season (June – September) whilst a significant decreasing trend occurred in the short rainy season (February – May). Likewise, the seasonal maximum temperature trends exhibited a significant increasing trend in all seasons whereas the minimum temperature showed inhomogeneous trend across seasons as well as stations. Despite significant increase of rainfall in summer season, the trend of growing season characteristics (onset, cessation, length of growing period and dry spell length) did not change significantly over the


Introduction
Trends of temperature and precipitation, the most important parameters for crop production, have been changed over time in Ethiopia. For the past four decades, Ethiopia's annual average temperature has been increased by 0.37 0 C per decade since 1951 (NMA 2007;Aragie 2013). Moreover, Ethiopia has experienced a high degree of interannual rainfall variability (Cheung et al 2008;Gebremicael et al. 2017;Mekasha et al. 2014;Meze-Hausken 2004;Seleshi and Zanke 2004). The changes in temperature and rainfall patterns as a result of climate variability and change are very critical to most of the population of the country who are dependent on rain-fed agriculture for their livelihoods.
The impacts of increased temperature and changes in rainfall patterns are expected to reduce crop production and water availability for irrigation and other farming uses which is more pronounced in the north, northeast and eastern lowlands of the country (Aragie 2013).
Most of the droughts caused by climate variability and change have been occurred in the eastern and southern part of Tigray region (Gebrehiwot and van der Veen 2013). The climate of the study area is characterized by frequent droughts. Almost every year, localized droughts associated with variable and erratic rainfall has been the main reasons for crop productivity failures and hence jeopardizing livelihoods and development activities of the region (Awulachew et al. 2011;Gebrehiwot and Van der Veen 2013). Indeed, moisture stress stands to be the major limiting factor for crop and animal production in the eastern zone of the Tigray region (Meles et al. 1997). As a result, crop yield has been severely affected during part of its growing period .
Although rainfall variability and associated localized droughts have been the greatest concern in the study area, few attempts have been made so far to quantify the spatio-temporal characteristics of precipitation and temperature. Yet, the emphasis of most of the studies related to the trend analysis so far carried out in Tigray region has been limited to rainfall analysis (Abraha 2015; Cheung et al 2008;Gebrehiwot et al 2011;Gebrehiwot and van der Veen 2013;Hayelom et al 2017;Gebremicael et al. 2017;Mekasha et al. 2014;Meze-Hausken 2004;Seleshi and Zanke 2004).
Whereas temperature analysis has been ignored in many of these studies, although it is also vital for crop production and water-related issues.
In addition, the rainfall trend analyses made by many of the studies listed above are based on few station data and/or with few number of years especially regarding the study area (Eastern zone of Tigray) and many of the studies were restricted to even trends of annual or monthly or seasonal total values. Rainfall variability based on agricultural practices such as onset and cessation at an interval of days, length of growing period (LGP), and wet and dry spells weren't included in those studies with the only exception of  who determined LGP of two crops, Teff (Eragrostis tef) and barley (Hordeum vulgare) in Giba catchment of the Tigray region. Yet, the rainfall and rainy season characteristics are important to make proper crop-based decisions in seeding, fertilizing, selecting crop variety, selecting suitable cropping pattern, and selecting the best agro-techniques. Assessing long-term trends of rainfall and rainy season characteristics (including onset, cessation and LGP) can help to formulate farming strategies to efficiently use the available water (Fiwa et al. 2014). Rainfall statistics for dry spells are also very important for the planning and management of water resources (Almazroui et al. 2017).
Overall, assessing spatio-temporal of changes in temperature & rainfall, and rainfall characteristics was the main purpose of this research. Findings of this study will help to better understand the uncertainties associated with rainfall and temperature patterns and will favour knowledge-based management of agriculture, irrigation and other water related sectors in the region.
Three watersheds, namely Agulae, Suluh and Genfel were selected from the zone for this research purpose based on population growth, expansion of urbanization, and availability of small-scale irrigation schemes in each watershed ( Figure 1). The livelihood of the community is mainly dependent on rainfed agriculture. Common rainfed crops in the watersheds include teff, wheat, barley, maize, sorghum and pulses. However, irrigation agriculture has increased significantly at household level in the recent years (Nyssen et al. 2010). According to Gebreyohannes et al. (2013), the dominant soil texture classes in the area is clay loam (40%) followed by sandy clay loam (30%), clay (19%), loam (10%), and sandy loam (1%). recently implemented at Ethiopian National Meteorological Agency (NMA).The quality of provided climate data by ENACTS is improved by combining careful quality control of data from weather stations with that of satellite estimates. The combined data set is generated at a 10-daily time scale and for every 4 km grid across Ethiopia. This is the best available dataset for the country which is homogeneous and recommended for climate analysis (Dinku et al. 2014). The detailed information about ENACTS is elucidated in Dinku et al. (2014) and Dinku et al. (2018). The stations together along with their geographical locations and elevations are shown in (Table 1).

Data analysis
To identify linear and non-linear trends in monthly, annual and seasonal time scales of rainfall and temperature series, both parametric (linear regression) and non-parametric (Mann-Kendall and Sen's estimator of slope) statistical tests were considered. For homogeneity test of data the Van Belle and Hughes was also utilized.

Linear regression
A straight line is fitted to the data to determine whether the slope is different from zero or not. A simple linear regression method was used to determine the tendency (Eq. 1). Student t-test was applied to determine the statistically significance of the trend at a 5% significant level.

= + (1)
Where: Y is the dependent variable, a is the slope, x is the independent variable and b is the intercept.
The parametric test is commonly applied to normal data. Hence, before analysis, the rainfall and temperature data were tested for normality using the Shapiro-Wilk test ( Shapiro and Wilk 1965

Mann-Kendall and Theil-Sen's Slope estimator
Monthly, seasonal and annual trends were assessed using the Mann-Kendall trend test (Kendall 1975)

Mann-Kendall test
The Mann-Kendall test S of the series X was calculated by applying equation 2 and 3 (Mann 1945;Kendall 1975): Where, sgn (Xj-Xi) is the signum function, Xi ranked from i = 1, 2…N-1 and Xj ranked from j = i+1, 2…N.
Each data point Xi is used as a reference point and is compared with all other data points Xj values.
The Mann-Kendal test statistics (S) represents asymptotically normal distribution and give the mean of zero and the variance of one (Mann 1945;Kendall 1975).
The variance associated with S is calculated by applying equation 4:

Where: m is the number of tied groups and ti is the number of data points in group i
The standardized test statistic Z(S) is calculated by applying equation 5 (Mann 1945;Kendall 1975).

The Ho (null hypothesis) is rejected, if absolute value of Z(S) is greater than Zcritical (Zα/2), in which α represents the level of significant. Positive values of normalized test statistics Z(S) indicate an increasing trend and negative Z values indicate decreasing trends. In this study, statistically significant level at 95%
confidence level (at α = 0.05) was used.

Theil-Sen's Slope estimator
This method is used to quantify the linear trend in time series analysis. The slope Ɵi between two values in a time series X is estimated by equation 6:

Where Xk and Xj are data values at times k and j (K>J)
The median of these N values of Ɵi is known as Sen's estimator of slope and calculated by equation 7. The positive sign of Ɵmedian represents an increasing trend while negative sign shows a decreasing trend.

Van Belle and Huges' homogeneity of trend tests
To test homogeneity of trends, the most widely used Van Belle and Hughes (Belle and Hughes 1984)  To get the trend homogeneity of temperature and rainfall at multiple stations, Van Belle and Hughes proposed a procedure based on the partitioning of the sum of squares. The analysis procedure uses chi-square (X 2 ) test statistics of various chi-squares (X 2 homogeneous, X 2 month/season, X 2 station, X 2 month/season-station interaction) for testing the trend homogeneity.
The formulas from 10-15 were used to test the homogeneity of the trends by partitioning the X 2 total into respective source of variations Where: K is the number of month or seasons, M is the number of stations and Z is the standardized test statistics with KM d.f.

Test for significant of common trend (X 2 trend)
Where: K is the number of month or seasons and M is the number of stations and, ̅ 2 is the average square of standardized test statistics with 1 d.f.

Homogeneity trend test at different stations in different months/seasons (X 2 homogeneous)
The values of Zi and ̅ are calculated as: Where: Si is the Mann-Kendall statistic for month i, and k= 12 for monthly data values, with (KM-1) d.f.

Test for monthly or seasonal heterogeneity (X 2 month/season)
Where: M is the number of stations with (K-1) d.f.

Test for station heterogeneity (X 2 station)
Where: K is the number of month/season and with (M-1) d.f.

Test for the interaction (X 2 month/season-station)
During homogeneity test, four possible scenarios were examined based on Van Belle and Hughes (1984): When all values of X 2 for stations, X 2 months/X 2 seasons, and the interactions are non-significant (ii) When values of X 2 months/X 2 seasons is significant and X 2 values of stations is non-significant, different trend direction of each months/seasons was tested using M ̅ 2 , (i = 1,2, …, K) (iii) When X 2 values of stations is significant and X 2 months/X 2 seasons is non-significant , different trend directions at each station was tested using K ̅ 2 , (m = 1,2, … , M), and (iv) When both months/seasons and stations or the interactions are significant, the only meaningful trend test was possible for individual month/season-station using Zim, (i = 1, 2, …, K; m = 1, 2, …, M).

Crop risk assessments
In this study, crop risks associated with extreme events including dry spells and other growing season characteristics (e.g., Onset, Cessation and Length of Growing Period, LGP) was considered for the analysis in R-Instat (V.0.6.2) software (http://r-instat.org/Download) developed under the African Data Initiative (ADI) (https://africandata.org/).
Growth season characteristics such as onset and cessation date, Length of Growing Period (LGP), and dry spell length were determined using 30 years of data from seven stations. Specific rainfall characteristics information is vital for crop planning and for carrying out agricultural operations. These are important to make crop-based decisions during seeding, fertilizing, selecting crop variety, selecting suitable cropping pattern, and selecting best agro techniques, etc.
Onset and cessation date: The onset of the rainy season in the Eastern zone of Tigray region was assumed to start as of June 19 after the wet spells occurred for at least three consecutive days and when the total rainfall is 20 mm or more if there was no dry spell longer than 7 or more days within 30 days. Moreover, the cessation date was assumed as the date when the stored soil moisture reaches 100 mm after rainfall falls below ETo values as per Higgins and Kassam (1981), which considers annual crops utilize 75-100 mm sored soil moisture during their harvest stage. The possible onset day was selected based on the information collected from farmers during the field survey. A minimum threshold value of rainfall (<1mm) was considered as part of a dry spell, which is an insignificant amount for crop use (Siva Kumar 1992).

Length of Growing Period (LGP):
The duration between the onset of the rain season (OS) and cessation of total seasonal rainfall (CS) represents the number of rainy days or the length of growing season (16).

= − (16)
Where: LGP is Length of Growing Season, CS is Cessation season rainfall and OS is onset season rainfall Dry spell length is a time period with no rain or less than 0.1mm rain for more than 7 days within 30 days. The average value of the dry spells was computed at a seasonal time scale during the main rainy season.

Annual and seasonal rainfall variability
The mean rainfall amount throughout the study area was found to be 572 mm per annum. A minimum total rainfall amount of 554 mm was recorded in Edagahamus station while a maximum total amount of 617 mm was obtained in Hawzen station (Table 2). The contribution of "Kiremt"(local language) season to that of the annual total rainfall amount is very large in all stations which varies between 54% and 84%. In addition, the contribution of the "Belg" season was not to be underestimated. In the majority of the stations, it contributed to more than 25% of the total rainfall. The "Belg" season is very useful for long-cycle crops. The long cycle crops are planted during this season before "Kiremt". Moreover, farmers also used this season for land management practices, such as repeated ploughing and in-situ soil moisture conservation activities.
However, the coefficient of variability (CV) of the total rainfall amount was much higher for "Belg "season that ranged from 37-45% than Kiremt season rainfall (21-31%) and indicating a very higher temporal variability of the seasons (Table 2). Several studies also showed that the CV of "Belg" season is higher than "Kiremt" season in the     Annual rainfall showed a positive increasing trend in most of the stations and varied from 0.8-5.51 mm/year (Table   4). In addition, "Kiremt" season rainfall values also showed an increasing positive trend varied from 2.34-6.78 mm/season whilst "Belg" season showed a decreasing trend varied from 0.74-2.29 mm/season (Table 4). However, annual and seasonal trends are found to be non-significant at 5% significant level. Our results corroborate the findings of previous studies that indicated non-significant rainfall trend in northern Ethiopia (Cheung et al 2008;Gebremicael 2017;Gebremicael 2020;Seleshi and Camberlin 2006;Seleshi and Zanke 2004;Viste et al 2012).

Seasonal rainfall, maximum and minimum temperature trends
Compared to all stations, Hawzen station showed the highest increasing trend magnitude of annual (5.51 mm/year) and Kiremt (6.78 mm/season) rainfall. In contrast, Edagahamus station showed the highest decreasing trend magnitude during the Belg rainfall season (2.29 mm/season). Few stations (Adigrat, Sinkata and Wukro) showed a significant increasing Kiremt rainfall and significant decreasing Belg rainfall (Edagahamus) at p = 0.1 level.  The seasonal temperature exhibited non-uniform trend directions across the stations (Table 5). A significant increasing trend of maximum and minimum temperature, with a magnitude of (0.04-0.07 0 C/year, 0.024-0.06 0 C/year) and (0.07-0.1 0 C/season, 0.06-0.12 0 C/season, respectively) were observed during annual and Belg season in all stations at p = 0.01 significance level. Unlike the annual and "Belg" season, the maximum and minimum Kiremt season temperature showed an increasing and decreasing trend patterns. In most of the stations (5 out of 7), such as Illala, Edagahamus, Adigrat, Sinkata and Hawzen, the maximum temperature showed an increasing trend whereas in Wukro and Atsbi stations it showed a decreasing trend. In the case of minimum Kiremt temperature, a decreasing trend was observed in most of the stations (6 out of 7) except in Edagahamus. Yet, the Kiremt season maximum and minimum temperatures trends were not statistically significant at a 5% significant level in most of the stations. Few stations such as Edagahamus and Hawzen showed a significant increasing maximum temperature with a magnitude of 0.08 0 C/season, at p = 0.01 level and 0.02 0 C/season, at p = 0.1 level, respectively. Moreover, Adigrat and Hawzen stations showed a significant decreasing trend in minimum temperature by 0.04 0 C/season at p= 0.01 level and 0.03 0 C/season at p = 0.1 level, respectively. NN-Non-Normal data, *Statistically significant at 1% significance level, ** Statistically significant at 5% significance level, *** Statistically significant at 10% significance level

Monthly and seasonal rainfall homogeneity
In the monthly rainfall series (June-September), none of the station, season and station-season interaction components exhibited significant trend heterogeneity since X 2 station, X 2 month, and the interaction are less than the X 2 critical values (Table 6). But, the overall trend heterogeneity was found to be significant since X 2 trend > X 2 critical. The average of Mann-Kendall test statistics over months (k=1, 2, 3, 4) and stations (m= 1, 2,….7) was found to be positive ( ̅ . = 1.44) indicating an increasing trend.
However, in annual and seasonal (Belg and Kiremt) rainfall series, X 2 season only exhibited seasonal heterogeneity of rainfall trend since X 2 season is greater than X 2 critical values. However, the stations were found to have homogeneous trends (Table 7). Hence, according to scenario (ii) trend direction analysis at each season using K seasonal statistics becomes necessary while each season refers to the value of X 2 critical (with d.f.=1, i.e., 6.64) at 1% significant level. In evaluating the various homogeneity test it is best to use high significance level to obtain the same Z statistical sign by neglect small differences in the trend magnitude not seem too important (Belle and Huges, 1984).
Significance of trend homogeneity for each season was tested using the average Mann-Kendal test statistics (Zk).
The M ̅ 2 was obtained to test the overall seasonal trend homogeneity. Accordingly, the K seasonal statistics of M ̅ 2 becomes less than the X 2 critical values for annual and greater for Kiremt and Belg seasons (

Maximum and Minimum temperature homogeneity
The value of X 2 season for maximum and minimum temperature was significant which suggests heterogeneous (Table 9), In contrast, the value of X 2 station for maximum and minimum temperature was non-significant indicating homogeneous time series data. According to scenario (ii), testing trend direction at each season using K seasonal statistics becomes necessary. Table 10 and 11 indicated that the computed M ̅ 2 values for seasons were found to be greater than the critical X 2 (equal to 6.64) with d.f. = 1 at 1% significance level. Hence, the annual, Belg and Kiremt maximum temperature increased significantly (Table 10). Similarly, annual and Belg minimum temperature increased significantly while Kiremt decreased significantly (Table 11).
The trends in minimum temperature of the different seasons were not homogeneous in the study area as the trends in Kiremt were different from that of annual and Belg seasons. It is also noticeable 6 out of the 7 stations had negative trends in the season of Kiremt, while all stations experienced positive trends in annual and Belg season.    which may relatively a stable onset and cessation. Stable onset and cessation is advantageous for the farmers in searching other off farm activities once they have stabilized onset and cessation dates. for onset and cessation analysis and reported LGP in the north eastern stations ranged from 60-100 days (Berhe 2011;Gebre et al. 2013). The CV in LGP was larger than 15% in all stations, suggesting crop growing season are relatively unstable and relaying on one type of crop is risky in the study area. However, providing such kind of information is essential in selecting crop cultivars that can grow based on their maturity.

Trend characteristics of crop growing season
Similarly, the mean seasonal dry spell length was analysed and found to be in the range of 24 days at Adigrat to 30 days at Edagahamus. The dry spell length was highly variable with CV of 25 to 43%. This indicates availability of high risk on intra-seasonal water deficit in the study area. These findings agree with those of Gebreselassie and Moges (2016) who reported greater than 30% of variability in the dry spell length of daily rainfall. The highest variability of dry spell length was observed at Illala while the lowest was occurred at Edagahamus station.

Conclusions
This study analyzed the variability and trends of rainfall, temperature and characteristics of crop growth seasons in Eastern zone of Tigray region. Despite the observed significant trends in the rainfall and temperature in the study area, the trend of growing season characteristics did not change significantly in all stations over the study period. In addition, the coefficient of variability of the onset (CV, <10%) and Cessation (CV, <5%) indicated very small, possibly relatively stable onset and cessation. However, the coefficient of variability of LGP was (CV, >15%) indicates unstable and how risky was crop production under rain-fed condition relying in one type of crop in the study area. Moreover, the CV of the dry spell length was (CV, >25%) this also showed the high risk of intraseasonal water deficit situation in the study area. Overall, the higher variability of rainfall amount of Kiremt and Belg seasons and the higher variability of dry spell length of (CV, >25%) in association with short nature of LGP (68-85 days) had a negative impact on the agricultural activities of the study area during the study period . Hence, crop production in the study area demands appropriate adaptation strategies that considers the erratic nature of the rainfall, the long dry spell length in the season and increasing trend of temperature.