Rainfall data for the two stations: Perkerra and Nginyang rainfall station was obtained from Kenya Meteorological Services. Perkerra and Nginyang rainfall stations had data for the period 1970-2008 and 1974-2013 respectively. The choice of the rainfall station was influenced by agro-ecological stations and amount of missing data. The two stations are located within the study sites agro- ecological zones LM5 and IL6. The percentage of missing rainfall data in the two stations was less than 10% and the data spans over 20 years hence meeting the requirements of World Meteorological Organization with regard to climatological analysis.

The study used imputation to fill in missing rainfall data. Multiple imputations method was used to overcome underestimation of standard errors and confidence intervals typical of single imputation [23] . The method involves replacing missing data with substituted values. The imputed data sets were combined and average worked out. The method involves estimation of missing rainfall data from the observations of rainfall (rainfall data sets) at the same station and period but different years.

The missing rainfall data Px was estimated using the following formula:

P x = 1 / n ∑ i = 1 i = n P i (1)

where:

n = the number of rainall data sets

Pi = rainfall data from the ith data set

Px = missing rainfall data

The study estimated standardized anomalies (drought index) from rainfall data using the formula:

S A ( t ) = { S P ( t ) − μ } { σ } (2)

where:

SA(t) = time-series of standardized anomalies

SP(t) = cumulative precipitation during the season

µ = represents mean

σ = standard deviation

The interpretation of Standard Anomalies was based on World Bank (2013) definition:

Anomaly lower than −0.9 = Catastrophic drought

Anomaly between −0.9 and −0.6 = Severe drought

Anomaly above −0.6 = Not severe

The indices were plotted to generate time series graphs and determine trends of drought events within the study period.

To ascertain whether there was a correlation between the total annual rainfall and the annual drought indexes-SA(t), a correlational analysis was done and a plot presented to display the relationship.

An oscillating trend of rainfall with four major positive peaks was observed in 1977, 1988/89, 1997 and 2007 while very low mean annual rainfall was recorded in 2000 (21.4 mm) and in 2002 (19.8 mm) with a general declining rainfall trend from 1970 to 2008 periods in Perkerra (Figure 2(a)). This provides a strong signal that drought events have occurred and are likely to reoccur.

On the other hand, the year-to-year annual rainfall variations in IL6 zone showed significant anomalies in annual rainfall between 2000 and 2010 (Figure 2(b)). Nginyang rainfall station shows an oscillating trend with major rainfall peaks observed in 1982 and 1997, and others in 1979, 1981 and 1985. The results concur with the findings of [24] , an analysis of extreme climatic events in Kenya that estimated long-term annual mean of 23.52 mm. Very low rainfalls (3.29 to 13.33 mm) far much below the long-term annual average featured in 1974, 1976, 1980, 1991, 1996, 1999, 2000 and 2009 which corresponds to the years Kenya experienced widespread drought with a decreasing trend of total annual rainfall (Figure 2(b)) in the study area (IL6). This indicates that the area will continue to become drier over time, increasing vulnerability if appropriate response strategies are not identified and implemented. Drought events in Congo in the years 1970s, 1980s and 1990s, corresponds with the period that Nginyang registered very low rainfall [25] . This finding indicates that drought can be widespread and can cut across both national and international boundaries, requiring trans-na- tional mitigation and adaptation approaches. Drought trends in Burkina Faso,

Ghana, Mali, Togo and Volta Basin supports this finding by indicating that frequency of dry years have increased in Africa but occur at relatively shorter intervals [26] .

The results of rainfall variability imply that pastoralists have planned adaptation strategies implemented to secure their livestock assets from drought event effects. The information is informative and significant in guiding decision making in relation to pastoral livestock production in arid and semi-arid Baringo County.

Figure 3(a) & Figure 3(b) presents a plot generated from the annual drought index for Perkerra and Nginyang Rainfall Station. On a time-scale of 12-month (annual), for the period 1970-2008, four extremely catastrophic drought periods were observed: 1984, 2000, 2002 and 2004 with standard anomalies less than −0.9 (SA (t)< −0.9) as a function of the time scales in Perkerra (Figure 3(a)). Other noticeable catastrophic drought events were 1972/1973, 1976, 1980, 1986, 1995/1996, 2001, 2003, 2006 and 2008 with standard anomalies less than −0.9 as a function of the time scales. Apart from the case of year 2003 and 2008, these results concur with the findings of [24] on extreme climatic events for economic development in Kenya. The two cases of year 2003 and 2008 can be attributed to

locational variation in drought events in place and time. Similar findings were recorded in Morocco, a county that experienced catastrophic drought in 1984 and 1999-2003 [27] . Studies registered the most severe droughts during the period 1999-2002 in Morocco, Algeria and Tunisia, results that corroborates the present study [28] [29] . This is an indication that the drought event did not only affect Baringo County, but also different parts of Africa. It also signifies spatio-temporal nature of drought events and associated impacts.

Noticeably, all the catastrophic drought events were preceded by high rainfall events (Standard Anomalies greater than 1, SA(t) > 1.00). This phenomenon takes place when sea surface temperature in oceans increases anomaly, causing sudden heavy rainfall and thereafter rainfall decreases drastically followed by a prolonged severe dry spell [30] . The effect can be seen in two forms, increasing then suddenly decreasing rainfall amount. For instance, the catastrophic drought event in 2000-2004 period that was preceded by the El Niño rains in 1997-1998 and has been classified by many experts as the worst El Niño effect in 20th century. Previous studies established very severe seasonal never span over many years in Africa, a finding that concurs with the results of drought analysis for the present study [31] . The obtained drought indexes show breaks between successive drought events in the study area.

The Nginyang station data on a time scale of 12-month (annual) reveals five catastrophic drought periods in the study area observed in 1974, 1976, 1996, 2000 and 2010 period (Figure 4(b)). All these periods indicated standard anomalies less than −0.9 (SA (t)< −0.9) as a function of the time scales. Noticeable severe drought events were observed in 1980, 1999 and 1991 with standard anomalies between −0.9 and -0.6 as a function of the time scales (Figure 4(b)). Studies attributed the 1970s and 1980s droughts experienced in Western Africa? Sahel to southward shift of the warmer sea surface temperatures in the Atlantic and warming in the Indian ocean [32] . The Ethiopia and Somalia drought is attributed to Indian Ocean sea temperatures that influenced the East African rainfall [31] [33] [34] .

Correlation analysis results between the annual rainfall and the drought indexes-SA(t) had higher and positive correlation in Perkerra (r = 0.9218) rainfall station which depicts a strong positive correlation between the total recorded annual rainfall and the annual standardized drought anomaly indexes SA(t) as

illustrated in Figure 4(a). This test was important to justify the use of drought index from rainfall amounts. It also helps to establish that drought events are directly influenced by rainfall received.

Similarly, correlation analysis results between the annual rainfall and the drought indexes-SA(t) in Nginyang also posted a positive correlation (r = 0.6879) (Figure 4(b)). This result is significant in analyzing effects of drought in livestock asset. This is an indication that severity of drought increases with decrease in rainfall amount.

The plots for the March-April-May (MAM) and October-November-December (OND) seasonal drought index-SA(t) for the variation of rainfall in 1970-2008 periods for the two stations are shown in Figure 5(b) and Figure 6(b) and over the period 1974-2013 is shown in Figure 5(b) and Figure 6(b).

The MAM Seasonal Drought Index plot for Perkerra rainfall station shows intermittent trend of drought events with peaks observed in the years 1979, 1983, 2000 and 2005 and a declining trend indicating drier conditions, which implies that the region is vulnerable to drought events (Figure 5(a)).

The MAM Nginyang Seasonal Drought Index plot shows fluctuating trend of drought events with drought peaks observed in the years 2000 and 2009 and others visible in the years 1974, 1996, 2010 and 2011. The long-term MAM seasonal trend for rainfall shows a declining trend, which is an indication of drier conditions over long time scale and implies vulnerability of the region to drought (Figure 5(b)). This finding corroborates [34] , which revealed a statistically significant increase in drought for the African continent during the period 1901-2011.

The OND Seasonal Drought Index plot for Perkerra station shows few drought events with the major drought event peak being observed in the years 2003 (Figure 6(a)) with smaller peaks in the years 1985 and 1996. Compared to the MAM Seasonal Drought Index plot, the OND season seems to be having fewer drought events for the 39-year period for Perkerra rainfall station. Noteworthy, from the OND Seasonal Drought Index plot, it was deduced that the long-term OND seasonal trend for rainfall shows a constant trend of below mean rainfall for Perkerra rainfall station (Figure 6(a)).

On the other hand, the OND Seasonal Drought Index plot also shows a fluctuating trend of drought events in Nginyang with catastrophic drought events being observed in years 1980, 1983, 2000 and 2010 (Figure 6(b)). Severe drought peaks include: year 1974, 1974 and 1996. Compared to the MAM Seasonal Drought Index plot, the OND season seems to have more catastrophic drought events for Nginyang rainfall station. However, from the OND Seasonal Drought Index plot, it can be observed that the long-term OND seasonal drought trend shows a gentle upward trend, an indication that the conditions are improving. The trend indicates that Nginyang station rainfall totals with time is likely to display an upward trend with improving drought conditions. Drought impacts negatively on food security and pastoral livestock production, calling for an integrated approach that is geared towards a shift from crisis management to risk management approaches that includes use of early warning systems and seasonal climate forecasts [25] [34] [35] .

Comparing the two regions, Perkerra and Nginyang, Perkerra is getting drier compared to Nginyang across the study period. This can be attributed to the generally improving conditions of Nginyang during the OND seasons across the study period. Across all the seasons, Perkerra MAM is getting drier as depicted by the relatively sharp dip in the plot. This is an alarming situation calling for more attention. MAM being the main season in the study area, the Perkerra livelihood is likely to be significantly affected.