Monsoonal rainfall characteristics in the context of climate adaptation planning for rain-fed agriculture in the Sudano-Sahelian area of Northwestern Nigeria


 This study was aimed at assessing monsoonal rainfall real onset dates (RODs), real cessation dates (RCDs) and extent of association between cumulative rainfall (CR) and length of growing season (LGS) in the context of climate adaptation planning for sustainable rain-fed agriculture in the Sudano-Sahelian area of Northwestern Nigeria. Daily rainfall data of four stations purposively selected namely: Gusau, Kano, Katsina and Sokoto for the period 1981–2018 were collected from Nigerian Meteorological Agency. The data were analyzed and the Intra-seasonal Rainfall Monitoring Index (IRMI) was generated. IRMI was used in determining the RODs and RCDs of rainfall and LGS and CR. The Mann–Kendall test was used to detect trends in rainfall characteristics. Findings revealed that RODs, unlike RCDs of rainfall in the study area, show extensive variations from one station to another. There is a very low correlation (0.07 coefficient) between latitudes and early onsets (EOs). There is however a strong positive correlation (0.8 coefficient) between meridians and EOs of rains. Late onsets (LOs) recognize latitudinal differences to the extent that there is strong positive correlation (0.7 coefficient) between lines of parallels and LOs of rains. The three types of onsets interchanged with one another annually without a clear trend in the RODs and RCDs phases. We conclude that non-definite trends in RODs and RCDs pose a strong challenge to long term adaptation planning. The recommendations of the study are geared towards enhancing climate change adaptation in the context of complicated rainfall characteristics of the study area.


Introduction
The monsoonal rainfall season in the Sudano-Sahelian agro-ecological zone of Northwestern Nigeria lasts for only about four to ve months in a year. The distribution of rainfall, its commencement and ending in semi-arid and dry sub-humid zones of the study area have been highly variable in both space and time (Umar and Adamu, 2019). Alternating of a moisture-laden wind from Atlantic ocean with a dry wind originating from the Sahara desert derives a cyclic movement of rain-bearing mechanism that ensue high inter annual rainfall variability that frequently culminates into extreme events such as dry spells, droughts and oods (Ati, Strigter & Oladipo, 2002;Umar, 2016;Umar & Adamu, 2019). This has been occurring despite the fact that Northwestern Nigeria has been acknowledged for its high performance in rainfed agricultural productivity of cereals, legumes, vegetables and fruits. Rainfall characteristics such as real onset dates (RODs), real cessation dates (RCDs), length of growing season (LGS), and cumulative rainfall (CR) are critical factors for sustaining crop productivity in rainfed agriculture especially in this drylands region. It is increasingly acknowledged that very early and delayed planting may have damaging effects on the cultivated plants and farmers' nancial welfare. Farmers who engaged in early planting may guided by false onset of rainfall and this can lead to the wilting and failure of their crops. False onsets add to the hardship of farmers who have to replant after the destruction of their initial cultivation (Umar & Adamu, 2019). Likewise delayed planting by farmers may induce a reduced the LGS of their cultivated crops that can adversely affect its yields especially the late-maturing varieties.
Researchers (such as: Shevakuma, 1988; Oguntunde, Lischeid, Abiodun & Dietrich, 2014) established a signi cant positive association between the start of rains and the LGS. Thus, earlier onset (EO) most often leads to longer LGS and late onset (LO) is correlated with shorter LGS. There is an understanding that the LGS is more dependent on rainfall onset than on its cessation (Omotosho, 1992). One of the main determinants of planting suitability for a crop anywhere is its moisture requirement and LGS which is dependent upon RODs and RCDs. Rural households in the dryland of Northwestern Nigeria had been exposed to variety of climate risks such as harvest failure and death of livestock as a result of droughts resulting from extreme rainfall variability (Umar & Adamu, 2019). Thus, determining monsoonal rainfall characteristics is essential in adaptating to climate variability and change. The variability and unreliability of LGS in the study area exacerbate great risk to agricultural productivity (Mortimore, 2005; Haruna & Murtala, 2019). It is acknowledged that the extent to which climate variability is managed forms the bedrock for a wise adaptation to climate change and that "adaptation can be motivated by a diverse set of current and future climate hazards, including observed and expected changes in average climate, climate variability, and climate extremes" (Fu¨ssel, 2007, p266).
Water balance models have been extensively built and employed in determining RODs and RCDs by scholars (such as: Walter, 1967;Kowal & Knabe, 1972;Benoit, 1977;Stern, Dennett, & Garbutt, 1981;Faheun, 1983;Nicholls,1984;Mhita & Nassib,1987;Sivarkumar,1988;Nnoli,1996;Samba 1998;Omotosho, Balogun & Ogunjobi, 2000;Ati, Stigter & Oladipo, 2002;AGRHYMET, 2005). We can deduce from these works that for sure RODs and RCDs differ from place to place and they can be measured using multitude quantitative analytical frameworks ranging from simple to complex mathematical equations. Some of the reasons behind the disparity in these frameworks include that scholars differed in the agroclimatological settings in which their models were developed (predominantly tropical Savannas and forests of Australia, Kenya, Burkina Faso, Sothern Nigeria and Northern Nigeria, etc.). Even models that focus on a single setting may yield different results in calculation of RODs and RCDs because scholars differ in the variables they look into while constructing their models. Those variables employed in the models construction include amount of rainfalls (cumulative rainfall) period for distribution of amount of rainfall, local crop water requirement, duration of dry spell, usual time of crop planting and etc.
The model builders tried as much as possible to key into actual or real (actual) onset, avoid false onset and provide a reliable retrospective or proactive prediction of rainfall characteristics, especially the onset which is needed to determine a less risky planting date or planting method, or sowing of less risky types/ varieties of crops in responsive farming (Stewart, 1991). In this regard, using daily rainfall data that has been reduced or summarized into pentads rainfall amounts, Usman & Abdulkadir, (2012) proposed the Intra-seasonal Rainfall Monitoring Index (IRMI) model that determines RODs and RCDs of rainfall. Real or actual onset dates are determined using this index equation's that computes RODs, RCDs and CR commencing from the 1st May. This study adopted the IRMI because apart from being among the latest, when compared with other models, IRMI is very simple to use and has been speci cally developed to be applied in the study area and similar agro-ecological zones. Therefore, this study was aimed at assessing monsoonal rainfall real onset, real cessation and extent of association between CR and LGS in the context of climate adaptation planning for sustainable rain-fed agriculture in Sudano-Sahelian area of Northwestern Nigeria. The objectives of the research are to: determine the RODs and RCDs of rainfall in the study area; to examine spatiotemporal interannual and to establish (if any) nature and extent of trends in and relationships between LGR and CR.

Study Area
The study area as can be seen from Fig. 1  Crops produced include millet, sorghum, rice, cowpea, soy beans, wheat, groundnut, maize, cotton, sesame and vegetables (Umar, 2016, Umar & Adamu, 2018. In terms of agricultural land-use, the study area comprises of two wide belts of dominant staple cereals, millet and sorghum grown in varying proportions. There are other common cash crops that further distinguish the local economy namely cowpeas, groundnuts, cotton and sesame.

Procedures for sampling, data collection and analyses
A purposive sampling procedure was adopted in the selection of four synoptic meteorological stations out of the six of them in the Northwestern Nigeria ( Fig. 1) namely: Gusau, Kano, Katsina, Sokoto, Yelwa and Zaria. The selection was done while giving consideration to stations with longer consistent and most reliable daily rainfall records.
Daily rainfall data of four stations (Gusau, Kano, Katsina and Sokoto) for the period 1981-2018 were collected from Nigerian Meteorological Agency (NiMet) Head O ce, Abuja. The data analyzed and IRMI was generated. This determined RODs, RCDs, LGS and CR of rainfall in the study area. The rst step data analysis was compute pentad rainfall summations which formed the units of analysis of the study. This has been done with the aid of rainfall pentad calendar. After this, IRMI was employed to generate an index that determines the real onset date and the real cessation dates on a pentad-by-pentad basis beginning from the 1st May using the Eq. 1:

Where
Cumulative pentad rainfall since May 1 = The highest pentad total rainfall since May 1 = Number of breaks in rainfall (pentads with less than 5 mm of rainfall) and 100 = a factor The 'actual' or 'real' onset of rains is taken as the pentad within which the index is ≥ 1 for the rst time.
The actual or real cessation date is calculated when the cumulative pentad rainfall remain the same for two or more consecutive time.
With regards to detection of trend, a preliminary normality test we conducted on the distribution of RODs and RCDs (Table 1) revealed that these rainfall characteristics in the study area are not normally distributed. Hence employment of non parametric tests such as Mann-Kendall test to detect trend has become necessary. The Mann-Kendall is a nonparametric test for nding trends in time series. This test is widely used because one of the advantages of this test is that it does not require data to conform to any distribution before it can be used. The term trend has been de ned to stand for to a change characterized by a smooth, monotonic increase or decrease of average values over the period of record (Donaire, 2000: El-Tantawiy, 2006).
The equations for computation of Mann-Kendall Statistics , ( ) and normalized test statistic are as follows: The equation for calculation of the variance of S. VAR(S) is: Where n = number of data points; t i = are the ties of the sample time series; and m = number of tied value (a tied group is a set of sample data having same value) Equations 2 and 3 were then used to compute the test statistics Z. The computation for normalized test statistics Z is given as: A positive value of Z indicates an upward trend; a negative value indicates a downward trend, and a zero value indicates no trend. Alternatively, the trend can be determined using computation software applications such as: R-package, SPSS, and PAST3 statistical package.
RODs and RCDs were analysed in PAST3 statistical package where an index was generated through transformation (that is by getting the deviation of all the observations from the RODs and RCDs means). The result was interpreted as follows:

Results And Discussions
Result of a preliminary normality test we conducted on the distribution of RODs and RCDs (Table 1) revealed that these rainfall characteristics in the study area are not normally distributed.
who studied variabilities in rainfall onset, cessation and length of rainy season for the various agro-ecological zones of Ghana and found that onset of rainfall follows latitudinal pattern in Ghana from the forest coastland to savanna hinterlands and that LO is associated feature of the savannah zone. On average RODs commence earlier (21st May) in Gusau than in other stations of the study area, a difference of 10 days compared to 1st, June the date of the other three stations. This result is in agreement with Odekunle, (2004) who studied rainfall and length of growing season in Nigeria and found rainfall commences in Kano station around early June. Also the higher values of standard deviations and coe cients of variation of all the stations shown in Table 3 when compared with lower values of standard deviations and coe cients of variation for the cessation of rainfall as shown on Table 5 portray that though RODs have higher variability across the stations when compared with RCDs that occur almost the same time across stations in the study area. This result however, contrasts sharply or did not tally with Adejuwon, (2006) and Amekudzi, et al., (2015) who found distinctive dates for rainfall cessations across various agro-climatological zones of Nigeria and Ghana respectively. Another factor that account for the disparity in this study's nding with the two aforementioned results is that the differences in RCD between the Sudan (Latitude 12) and the Sahel (Latitude 13) agro-ecological zones is minimal when compared with differences that exist between say forest and Guinea Savanna or Guinea Savanna and Sudan/Sahel zones.
The distribution of RODs (Table 3)   The obtained values of skewness show that only RODs of Katsina are near normally distributed (-0.091). Two stations (Gusau and Sokoto) have one very low positive and another negative kurtosis and the other two stations (Kano and Katsina) have one low positive and one negative kurtosis respectively. This implies a tendency towards having light tails, or lack of outliers (see Fig. 4A-D). This is an indication of low occurrences of extreme events as per very early rainfall onset that could generate confusion to farmers regarding the safe period to plant or very late onset that could drastically affects LGS. However, in terms of variability, the RODs in Katsina are the most variable with 11.47 coe cient of variability and standard deviation of 3.56 while the RODs in Gusau are the least variable with 7.58 coe cient of variability and range of 25 to 34 (Table 3). High variability in the RODs of stations has deleterious implications on crops (maize, rice, sorghum, millet, groundnut and cow peas) yields the least sensitive has been rice because it enjoyed some irrigation supplementation in Kano and other areas of the study area (Adejuwon, 2006). It has been noted that in the study area, the crops whose yields are most sensitive to rainfall variability (essentially it de ciencies) in the months of June, September or both that is onset and cessation periods are cowpeas, maize and millet (Adejuwon, 2006 Table 5 when compared with higher values of standard deviations and coe cients of variation for the onset of rainfall as shown on Table 3 portray that RCDs occur almost the same time across stations in the study area. As stated earlier, this result did not tally with Adejuwon, (2006) and Amekudzi, et al., (2015) who found distinctive dates for rainfall cessations across various agroclimatological zones of Nigeria and Ghana respectively. The obtained values of skewness show that no any station has RCDs that are near normally distributed.
Only Gusau station has low positive kurtosis and the other three stations have low negative kurtosis. These show tendencies towards having light tails, or lack of outliers with regards to the all stations except Gusau. This has been con rmed by the station's highest occurrence of ECs (26.3%). This trend has serious implications on the sustainability and yields of late maturing crops such as sorghum, cowpea and soybeans. The RCDs are less variable than the RODs that is why there are low values of standard deviations and consequently also low values of coe cient of variability in the RCDs (Table 5) than in the RODs ( Table 3). The RCDs in Gusau are the most variable with 4.43 coe cient of variability and range of 48 to 61. This has sharply contrasted with the station's RODs least value of 7.58 coe cient of variability (Table 3).

Spatiotemporal interannual variability of rainfall characteristics
Since the normality test in Table 1 informed us that our data essentially RODs and RCDs are not normally distributed, consequently, in the process of detection of trend in the values of RODs, RCDs we have to use Mann-Kendall test because of its distribution non-bias. Result in Table 6 shows that, at a 95% con dence level, there is no statistically signi cant trend in both RODs and RCDs. But considering that there have been variations in those dates and in some stations (all the stations in the cades of RODs and Gusau in the case of RCDs) these variations have been signi cant, our study answered two questions one of them whether EOs of rainfall leads to longer LGS and invariably LOs shorter LGS and secondly to what extent LGS is dependent on rainfall CR in the study area? Exploring the variability the foregoing results and discussions show that both RODs and RCDs exhibited varying extent of disparity in their characteristics anomalies. Both Table 6  in Gusau that there is an increasing trend in the LGS. That is why the correlation between CR and LGS in all the stations is either very low or low correlation (see Table 6).
This result has tallied with Chamberlin and Diop (2003) that a late onset is not necessarily associated with a low seasonal rainfall amount. The implication of this lack of de nite trends during both phases of onset and cessation is that long-term adaptation planning will be di cult and rigorous exercise.

Conclusion And Recommendations
Findings of the study revealed that RODs unlike RCDs of rainfall in the study area so much vary from one station to another. There is this variation even between stations on the same latitude. There is very low (0.07) correlation coe cient between latitudes and EO. Early Onsets (EOs) of rainfall are characteristic features of Kano and Katsina stations where their frequencies and percentages of occurrences are both 26.3% respectively. There is however, a strong positive (0.8 coe cient) correlation between meridians and EO of rains. LOs recognize latitudinal differences to the extent that there is strong positive (0.7 coe cient) correlation between lines of parallels and LO of rains. All the three types of onsets EOs, NOs and LOs interchanged with one another from year to year without any explicit trend with non-de nite trends in the RODs as well as RCDs phases. There is a signi cant increasing trend in CR across all the stations with the exception of Gusau and conversely, there is no signi cant trend with regards to LGS in all the stations with the exception of Gusau. It is only in Gusau that there is an increasing trend in the LGS. That is why the correlation between CR and LGS in all the stations is either very low or low correlation.
We conclude that non-de nite trends in RODs and RCDs pose strong challenge to long term adaptation planning. In view of the complicating rainfall characteristics of the study area our study recommend as follows: a) There is the need to blend the traditional weather forecast system in the study area with modern scienti c one plus strengthening of other coping and adaptation strategies like off-farm livelihood diversi cation such as ownership of small ruminants in order to reduce climate risks; b) We also recommend enhancing of small and medium scale irrigation schemes and micro-nance to smallholder farmers in the study area and c) Finally this research suggest the need for a further study that will explore the occurrences of extreme events in respect of rainfall of the study area speci cally within inter-seasonal time-scale (dekads, pentads and daily) occurrences and their implication to rainfed agriculture.

ACKNOWLEDGMENTS
The principal researcher and co-authors are grateful to the Nigerian Tertiary Education Trust Fund for sponsoring this study under the Institutional Based Research initiative.

Declarations
Not applicable.
Con icts of interest/Competing interests I declare there is no con icts of interest/competing interest