Effect of Rainfall Variability on the Maize Varieties Grown in a Changing Climate: A Case of Smallholder Farming in Hwedza, Zimbabwe

Rain-fed maize production has significantly declined in Zimbabwe especially in semi-arid and arid areas causing food insecurity. Erratic rainfall received associated with mid-season dry spells largely contribute to low and variable maize yields. This study involved a survey of current farmers’ cropping practices, analyses of climatic data (daily rainfall and daily minimum and maximum temperature) of Hwedza station and simulation of maize yield response to climate change using DSSAT CERES crop growth simulation model. The climatic and maize yield data was analysed using mean correlation and regression analyses to establish relationships between rainfall characteristics and maize yield in the study area. Survey results showed that maize was the staple food grown by 100% of the farming households while 8.7% also grew sorghum. The survey concludes that 56.2% of the farmers grew short season varieties, 40.2% medium season varieties and 3.6% long season varieties. The result of the correlation analysis of climatic data and maize yield showed that number of rain days had strong positive relationship (r = 0.7) with maize yield. Non-significant yield differences (p > 0.05) between maize cultivar and planting date criteria were obtained. Highest yields were obtained under the combination of medium season maize cultivar and the DEPTH criterion in all simulations. The range of simulated district average yields of 0.4 t/ha to 1.8 t/ha formed the basis for the development of an operational decision support tool (cropping calendar).


Introduction
In Zimbabwe, agriculture is one of the key sectors of the economy (Matutu, 2014), with an estimated 80% of the population directly dependent on agriculture.Over 65% of farmers practice rain-fed farming for their livelihood (Johnson and Chivandi, 2010;Mapfumo, 2013).Agriculture sector contributes 20.1% to the gross domestic product (GDP) (Crop, 2013).It is apparent that an agricultural production crisis is developing in Zimbabwe due to a steep decline in agricultural production (Mapfumo, 2013).This conclusion is based on declining income derived from agriculture as compared with non-agricultural sectors and the fact that the growth rate of agriculture has deteriorated relative to the population growth rate.In addition the status of the country as, "the food basket of the Sothern Development Community (SADC)" is gradually dwindling largely due to several factors and amongst them are climate related factors.Zimbabwe's climate is changing with the rate of change seemingly faster than what is normally expected (Asseng, 2015;Geerts and Raes, 2012;Hertel and Lantz C. Baldos, 2016;Sithole and Murewi, 2009).As a result, the climate of Zimbabwe is now characterized by many extreme events such as severe storms, floods and droughts, strong winds, hail, frost and heat waves (Sithole and Murewi, 2009;Zinyengere et al., 2014).
Low agriculture production in Zimbabwe may be attributed to several factors such as market forces that are disincentives, lack of irrigation facilities, inability of the farmers to access inputs in time, conservative attitudes towards extension services, field operations, timing of planting, and superimposed on these is the lack of use of adequate climate information (Conway et al., 2008).
Climate change and variability have become topical issues in recent time because of their largely detrimental impacts on natural and human systems (Mugalavai et al., 2008;Phillips et al., 1998).The most frequently cited activities or operations likely to be affected by climate change are agriculture, forestry, hydrology and fisheries especially in semi-arid regions (Laura, 2017).Within the agriculture sector, an extreme event like drought is arguably the most important climatic challenge and has major impacts on rain-fed cereal yields (Chimonyo et al., 2016;Kosgei, 2009;Mabhaudhi et al., 2018;Mendelsohn, 2008).
Use of agro-climatic information can improve rain-fed production but there is a general lack of readily available useable information accessible to farmers to achieve this (Keating et al., 2003;Mhizha et al., 2014).In agriculture, decision support tools for making choices such as crop variety selection, planting dates and timing of fertilizer application can contribute to increased yields under rain-fed conditions (Geerts and Raes, 2012;Hsiao et al., 2009;Vanuytrecht et al., 2014).However these decision support guidelines for rain-fed farming systems are not available to resource-poor farming communities such as subsistence farmers in Zimbabwe (Mhizha et al., 2014;Mupangwa et al., 2011;Raes et al., 2004;Zinyengere et al., 2014).Optimum management practices, such as planting criteria and cultivar selection can be accessed through validated models such as DSSAT for making seasonal or within-season decisions (Pickering, 1996;Vanuytrecht et al., 2014).One tool which can fill this gap is a crop calendar.A crop calendar is a tool that provides timely climate related information about crops and varieties to promote local crop production.It contains information on planting dates, sowing and harvesting periods of locally adapted crops in specific agro-ecological zones (Mendelsohn et al., 2006).
Previous attempts to study effects of rainfall on crop yield in Zimbabwe (Alumira and Rusike, 2014;Crop et al., 2012a;Mhizha, 2010;Mhizha et al., 2014;Mupangwa et al., 2011;Raes et al., 2004;Zinyengere et al., 2011) were based on some rainfall characteristics such as dates of onset and end of rainy season, and temperature.Though these studies were comprehensive, farmers' practices and important rainfall characteristics, like rain days that are also vital to crops were not considered.It is against this backdrop that this study focuses on examining the length of growing seasons and matching them with crop growth cycles in Hwedza district of Zimbabwe.Planting windows, best planting combination and variety selection were developed into a maize crop calendar that will guide farmers on planting dates, dry spell probability during critical stages of the crop growth cycle and rainfall cessation.In addition, current common farmers' practices were investigated through a sampling survey.Historical climate was characterized from observed daily temperature and rainfall data and a crop growth simulation model DSSAT (Decision Support System for Agro-technology Transfer) version 4.5 was used to explore implications of observed changes in daily rainfall records on district maize average yield.

The study area
The study was carried out in Hwedza district of Zimbabwe, (Figure 1).The geographic position for Hwedza climate station is: latitude -18 o , longitude 31 o and altitude 1 425 m.The average total annual rainfall for Hwedza is 807 mm.The rainy season starts in November and tails off in April.Ther is wide spatial and temporal variation in rainfall.The mean annual temperature ranges from 18-19 ºC at about 1400 m above sea level.Maximum temperatures are low in the winter months (June-July) and highest in the month of October (Crop et al., 2012b).

Importance of socio-economic surveys in agriculture
In research farmers experience and practice are the backbone of research questions.One of collecting data is through surveys.Survey research involves the collection of information from a sample of individuals through their responses to questions.Survey research owes its continuing popularity to its versatility, efficiency, and generalizability (Quantity and Marsh, 2017).Surveys are efficient in that many variables can be measured without substantially increasing the time or cost (Bank and Change, 2018;World and Annual, 2008).For a survey to succeed, it must minimize the risk of two types of error: poor measurement of cases that are surveyed (errors of observation) and omission of cases that should be surveyed (errors of non-observation) (Division and Banda, 2003;Peytchev et al., 2011) .
Potential problems that can lead to errors of observation stem from the way agro-meteorological questions are written, the characteristics of the farmers who answer the questions, the way questions are presented in questionnaires, and the interviewers used to ask the questions.

Sampling design
A stratified two stage random sampling design was used for the survey.The sample size was based on central limit theorem and sample size calculation formula was based on the assumption of normal Approximation, (Davis et al., 2016;Division and Banda, 2003).

Meteorological data
One meteorological station was selected for this study; the station was selected based on availability of meteorological data.Forty nine years of daily rainfall data of Hwedza station for the period 1963 to 2012 was collected from Meteorological Service Department (MSD).The choice of this length of period is in line with the convention of using 30 years' weather data in characterizing the climate of an area, as adopted by the World meteorological Organization (Scafetta and B. J. West, 2009).

Rainfall
The following definitions were used during weather characterizations.
• agricultural rain (wet) day: when an area receives 4.95 mm or more of rainfall • rain day: when an area receives 2.95 mm or more of rainfall • dry day: any day that accumulates less than 2.95 mm of rain • water logging rains: At least 100 mm of rain within five consecutive days The season was considered to have started if an area receives effective planting rains.In the study, INSTAT "Climatic Events" menu options was used to define events of interest (start of season, dry spell, water balance) (Gallagher and Stern, 2015).Then sowing dates were based on the following definitions commonly agreed upon with the farmers, 1. AREX criterion, first date from 1st October when an area receives more than 25 mm in 7 days with the condition that there is no 10 day dry spell or longer within the next 20 days.

MET criterion, first date from 1st
October when an area receives at least 40 mm in 15 days, with the condition that there is no 10 day dry spell or longer within the next 20 days.
3. Depth criterion, first date from 1st October when an area receives at least 40 mm in 4 days (Mugalavai et al., 2008).

Climate impact scenario assessment with DSSAT and Simulation procedure
In this study the DSSAT model was calibrated and validated using the district average observed data.

Weather data
The three main weather parameters required for use by the DSSAT model are: daily rainfall (mm), daily minimum temperatures (T°C) maximum temperature (T°C) and daily solar radiation (MJ/m-2 ).

Soil characteristics
The soil texture characteristics at Hwedza were considered to be in Group G5-felsic igneous and metamorphic rocks, moderately deep coarse loam (Mtambanengwe and Mapfumo, 2005).The field capacity and wilting point for textural classes of Zimbabwe are found at -10 kPa and -1500 kPa suction pressures respectively (Journal, 1983;Mtambanengwe and Mapfumo, 2005) (Table 1).

Management and Experiment Data
Planting dates The planting dates used for the study were based on the optimal plantings commonly used in Zimbabwe.In this study, method applied are the Depth criterion, MET criterion and AREX criterion dates (Raes et al., 2004).

Fertilizer management
Two levels of fertilizer application were used: a ´optimal´ level and a lower level more representative of resource poor farming communities in Hwedza.

Model calibration and validation
The DSSAT model was calibrated by continuously adjusting the maize genotypic input parameters so as to mimic observed average district grain yield.The model was calibrated using the observed average district grain yield data collected from 1999 to 2008.
The CERES groups of models require a set of parameters describing the crop environment interactions, termed genetic coefficients.To use the CERES maize models in Zimbabwe, the genetic coefficients of the widely grown varieties were obtained from CIMMYT, (CAO et al., 2012;Jones et al., 2003) (Table 2).

Model validation
Validation of the model was done by use of the root mean square error (RMSE).The RMSE is a measure of the difference between the simulated predicted outcomes.

Gender
Out of 438 sampled farming household, 34.9% were females farming households while 65.1 % were males.In Hwedza there fewer females who are household heads.

Cropping information
The descriptive statistics from SPSS, showed that 100% of farming households in Hwedza grew maize because it is a strategic and staple grain food in Zimbabwe.Sorghum is a drought tolerant cereal crop but is grown by only 8.7% of farming household in Hwedza district, especially in communal area and small scale sectors.

Start of season from the survey
In Hwedza, the start of the season varies depending on farmers location micro climates .Out of 438 farmers interviewed (Table 3) the start of season is very variable.There is no consistent of start of season.The inconsistent of start of season might attributed to climate variability in Zimbabwe (Zinyengere et al., 2015) .The t-tests of correlations (at the 5% level of significance) (figure2) exposed that there was no significant trend in the planting dates computed by any of the three criteria over the past 49 years (Fiwa et al., 2014).From the analysis of Depth criterion with less risk of crop failure after the start of there is evidence that the climate variability (Sithole and Murewi, 2009)

Occurrence of dry spells
The probability and occurrence interannual fluctuations in precipitation-dry spells were reported in Hwedza (Table 4).Dry spells cause water stress in maize.Water is one of the main yield reducing factors in maize production.Availability of adequate moisture at rootzone is essential for maize development.Scarcity of water (drought stress) at critical stages (anthesis and grain filling) of maize cause yield loss (Basir et al., 2018).Potential yield can be reduced by 10-50% if dry spells occurs at 8-16 leaf stage of maize phenology.Water stress during this period will reduce ear size and potential Results from survey and analysis of historical data (Table 4) rainfed maize anthesis period in Hwedza takes place between January and February.The critical maize growth stages (silking and grain filling) often coincide with mid-season dry-spells.Silking stage is the most sensitive stage for water stress.
Water stress during silking coupled with high temperatures can result in 100 percent loss (Fiwa, Resources and Raes, 2014;Zinyengere et al., 2015;A Ahmed et al., 2018;Basir et al., 2018).This explains the importance of growing varieties which match with length of growing period.Dry weather that starts early and covers several growth periods will have a compounding effect with severe reductions in maize and sorghum yields.False starts of the rainy season cause replanting and increase cost of production of rain fed crops in Zimbabwe, therefore starts of season need to be monitored to reduce risk of crop failure (Mupangwa et al., 2011).

Agricultural seasonal rainfall characteristics and maize phenology
There is a marked decrease in total amount rainfall received in January and February for 49 years analyzed (Figure 3a).A marked decrease in total rainfall received in January and February signify characteristics of climate change.There is a gradual decrease in total rain received in month of January from selected 10 year period (1992, 2002 and 2012) which is a sign of climate change in Hwedza (Sithole and Murewi, 2009).In rainfed maize production January and February months are critical because the dates usually correspond with critical stages of cereal crop growth.The average seasonal rainfall showed a significant increase (Figure 3).A significant increase of total seasonal precipitation is not important in rainfed maize production.What matters in crop phenology development is the rainfall distribution (Geerts and Raes, 2012).Analyzed climatic data of Hwedza station there is a marked decrease in number of rain days in a season (Figure 4).Two definitions were used to define the agricultural rain day: (a) any day when an area receives 4.95 mm or more of rainfall, (b) any day when an area receives 2.95 mm or more of rainfall.

298
The long season cultivar which requires greater than 150 days to mature gave lowest yields (0.3 t/ha)

299
(Table 6) when planted late because the average length of growing season is 139 days therefore the of experiencing dry spells greater than 14 days is very high.Water stress at anthesis period and grain filling reduce yield drastically.

3.3.2: Validation of simulated results
By using GenStat, the separation of the mean was carried out to find which varieties are most suitable in Hwedza and were compared with results from observed maize district average yields.The simulated results for medium season and short season varieties showed no significant difference from the observed maize district average yield, P = 0.6 and P = 0.16, respectively while significantly difference were obtained between simulated results for long season varieties and district average, P = 0.004 (Fiwa et al., 2014;Zinyengere et al., 2015).The results from the survey farmers who grew long term varieties were 3.6 %, medium term varieties 40.2 % and short-term varieties 56.2 %.In research, finding farmers experiences and practices of farmers are important in crop modelling and in development of cropping guidelines.Farmers perceptions make calibration and validation of model easy.In science of modelling garbage in garbage out scenarios are common(K.L. MACKERRON, 1999;Mukhtar, 2012).During fine tuning a model it's important to compare simulated results and data from baseline surveys.

Evaluating planting criteria and maize cultivar
STATA version 15.0 was used to evaluate the effects of planting criteria and maize cultivar (Figure 5).The variance ratio for planting criteria to maize varieties was not significant as illustrated by standard mean errors bars (Figure 5).The combination of AREX criterion and maize varieties produced highest yields followed by MET criterion and lastly, the depth criterion.The results show less significant differences in yield obtained from simulation with different combination of planting criteria.The combination of depth criterion and medium season cultivar is recommended because depth criterion has less risk of crop failure (Raes et al., 2004).

Figure 5. Combination of planting Criteria and cultivar (AREX, MET, DEPTH criterion) effects on yield
By planting late, farmers avoid the potential false starts which may occur early in the season.This ensures that crops do not suffer moisture stress in the initial and crucial stages of development (Chivenge et al., 2015).The medium and short varieties obtain better yields when planted late and early.The maize yield of short and medium varieties was insignificantly affected by time of planting (Figure 5).

Short. Var
Medium.var

3.3.4: Cultivar optimization for maize crop calendar
Testing of null hypothesis that mean of medium, short and long-term varieties were equal to mean of observed yields (Table 7).Medium and short varieties (P=0.6333 and P=0.169) respectively (Table 7) there's no difference between the means and conclude that a significant difference does exist.The results (Table 7) are consistent with observation by (Mhizha et al., 2014;Phillips et al., 2006) that medium-term varieties are dominant in agro-ecological region IIb and III in Zimbabwe, while in drier zones the more drought-tolerant crops, pearl millet (Pennisetum glaucum.L) and sorghum (sorghum bicolor.L) are recommended.Some farmers are risk takers they grow long varieties (P value < 0.05) (Table 7) where there's difference between the means and conclude that a significant difference exist.Climate variability confuse farmers to extent of choosing maize varieties which don't match with crop varieties water requirements.In some LaNina years farmers benefits from well distributed rains which favors better yields from long maize variety.

Relationship between total seasonal rainfall, simulated yield and observed yields
The simulated results showed that DSSAT is sensitive to moisture stress, it failed to simulate 1992 yields because of the drought experienced in that season (Figure 6), shows that in years where total seasonal rainfall was greater than 700 mm simulated and observed yields were greater than 1 tonne.
Not all years with totals seasonal rainfall greater than 700 mm had good yields, this proves that rainfall distribution (rain days) is more important than seasonal totals.In a season, if severe midseason dry-spells greater than 14 days occur at anthesis and grain filling stages, final maize yields are affected.Maize requires about 600 mm of rainfall of even distribution to reach physiological maturity.
In case of low seasonal total rainfall (less than 600 mm) and few rain days (erratic rainfall distribution) growing of small grains is highly recommended.In Hwedza only 7.8 % of farmers grew sorghum (or other small grains) which implies that farmers are advised to grow both maize and sorghum to reduce the negative impact of climate variability on yields.The correlation (r = 0.7) between rainfall characteristics and maize yield shows that annual number of the rain days had strong (significant) positive correlation with maize yield; while dates of onset (r = 0.003) had weak correlations with maize yield (Table 8).This implies that variations in annual rain days and annual rainfall amounts account for most for the annual variations in maize yield.This confirms the findings by (Geerts and Raes, 2012;Mhizha et al., 2014;Raes et al., 2004;Zinyengere et al., 2014) who reported that annual rain days and annual amount had the greatest effect on maize yield.Thus, the higher the amount of rainfall spread over the number of rain days in a year, the greater the maize yields.
With the seasonal length varying from 90-139 days, the results show that the length of growing period is not very reliable to maize production but very suitable for drought tolerant crops like sorghum and finger millet.Seed companies in Zimbabwe have a range of sorghum varieties which can be grown throughout the country but particularly in areas where maize is a high-risk crop.In these areas where the higher drought tolerance of sorghum is so important it is more likely to produce a crop in years with less than average rainfall.(Mhizha et al., 2014;Mugalavai et al., 2008;Raes et al., 2004).Number of raindays are strongly correlated (r = 0.7) to district maize average yield.The more the rain days the more the more the water availability in the root zone.

Maize (Crop Calendar) in Hwedza (Decision support tool)
The decision support tool for maize production was developed for Hwedza district (Figure 7).The tool was developed based on triangulation of the survey results, analysis of 49 years' historic climatic data for Hwedza climate station and simulation results by DSSAT model.The tool is meant to help maize producers, to make decisions for maize production management based on the climate variability in Hwedza.

Figure 7. Maize crop calendar for Hwedza District
In applying the decision support tool, farmers, it should be noted that when they plant early the district average yields as high as 1.8 t can be attained if the long, medium and short varieties are planted between 1 October to 10 November) with their current fertilizer application.
In general , highest maize yields in the crop calendar are obtained after averaging few success years with October plantings due to favorable heat units for hybrid maize growth (Geerts and Raes, 2012;Mhizha et al., 2014;Zinyengere et al., 2014), but these can usually only be sustained with irrigation.
However, plantings in October may be feasible on vleis or when there are good early rains (over 40 mm).Early planting in October in figure 3.7 has fewer chances of successes in crop establishment due to dry spells which often occurs after early plantings.
The red color represents the chances of having mid-season dry spells, which are greater than 14 days.
Planting in December produce low yields and is very risky because the crop may suffer from water stress at the end of season, while pests like stalk borer and disease (maize streak virus are more prevalent to late planted maize crop. The blue lines represent period from planting to anthesis stage.The brown strip represents the common end of season.It is clear that with the use of DSSAT model, the length and the quality of the growing period can be assessed for its suitability to rainfed production of maize varieties (Chisanga et al., 2017;Chivenge et al., 2015;Ndawayo et al., 2017;Zinyengere et al., 2015).As a result, a crop calendar that can be used to indicate whether to grow maize varieties of particular length of growth cycle was developed for Hwedza district.

Conclusions
The maize cropping calendar (guidelines) was successfully developed as a general guide applicable to the study area.
The survey helped to find agro-meteorological challenges which are affecting maize production in Hwedza district of Zimbabwe.Results of this study indicate that rain fed maize farming in Hwedza is constrained by inadequate precipitation.The survey results showed that the start of the season was very variable generally ranging from the 1 st dekad of October to the 1 st dekad of December.The survey concludes that 56.2 % of the farmers grew short season varieties, 40.2 % medium season varieties and 3.6% long season varieties.
The study has established that there is high variability in rainfall characteristics (start of season, number of rain days and dry spell length) which translates into high variability in maize yield per hectare.The result also reveals that the number of rain days the strongest influence on maize yield per hectare in the study area (r = 0.7) while start of season has less effect on district average maize yield (r = 0.003).
Use of one station based climatic data for the study area does not represent the whole study area, therefore use spatial data might help to come up with cropping guidelines.
The validated DSSAT model was applied together with an optimization to select the best combination of planting criteria and maize cultivar.The study showed that maize cultivar and planting date affect maize yields considerably.The DEPTH criterion and medium season cultivar produced the highest district average yield of 0.9 t/ha.Though crop modelling approach can be used to develop cropping guidelines, this study reveals that dedicated local knowledge is necessary to calibrate and validate models.

Figure 1 :
Figure 1: Map of Zimbabwe showing agro ecological zones and Hwedza district is significantly affecting the onset of the cropping season (figure2).The variability of onset dates with time on a station further proves a well-known point that rainfall pattern and distribution from one station cannot be used as representative enough to give operational guidelines of the whole district.There are different microclimates in Hwedza district which was proved by the survey results.

Figure 2 .
Figure 2. Vertical boxplot showing criteria used to determine start of season

Figure 3 .
Figure 3. Hwedza monthly rainfall changes.Error bars show mean deviation

Figure 4 .
Figure 4. Number of rain days in a season (dashed line = rain days> 2.95; continuous line = rain days> 4.95 mm)

Table 2 .
Genetic coefficients of Zimbabwe maize varieties P1: length of the juvenile period (degree days above 8 o C P2: Factor to account for delay in development when day length is less than the optimum P5: Time in degree days from silking to maturity G2: Maximum kernels per plant G3: Kernel filling rate (mg/day) during grain filling under optimum conditions PHINT: Phyloochron interval (time in degree days between successive leaf tip appearance)

Table 3 .
Start of season in Hwedza district

Analyzed start season from historical data using INSTAT Three
criteria for determining the start of season were analyzed using INSTAT.According to the AREX criterion, the season often starts on the 2 nd dekad of November (11 November), while it starts on 2 nd dekad of November (16 November) based on the MET criterion with the Depth criterion giving a common start of season on the 3 rd of November (day 148) of the Julian calendar starting from 1 July.

Table 4 .
Risk of 14 day, 21 day and 28 days dry spells in a season at Hwedza station

Table 5 .
Average district observed yields and simulated grain yields

Year District Observed Yields (T/Ha) Simulated Yields (T/Ha)
(Geerts and Raes, 2012;Zinyengere et al., 2014;Chisanga, Phiri and Chinene, 2017;Ndawayo et al., 2017)ene, 2017;Ndawayo et al., 2017).Calibration of district grain yield simulation was done by assessing the goodness of fit of simulated against grain yield and also two sample t-test at 95 % confidence interval.During calibration the model initially failed to correctly simulate crop yields and

Table 8 .
ANOVA for regression of observed average yield on number of rain days