1. Introduction
Crop genetic diversity is a key factor for long-term viability and adaptation to changing environmental conditions (Hammer and Teklu 2008). It prevents crop failure (Altieri 1994; Vandermeer 1989), contributes to more resilient systems, and limits susceptibility to pests and diseases (Tonhasca and Byrne 1994). A wide gene pool is essential for world food security and, as a source of materials for breeding new plant varieties, to mitigate current and future production risks. In particular, the ability of agricultural systems to mitigate climate risk lies in genetic diversity (Kotschi 2007; Lin 2011; Sgrò et al. 2011). Genetic diversity is therefore crucial to ensure the resilience of farming systems in the present context of environmental change. This is also true at the farm level, particularly where local landraces still exist. Selected and reproduced by farmers for many generations, landraces are likely to be locally adapted (Wood and Leene 1997) and maintain a high adaptation potential.
Climate changes impose a significant effort of adaptation (Funk et al. 2008). In smallholder agriculture systems, rainfall distribution is a key factor to the success of a crop growing cycle. Prolonged dry spells during the wet season are common and can cause crop failures and seed losses (Araya et al. 2010; Barron et al. 2003; Segele and Lamb 2005). One main concern for farmers is that crops obtain adequate soil moisture for seedling emergence, and throughout the vegetative and reproductive phases (Maracchi et al. 1993; Marteau et al. 2011). Marteau et al. (2011) showed that most of the sowing failures of pearl millet in Niger were related to long dry spells (>7 days). Long dry spells that occur after sowing induce a water stress to the newly emerging seedlings, contributing to their mortality.
Many farmers around the world observe rainfall patterns at the beginning of the rainy season with one question in mind: is it the right time to sow? Anticipating rainfall and synchronizing the sowing to ensure better water conditions for crop emergence remains a key challenge for farmers. The interannual variability of the onset of the rainy season points out the risk of sowing too early or too late and subsequently the risk of sowing failures. Many strategies are used by farmers to mitigate this risk. The ways of interpreting rainfall signals at the beginning of the season, and the decision of sowing, are not uniform among farmers. They do not sow all crop species at the same time. As such, the length of dry spells after sowing cannot be considered a factor that is independent of farmers’ practices. Farmers directly intervene in the length of dry spells through the choice of sowing dates.
Farmers’ sowing practices and the origins of genetic resources are potentially different from one farming community to another; however, such a social factor is rarely considered. Yet, studying how societies intervene between crops and climate to mitigate the impacts of rainfall variations should allow a better understanding of the social process of adaptation to climate changes in the future.
The eastern slope of Mount Kenya is a particularly relevant region to study how societies mitigate the risk of crop failure during seedling emergence. Over the mountain, agriculture is rainfed, making rainfall a key factor in crop production. The region, at altitudes ranging from 750 to 1100 m MSL, receives relatively low (600–900-mm mean annual rainfall) and highly variable rainfall (Camberlin et al. 2012), and farming communities are characterized by a high cultural diversity (Fadiman 1982, 1993; Heine and Moehlig 1980; Moehlig 1974; Peatrik 1999).
Combining ecological anthropology and climatology, Leclerc et al. (2014, manuscript submitted to Wea. Climate Soc.) implemented a retrospective survey from 1961 to 2006 and concluded that the cropping system dynamics (a social factor) induced over time an increasing risk of local crop variety losses due to the drought. The present study is complementary in identifying, at a smaller time scale, social and environmental factors that can mitigate this risk, considering farmers’ cultural practices and the origin of crop genetic resources they manage. Historical, sociological, and ecological variables are controlled with a double comparative approach (Leclerc and Coppens d’Eeckenbrugge 2012), and this allows a description of what a social process of adaptation to climate change can be. As studies on environmental changes usually occur on a long-term scale and are costly to conduct, we propose an alternative space and time substitution design. We compared two communities that moved along the slope of Mount Kenya into a new climatic environment, representing for both a rapid environmental change. Change in space, involving ecological contrasts, is similar to that induced in time by global environmental change. This context does allow analyzing factors at work in the social process of adaptation to climate change.
Around 1960, Mwimbi farmers began to move from 1100 m (highland climatic zone) to 950 m (midland climatic zone), whereas, about 10 yr later, Tharaka farmers began to move up from 750 m (lowland climatic zone). The migration of the communities was driven by the government agricultural policy on land consolidation and adjudication. The aim was at gathering land fragments into sizeable plots to intensify agricultural development and to ease population pressure in the highlands (Bernard 1972). The two communities, originating from different climatic environments, are today geographically close to each other in the midlands. They now live in the same climatic environment, but their experience of drought events and their genetic resources differ. Indeed, the Mwimbi came down from higher altitude, with their crop genetic resources, knowledge, and practices adapted to the highland climatic zone, whereas Tharaka farmers came up, with their crop genetic resources, knowledge, and practices adapted to the lowland climatic zone. By moving down, the Mwimbi faced a rapid warming of about +0.8° to +1.09°C (100 m)−1 and decreasing precipitation of 26.7 to 32.9 mm (100 m)−1 (Camberlin at al. 2012). According to McCarthy et al. (2001), such a variation corresponds with the warming observed in Africa in the last 40 yr. Reversely, for the Tharaka who moved up, precipitation increased and temperature decreased. The contrast between the two communities living today in the same climatic environment allows for studying the social process of adaptation to environmental changes.
The historical and social contrast between the two communities is amplified by their respective social networks, which still remain differentiated today. Midlands Tharaka farmers maintain contacts and relations (notably through intermarriage) with Tharaka farmers living in the lowlands. The same happens for midland Mwimbi with highland Mwimbi farmers, whereas intermarriage between the two cultural communities is unlikely in the midlands, even though they are geographically close to each other. Their respective social identity is commonly reaffirmed by farmers. Each community talks about the other maintaining a dichotomy between “us” and “them.” Such a social opposition is usual (Barth 1969), but its consequence on crop management and genetic resources have been the object of few studies (for instance, see Brush and Perales 2007; Longley 2000; Perales et al. 2005; Pressoir and Berthaud 2004a,b).
Leclerc and Coppens d’Eeckenbrugge (2012) have suggested that social factors organizing crop genetic diversity in situ operate less at farmer individual level than at the community level. They proposed in place of the usual two-way genotype by environment interaction (G × E), a three-way interaction model G × E × S in order to consider explicitly the social component (S). Under similar environmental conditions (E), crop genetic diversity (G) should be portioned between cultural communities (S). Following such a model, the hypothesis to be tested in the present context was that the coexistence of two historically differentiated communities of farmers, Mwimbi and Tharaka, should be reflected in differential crop genetic adaptability to long dry spells during the seedling emergence phase. Six main crops, namely beans—Phaseolus sp, cowpea/Vigna unguiculata (L.) Walp, green gram/Vigna radiata (L.) R. Wilcz, maize/Zea mays (L.), pearl millet/Pennisetum glaucum (L.) R. Br, and sorghum/Sorghum bicolor (L.) Moench—were considered in our analysis. If Tharaka farmers use their own varieties that are more adapted to drought conditions than those of the Mwimbi, the probability of sowing failure should thus be lower for the Tharaka than for the Mwimbi, in their common midland climatic zone. Testing this hypothesis is only possible if the sowing practices of Tharaka and Mwimbi farmers are similar, allowing control of one factor at a time.
Our strategy and methods, described in the next section, allowed the comparison of practices of individual farmers from Tharaka and Mwimbi communities and the testing of the differences in probability of sowing failure. Therefore, comparing these two communities in the same environment (same climate and soil conditions) should allow the identification of social factors involved in adaptation to climate change and how societies intervene between crops and climate variability.
2. Materials and methods
On the eastern slope of Mount Kenya, farmers manage two cropping seasons per year corresponding to the Long Rains (LR) and the Short Rains (SR) from March to May and from October to December, respectively [see Leclerc et al. (2014, manuscript submitted to Wea. Climate Soc.) for additional information]. Surveys were carried out over 2 yr (2009–11), including two Short Rains seasons and two Long Rains seasons (SR 2009, LR 2010, SR 2010, and LR 2011).
A double comparative approach was used to isolate climatic and social factors in the analysis. A total of 40 farming households were surveyed (Fig. 1), 20 at each of the two selected environments (950 m, hereafter the midland climatic zone, and 1100 m, hereafter the highland climatic zone). On the one hand, our sampling strategy aimed at having within the same midland agroecological zone (AEZ; after Jaetzold et al. 2007) an equal representation of the Mwimbi and Tharaka (one environment, two communities). On the other hand, it aimed at comparing Mwimbi farmers living in the midlands and highlands, that is, distributed between two distinct AEZs, with a semihumid climate in the midlands and a subhumid one in the highlands (one community, two environments). Figure 2 summarizes this double comparative approach. Within a given AEZ, households were selected at random, and we interviewed the person responsible for sowing.
Location of the study site with surveyed households. Automatic weather stations were installed in two different agroecological zones corresponding to the AEZ semihumid (midlands, 950 m) and the AEZ subhumid (highlands, 1100 m). The farmer sample allows studying two communities (Mwimbi and Tharaka) in one environment (midlands), as well as a single community (Mwimbi) in two environments (midlands and highlands).
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
Double comparative setting at the study site allowing the study of two communities (Mwimbi and Tharaka) in one environment (midlands), as well as a single community (Mwimbi) in two environments (midlands and highlands).
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
a. Farmer seed management and practices
Farmers’ crop varieties were inventoried, as well as their seed sources over the four seasons. The proportion of seeds that were locally produced by farmers was also recorded in order to assess how similar farmer practices and seed management were within and between communities.
Farmers were observed throughout the cropping season to characterize their sowing practices, by noting the dates of each sowing event, and their choice of crop species and varieties. Observations were carried out during four consecutive growing seasons. The observation units corresponded to parts of the field sown with one crop variety at a given date (hereafter plots). A total of 1691 plots were surveyed with a mean of 42 plots per farmer. Each plot was assessed for the proportion of empty planting hills 15 days after germination. We confirmed the reliability of this method by comparing independent observations by different field assistants. In the analysis, plots with sowing failure were recorded as 1, and those without sowing failure were recorded as 0. The proportion of sowing failure events was analyzed with a logistic regression model.
b. Climatic data collection
A Davis Vantage Pro2TM weather station was installed at each of the two environments, the midlands and highlands. The automatic weather station included temperature and humidity sensors, a rain collector, and an anemometer. It used frequency-hopping spread spectrum radio technology to transmit data to a console in which a data-logger was installed for data storage. In the midlands, the mean distance between the households surveyed and the weather station was 910 m for the Tharaka and 1340 m for the Mwimbi. Given this short distance, we assume that weather exposure was similar and that the rainfall variation between the two communities was negligible.
Rainfall variability was assessed on a daily basis to define the length of dry spell during crop emergence for the four growing seasons. The onset and the cessation dates and the duration of the rainy season were defined following Camberlin et al. (2012; see for details). A threshold of 2 mm was used to define a rainy day.
The germination date was defined as the first rainy day after the sowing date. The length of dry spell was defined as a run of consecutive dry days, using the germination date as the starting point and the next rainy day (>2 mm) as the ending point. The length of the dry spells was computed separately for each sowing event. Doing so, climatic conditions were specifically defined for each crop-sowing date. Soil water holding capacity was not used in the analysis as it is similar within the same AEZ (Jaetzold et al. 2007).
c. Statistical analysis
Multivariate analyses were carried out considering cultural communities (Mwimbi and Tharaka), agroecological and climatic zones (semihumid in the midlands and subhumid in the highlands), climatic season (SR 2009, LR 2010, SR 2010, and LR 2011), crop species (maize, bean, pearl millet, cowpea, green gram, and sorghum), and the length of dry spells after germination. The response variable, that is, the proportion of sowing failures, was modeled by a logistic regression with quasi-binomial error in place of binomial error to consider the overdispersion of our data. The regression was weighted by sample sizes (number of plots), and the logit link function was used to ensure linearity. Factors considered in the specific models were selected using a backward elimination procedure (Agresti 2007). This procedure begins with a saturated model, and the nonsignificant factors are sequentially removed to get the minimal adequate model. This model of analysis was run over the four seasons to compare the probability of sowing failure as a function of the (i) crop species, given the length of dry spell (lds in equations) and formalized as log(p/q) = α + β1cropi + β2ldsj + ϵ; (ii) agroclimatic zones (considering only Mwimbi), given the crop and the length of dry spell and formalized as log(p/q) = α + β1cropi + β2ldsj + β3AEZk + ϵ; and (iii) farmer community (considering only the midlands), given the length of dry spell and formalized as log(p/q) = α + β1communityl + β2ldsj + ϵ, where p is the proportion of sowing failure, q is the proportion sowing success, i = 1 to 6 crops, j is the length of dry spell (expressed in days), k = 1 to 2 AEZ (midlands or highlands, respectively), and l = 1 to 2 farmer communities (Mwimbi or Tharaka).
The goodness of fit D2, similar to R2 in linear regression, corresponds to the percentage of explained deviance. It was estimated by comparing the null deviance to the residual deviance of the minimal adequate model D2 = 1 − (model deviance/null deviance). Statistical analyses were performed with R data analysis software (R Core Team 2013).
3. Results
a. Farming system and sowing practices
The total number of varieties per crop species was similar between the Mwimbi and Tharaka in the midlands, except for bean (Table 1). Respectively, the two communities cultivated during the four seasons a mean of 5.4 and 1.2 bean varieties, 3.4 and 2.8 cowpea varieties, 1.6 and 2.5 green gram varieties, 7.8 and 5.7 maize varieties, 2.9 and 2.6 pearl millet varieties, and 5.5 and 4.0 sorghum varieties.
General characteristics of the surveyed population. Statistics were compiled for the four climatic seasons by combining the midlands and highlands (left part of the table). For the midlands, statistics were compiled to compare the four-season averages of the Mwimbi and Tharaka (right part of the table).
As shown in Table 1, the mean number of crops per farmer and the percentage of local/improved varieties were not significantly different between the Mwimbi and Tharaka in the midlands. The proportion of crop cultivated in the midlands by the Mwimbi and Tharaka differ for bean and green gram, while for both communities maize is the dominant crop species (about 29%). There was only small variation in the relative importance of crop species among seasons. The proportions of seeds saved by farmers from their own harvests or obtained from relatives or friends were similar between the Mwimbi and Tharaka (66.2% vs 69.5%). The proportion of seed obtained from the three local seed markets (one in each altitudinal region) differ slightly. Most of it (70% and 80%, respectively) comes from the lowland seed market. Almost all seed sold in the local markets is produced by individual farmers living in the area.
The average sowing dates for the Mwimbi and Tharaka living in the midlands were similar, except during the Short Rains in 2010 where it was 1 week earlier for the Mwimbi (mean date 11 October) than for the Tharaka (mean date 18 October). Differences in sowing dates between the two communities were significant for only 6 out of the 24 crop × season combinations tested (Fig. 3). The Tharaka sowed cowpea and maize later than the Mwimbi during the Long Rains 2010 and Short Rains 2009, respectively. Mean sowing dates were the same during the Long Rains 2011. During the Long Rains 2010, the Mwimbi sowed first bean, maize, and green gram, while the Tharaka sowed first sorghum, pearl millet, and maize. The sowing dates were on average 3.75 days later in the highlands than in the midlands.
Crop-sowing order per season comparing the Mwimbi and Tharaka. (a) Short Rains 2009 and 2010. (b) Long Rains 2010 and 2011. The date between brackets is the seasonal-mean sowing date computed for all crops. The horizontal axis represents the sowing dates per crop as compared to the seasonal mean. Crops sown first have a negative values and those sown last a positive values. The stars indicate when the difference of sowing dates between the Mwimbi and Tharaka are significantly different.
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
The statistical summary of rainfall variables (Table 2) shows how unpredictable these variables can be among seasons and years. In 2011, the Long Rains duration was 30% shorter than in 2010. In the midlands, the frequency of rainy days is likely to be higher during the Short Rains than during the Long Rains. The number of rainy days was also highly variable among seasons and years (e.g., 6 vs 16 days in the highlands during the Short Rains in 2009 and 2010). The spatial, seasonal, and interannual variability of rainfall increases the difficulty for farmers to predict the best sowing time.
Statistical summary of rainfall variables for the Short Rains (SR) and the Long Rains (LR) from 2009 to 2011. (a) Highland climatic zone. (b) Lowland climatic zone.
We have observed no contrasts in sowing strategies between the Mwimbi and Tharaka in the midlands. Both communities may start sowing before the beginning of the rainy season (dry sowing, well illustrated during the Short Rains 2009 and 2010; Fig. 4a) or after the beginning of the rainy season (wet sowing, well illustrated during the Long Rains 2010 and 2011; Fig. 4b). Figure 4b shows how risky was their common positive interpretation of an isolated first rainy event at the beginning of the Long Rains 2011. This contrasts with the Long Rains 2010, when both communities avoided sowing immediately after the early first rains. They preferred a prudent strategy, with highly scattered sowing dates.
Plots of the observed sowing dates (gray histogram) and rainfall distribution (black bars) comparing the Mwimbi and Tharaka in the midlands. (a) Sowing date and daily rainfall during the Short Rains in 2009 and 2010. (b) Sowing date and daily rainfall during the Long Rains in 2010 and 2011. Horizontal axis shows the first day of each month from January to May (Long Rains) or from September to December (Short Rains); the left vertical axis shows the frequency of sowing events; and the right vertical axis shows the daily rainfall amount.
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
b. Length of dry spells after seed germination and crop sensibility
The mean length of dry spells was significantly longer during the Long Rains (10.3 ± 0.12 days, n = 658) than during the Short Rains (7.6 ± 0.09 days, n = 1033), and in the midlands (9.0 ± 0.14 days, n = 916) than in the highlands (8.2 ± 0.13 days, n = 775). In the midlands, the mean length of dry spells after seed germination was not different between the Mwimbi and Tharaka, except for the Long Rains 2010, during which it was significantly longer for the Tharaka than for the Mwimbi (Table 3).
Mean length of dry spells after germination comparing the Mwimbi and Tharaka in the midlands. (a) Short Rains in 2009 and in 2010. (b) Long Rains in 2010 and in 2011. A t test shows that the mean number of consecutive dry day after seed germination was significantly greater for the Tharaka than for the Mwimbi during the Long Rains 2010 and not different for other seasons. Mean standard errors are SE and Pr is the p value. Significance codes: *** indicates <0.001 and NS indicates not significant.
We implemented a logistic regression to analyze the probability of sowing failure as a function of the length of dry spells. The model explained 56.7% of the null deviance (D2). Maize was more impacted than other crop species (Fig. 5). Model parameters suggest that while short dry spells (less than 5 days) impact maize during emergence, other crops are not impacted at all. The number of consecutive dry days needed to reach 50% loss is 8.3 days for maize, whereas it is 9.6 days for sorghum and pearl millet. When the dry spell is more than 10 days, all crop species are equally impacted.
Probability of sowing failure as a function of the mean length of dry spells after seed germination, comparing maize to other crop species. Model explained 56.7% of the null deviance. Each crop was assessed during the four climatic seasons at two elevations (8 dots per crop). Maize is significantly more impacted than other crops (p = 0.035).
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
c. How societies intervene between crops and climate
The highest proportion of sowing failure was recorded for the Long Rains with 65.8% ± 3.62% [0.95 confidence interval (c.i.)] as compared to 22.2% ± 2.53% (0.95 c.i.) for the Short Rains. Even though the Long Rains were associated with higher amounts of rainfall, the high level of sowing failure can be attributed to rainfall irregularities and the prolonged dry spells observed after seed germinations during this season. The proportion of plots where the crop failed was significantly higher (0.95 c.i.) in the midlands (42.9% ± 3.2%) than in the highlands (34.7% ± 3.35%).
1) One community, two environments: Controlling farmers’ social identity to assess the altitudinal effect
Figure 6 compares sowing failures of Mwimbi crop varieties in both the highland and lowland AEZ as a function of the length of the dry spell after sowing. The effect of prolonged dry spells (more than 8 days) appears more severe in the midlands than in the highlands; however, the difference was significant only for maize. The fact that maize is a highland crop and less well suited to the semiarid and subhumid conditions can explain its different response as compared to other crops. To confirm the null effect of agroecological zones on how other crops respond to the increasing length of dry spells, maize was removed from the model, and the logistic regression recomputed. The effect of agroecological zones on crop response remained nonsignificant (p = 0.38). The comparison of Fig. 6 curves not only shows a similarity in the response of Mwimbi crops at both altitudes, but also shows that all Mwimbi farmers interpret the rainfall signal in the same way for sowing.
Probability of sowing failure as a function of dry spell length, comparing Mwimbi plots in the midlands and highlands. The model explains 61.2% of null deviance. The effect of agroecological zones and altitude on crop response was not significant (p = 0.058).
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
2) One environment, two communities: Controlling altitude to assess the social effect
Controlling altitude, sowing failure was observed in almost all farmer plots during the Long Rains in 2011, following the risky strategy adopted by the Mwimbi and Tharaka (see section 3a). For all other seasons, the percentages of lost seeds were significantly higher for the Mwimbi than for the Tharaka (Table 4).
Proportion of plots where crop failed, comparing the Mwimbi and Tharaka at 950 m MSL during the Long Rains and the Short Rains. Binomial errors (SE) were reported in the table. Chi-squared test shows that the proportions of sowing failures were significantly greater for the Mwimbi than for the Tharaka. The p value is Pr and DF is degrees of freedom. Significance codes: *** indicates <0.001 and NS indicates not significant.
Comparing percentages of sowing failure in different seasons (by grouping the 2 yr), the difference (0.95 c.i.) between the Mwimbi and Tharaka in the midlands was as high as 25.9% ± 9.4% during the Long Rains and 24.97% ± 6.45% during the Short Rains. The relative risk of sowing failure was 3.3 times more for the Mwimbi than for the Tharaka during the Short Rains and 1.5 times more during the Long Rains.
Figure 7 provides a global comparison (across all seasons), showing that the probability of sowing failure differs significantly between the Tharaka and Mwimbi in the midlands. To confirm this effect of farmer community on crop response to the increasing length of dry spells, maize and bean were successively removed from the model and the logistic regression recomputed. The effect of farmer community on crop response remained highly significant. The seeds sown by the Tharaka failed less than those sown by the Mwimbi.
Proportion of sowing failure as a function of dry spell length, comparing the Mwimbi and Tharaka in the midlands. The model explains 69.9% of the null deviance. The probability of sowing failure is significantly higher for the Mwimbi than for the Tharaka (p = 0.001).
Citation: Weather, Climate, and Society 6, 3; 10.1175/WCAS-D-13-00034.1
4. Discussion
Our study confirmed the close relationship between rainfall distribution and farmers success in establishing cultivation. Thus, the rainfall variability and prolonged dry spells during the Long Rains explain the higher proportion of sowing failure as compared to the Short Rains. Out of the six crop species considered, maize, the dominant food crop for both the Mwimbi and Tharaka in the midlands, was significantly more impacted by dry spells. Considering globally all six crops, the odd ratio of sowing failure was 6.7 times more for the Long Rains than for the Short Rains. These results, based on 2 yr and four seasons, are clearly consistent with those obtained over a 46-yr period by Leclerc et al. (2014, manuscript submitted to Wea. Climate Soc.), where farmers’ oral report of seed losses was associated with climate variability of the Long Rains, with a stronger impact on maize variety losses. Altogether, our results underline the impact of rainfall variability and thence the crucial role of germplasm adaptation and sowing practices in crop success. Understanding the processes at work is essential in the present context of relatively rapid global climatic change, with the Mwimbi moving from the wet highlands to drier midlands and the Tharaka moving from the dry lowlands to wetter midlands.
Our results provide surprisingly strong elements on the importance of social factors in this adaptation. Indeed, the double comparative approach revealed that social factors have more impact in reducing the risk of sowing failures due to the rainfall variation than AEZs. The Mwimbi crop-sowing failures are more frequent in the midlands than in the highlands, as dry spells tend to be longer in the midlands (see distribution of data points in Fig. 6); however, the logistic model shows that the relationship between the length of dry spells and the subsequent sowing failure is the same at both altitudes, except for maize (section 3c). This shows a lack of differentiation between the midland and highland seeds used by Mwimbi farmers for the trait under study, that is, the capacity of Mwimbi crops to face dry spells in the emergence phase. In contrast, the logistic model shows a significant difference between the response of Mwimbi and Tharaka seeds in the same midland environment.
Thus, the logistic models have demonstrated that the variation in crop response to dry spells was lower among farmers within each community than between the two communities. Explanations of this social differentiation in sowing success can be sought in better-adapted farming practices and/or in better-adapted crop genetic resources. Indeed, the two communities differ in their experience with drought and in the origin of their genetic resources. Thus, the Tharaka, migrating from a drier environment, may have a better experience in managing rainfall variability and/or they may have developed more drought-resistant landraces, as compared to the Mwimbi, who come from a more humid environment. Alternatively, the Tharaka may have adapted by basing their farming system on a higher proportion of crops that are less susceptible to dry spells.
The farming system and relative importance of crops are comparable among individual farmers and between Mwimbi and Tharaka communities, even if the former favor beans while the latter favor green gram (section 3a; Table 1). Maize occupies the same dominant position (29%) among food crops for both communities in the midlands, so this drought-sensitive crop does not induce a difference between them when they face long dry spells. No appreciable difference was observed at the varietal level either. The total number of varieties per crop species was similar for both communities, as well as seed sources, with a dominance of seeds saved from previous harvest or supplied within the community and similar proportions of local versus improved varieties. The proportion of seeds obtained from market was slightly different between the Mwimbi and Tharaka (7.8% vs 15.2%), but both preferred the lowland market to supply seeds.
Even if the choice of the cropping system and sowing date can be used to cope with climate variability (Laux et al. 2010; Waha et al. 2013), individual practices did not appear to play a role either in the process of adaptation to climate change. Sowing strategies showed no significant differences between communities, except for the Short Rains season 2010, when the Mwimbi started sowing 1 week earlier. No community-specific sowing strategies were detected, neither when analyzing the relative timing of sowing different crops (section 3a; Fig. 3), nor when comparing the distribution of sowings and that of rains, with wet and dry sowings in both communities (Fig. 4). Thus, the way the Tharaka and Mwimbi interpret rainfall signals at the beginning of the rainy seasons and their choices of the sowing date were very similar. As a result, the mean length of dry spells after seed germination was not different between the Mwimbi and Tharaka in the midlands, except for the Long Rains 2010 during which they were significantly longer for the Tharaka than for the Mwimbi (section 3b, Table 3).
Despite the high similarity in crop management, at both specific and varietal levels, and in sowing strategy, we have observed a high contrast in sowing failure between the Mwimbi and Tharaka, with a relative risk of sowing failure that is 3.3 times more for the Mwimbi than for the Tharaka during the Short Rains and 1.5 times more during the Long Rains. Even in the Long Rains 2010, when dry spells were significantly longer for Tharaka farmers, the latter were considerably less impacted than Mwimbi farmers (30% vs 74% sowing failure). The only possible explanation of such a difference is that Tharaka genetic resources are much better adapted to long dry spells than those of the Mwimbi. This conclusion is not related to the importance of maize, as a dominant and drought-sensitive crop, as this social effect remained significant after removing maize from the logistic regression analysis.
The fact that the crops cultivated by the two communities in the midlands were differently impacted by dry spells can be related to the different origins of the crop genetic resources managed by the two communities. Indeed, the differential genetic adaptability to droughts of Mwimbi and Tharaka seeds is likely to correspond to their historical differentiation as communities of farmers. In a way, the history of the two communities gets reflected in the “genetic history” of their seeds. Tharaka seeds better endure drought conditions than those of the Mwimbi, probably because they originate from semiarid lowlands where droughts are usual.
The orientation of the seed exchange system must also be considered to explain why crop genetic adaptation to droughts differs between the two communities. Indeed, social barriers with limited intermarriage restrict seed exchanges, and hence the seed-mediated gene flow, between the two communities, favoring crop genetic differentiation. The centripetal orientation of seed exchange systems is related to the social relations that already exist out of the agricultural domain (Leclerc and Coppens d’Eeckenbrugge 2012). Indeed, farmers have to trust the supplier when exchanging information and seeds that are so important for their subsistence (Badstue et al. 2007). The fact that exchanged seeds are mainly obtained through trusted persons, members of the same family, the same village, or the same community, has been documented for maize in Mesoamerica (Badstue et al. 2006; Perales et al. 2005; Perales et al. 2003; Van Etten 2006), Andean tubers in Peru (Zimmerer 2003), sorghum in Ethiopia (McGuire 2008), and rice in Gambia (Nuijten and Almekinders 2008) and Sierra Leone (Longley 2000). Where several ethnic groups live in the same village, seed exchanges are preferentially (up to 90%) concluded with members of the same ethnic group (Almekinders et al. 1994; Delaunay et al. 2008). Reversely, seeds are rarely supplied by outsiders. In the cases studied by Badstue et al. (2006), only 1% of the seeds come from such sources. Considering these studies from different countries all together, most seeds (52% to 99%) are produced on the farm, and those obtained from outside mostly come from within the community of the farmer.
In the context of this study, differential genetic adaptation to droughts of Mwimbi and Tharaka seeds was probably maintained over time first because seed exchange systems have been oriented more within than between the two communities. Second, the cultivated fields are scattered, which limits pollen-mediated gene flow and also contributes to maintaining genetic differentiation. Another possible explanation of why the Tharaka show lower sowing failure than the Mwimbi could be that the Tharaka have got a higher percentage of seeds from the market than the Mwimbi (see section 3a; Table 1), mostly for maize and bean. However, removing maize and bean from the analysis did not affect our result. The effect of the farmer community on crop response remained highly significant.
Seed genetic differentiation is thus favored both by the low level of seed exchanges between communities and the high level of exchanges within the community. So far, such a factor was not well considered in crop genetic studies, and sampling strategies rarely allowed the consideration of both environmental and social variations. When social factors were taken into account, significant progress was achieved in our understanding of social and biological processes, as in the study of Deu et al. (2008) or those of Brush and Perales (2007) and Benz et al. (2007).
In the present case, contacts and relations favored by intermarriage directly foster seed exchanges between the midland and lowland Tharaka. The seed system in this case is oriented to the lowlands and drought tolerance. On the contrary, the social relations and intermarriage of the Mwimbi favor moving seed from the highlands to midlands, and the seed system is in this case oriented toward the highlands and drought susceptibility. The effect of this within-community gene flow may be negligible when farmers essentially use their own seeds for the next season, maintaining them in the same environment, but it can be much stronger in the case of seed loss. Then, renewing seeds with genetic resources originating from the highlands implies decreasing adaptability to droughts at the lower altitude for the Mwimbi, whereas renewing seeds with genetic resources originating from the lowlands implies increasing adaptability to droughts at the higher altitude for the Tharaka (notably with seeds obtained from lowlands markets). The differential adaptability to drought of crop managed by the two communities is thus maximized in the midlands.
Additional studies are needed to confirm that the rate of seeds that are annually self-produced within a community is a key factor in their adaptation to local conditions. Other crop traits could be taken into account, beyond the tolerance to dry spells in the emergence phase, to assess crop adaptability and economic potential in contrasted social and environmental settings. Crop genetic populations that have been managed within the community over time can be assimilated as reservoirs of adaptability. Each farmer annually selects the best plants for his/her seeds and transfers their genetic adaptability to the next season crop. The social dimension of the process is associated with farmer-to-farmer seed exchanges, which occur more within than between communities. As such, farmer communities and crop communities operate hand in hand.
5. Conclusions
Our double comparative approach fully supported our working hypothesis, even showing that not only the process of adaptation to climate change involves a social component, but that this component may be more important in explaining the differential success of farmer social groups than environmental variation itself.
The migration of Mwimbi farmers along the slope of Mount Kenya from the wet highlands to drier midlands, and that of the Tharaka from the dry lowlands to wetter midlands, provided a favorable context, with a useful contrast allowing the isolation of this social component in the adaptation to rapid environmental changes. Both communities are today geographically close to each other in the midlands, but their experience with drought and the origin of their genetic resources differ.
The impact of rainfall variations and droughts on sowing failure is not uniform. Societies intervene between crops and climate, and the social process of adaptation to climate variability involves farmers’ practices and knowledge, as well as the genetic resources that they historically manage. Crop adaptation to drought, and more generally to climate changes, should thus be not only a biological, but also a social process. Further research needs to implement the genetic analysis of the crop varieties together with an ethnographic approach of farmers’ practices in order to identify traits (or practices) that allow crop varieties adapting to the local environment.
The results obtained from this study highlight some needs for future breeding programs. Most critically, such programs should consider crop genetic adaptability together with the history and sociology of farmer communities. This must be reflected more accurately in crop genetic sampling strategies used to collect new materials and to improve crop adaptation to climate variability.
Acknowledgments
This study is a contribution to the project entitled “Predictability of the climatic information for reducing tropical agriculture vulnerability” (PICREVAT), funded by the French National Research Agency (ANR 08-VULN-01-008). We thank Geo Coppens d’Eeckenbrugge (CIRAD, UMR CEFE, France), Pierre Camberlin (CNRS, Biogeosciences, CRC, France), Françoise Dosba (Montpellier SupAgro, UMR AGAP, France), and anonymous reviewers for their useful comments on the manuscript.
REFERENCES
Agresti, A., 2007: An Introduction to Categorical Data Analysis. Wiley, 400 pp.
Almekinders, C. J. M., Louwaars N. P. , and de Bruijn G. H. , 1994: Local seed systems and their importance for an improved seed supply in developing countries. Euphytica, 78, 207–216, doi:10.1007/BF00027519.
Altieri, M. A., 1994: Biodiversity and Pest Management in Agroecosystems. Food Products Press, 185 pp.
Araya, A., Keesstra S. D. , and Stroosnijder L. , 2010: A new agro-climatic classification for crop suitability zoning in northern semi-arid Ethiopia. Agric. For. Meteor., 150, 1057–1064, doi:10.1016/j.agrformet.2010.04.003.
Badstue, L. B., Bellon M. R. , Berthaud J. , Juárez X. , Rosas I. M. , Solano A. M. , and Ramírez A. , 2006: Examining the role of collective action in an informal seed system: A case study from the Central Valleys of Oaxaca, Mexico. Hum. Ecol., 34, 249–273, doi:10.1007/s10745-006-9016-2.
Badstue, L. B., Bellon M. R. , Berthaud J. , Ramírez A. , Flores D. , and Juárez X. , 2007: The dynamics of farmers’ maize seed supply practices in the Central Valleys of Oaxaca, Mexico. World Dev., 35, 1579–1593, doi:10.1016/j.worlddev.2006.05.023.
Barron, J., Rockstrom J. , Gichuki F. , and Hatibu N. , 2003: Dry spell analysis and maize yields for two semi-arid locations in East Africa. Agric. For. Meteor., 117, 23–37, doi:10.1016/S0168-1923(03)00037-6.
Barth, F., Ed., 1969: Ethnic Groups and Boundaries: The Social Organization of Culture Difference. Allen and Unwin, 153 pp.
Benz, B. F., Perales H. R. , and Brush S. B. , 2007: Tzeltal and Tzotzil farmer knowledge and maize diversity in Chiapas, Mexico. Curr. Anthropol., 48, 289–300, doi:10.1086/512986.
Bernard, F. E., 1972: East of Mount Kenya: Meru Agriculture in Transition. Weltforum Verlag, 176 pp.
Brush, S. B., and Perales H. R. , 2007: A maize landscape: Ethnicity and agro-biodiversity in Chiapas, Mexico. Agric. Ecosyst. Environ., 121, 211–221, doi:10.1016/j.agee.2006.12.018.
Camberlin, P., Boyard-Micheau J. , Philippon N. , Baron C. , Leclerc C. , and Mwongera C. , 2012: Climatic gradients along the windward slopes of Mount Kenya and their implication for crop risks. Part 1: Climate variability. Int. J. Climatol., doi:10.1002/joc.3427, in press.
Delaunay, S., Tescar R.-P. , Oualbego A. , Vom Brocke K. , and Lancon J. , 2008: La culture du coton ne bouleverse pas les échanges traditionnels de semences de sorghos. Cah. Agric., 17, 189–194.
Deu, M., and Coauthors, 2008: Niger-wide assessment of in situ sorghum genetic diversity with microsatellite markers. Theor. Appl. Genet., 116, 903–913, doi:10.1007/s00122-008-0721-7.
Fadiman, J. A., 1982: An Oral History of Tribal Warfare: The Meru of Mt. Kenya. Ohio University Press, 185 pp.
Fadiman, J. A., 1993: When We Began, There Were Witchmen: An Oral History from Mount Kenya. University of California Press, 395 pp.
Funk, C., Dettinger M. D. , Michaelsen J. C. , Verdin J. P. , Brown M. E. , Barlow M. , and Hoell A. , 2008: Warming of the Indian Ocean threatens eastern and southern African food security but could be mitigated by agricultural development. Proc. Natl. Acad. Sci. USA, 105, 11 081–11 086, doi:10.1073/pnas.0708196105.
Hammer, K., and Teklu Y. , 2008: Plant genetic resources: Selected issues from genetic erosion to genetic engineering. J. Agric. Rural Dev. Trop. Subtrop., 109, 15–50.
Heine, B., and Moehlig W. J. G. , Eds., 1980: Geographical and Historical Introduction: Language and Society. Vol. 1, Language and Dialect Atlas of Kenya, Reimer, 10 pp.
Jaetzold, R., Schmidt H. , Hornetz B. , and Shisanya C. , 2007: Natural Conditions and Farm Management Information. Vol. 2, Farm Management Handbook of Kenya, 2nd ed., Ministry of Agriculture, 411 pp.
Kotschi, J., 2007: Agricultural biodiversity is essential for adapting to climate change. Ecol. Perspect. Sci. Soc., 16, 98–101.
Laux, P., Jäckel G. , Tingem M. R. , and Munstmann H. , 2010: Impact of climate change on agricultural productivity under rainfed conditions in Cameroon—A method to improve attainable crop yields by planting date adaptations. Agric. For. Meteor., 150, 1258–1271, doi:10.1016/j.agrformet.2010.05.008.
Leclerc, C., and Coppens d’Eeckenbrugge G. , 2012: Social organization of crop genetic diversity. The G × E × S interaction model. Diversity, 4, 1–32, doi:10.3390/d4010001.
Leclerc, C., Mwongera C. , Camberlin P. , and Moron V. , 2014: Cropping system dynamics, climate variability, and seed losses among East African smallholder farmers: A retrospective survey. Wea. Climate Soc., submitted.
Lin, B. B., 2011: Resilience in agriculture through crop diversification: Adaptive management for environmental change. Bioscience, 61, 183–193, doi:10.1525/bio.2011.61.3.4.
Longley, C., 2000: A social life of seeds: Local management of crop variability in North-Western Sierra Leone. Ph.D thesis, University of London, 309 pp.
Maracchi, G., Bacci L. , Cantini C. , and Haimanot M. T. , 1993: Effects of water deficit in vegetative and post-flowering phases on pearl millet grown in controlled environment. Agric. Mediterr., 123, 65–71.
Marteau, R., Sultan B. , Moron V. , Alhassane A. , Baron C. , and Traor S. B. , 2011: The onset of the rainy season and farmers sowing strategy for pearl millet cultivation in southwest Niger. Agric. For. Meteor., 151, 1356–1369, doi:10.1016/j.agrformet.2011.05.018.
McCarthy, J., Canziani O. F. , Leary N. A. , Dokken D. J. , and White K. S. , Eds., 2001: Climate Change 2001: Impacts, Adaptation, and Vulnerability. Cambridge University Press, 1032 pp.
McGuire, S. J., 2008: Securing access to seed: Social relations and sorghum seed exchange in eastern Ethiopia. Human Ecol., 36, 217–229, doi:10.1007/s10745-007-9143-4.
Moehlig, W. J. G., 1974: Die Stellung der Bergdialekte im Osten des Mt. Kenya: Ein Beitr z. Sprachgliederung im Bantu. Reimer, 249 pp.
Nuijten, E., and Almekinders C. J. M. , 2008: Mechanisms explaining variety naming by farmers and name consistency of rice varieties in the Gambia. Econ. Bot., 62, 148–160, doi:10.1007/s12231-008-9012-0.
Peatrik, A.-M., 1999: La vie pas Contés. Société d’ethnologie, 571 pp.
Perales, H. R., Brush S. B. , and Qualset C. O. , 2003: Dynamic management of maize landraces in central Mexico. Econ. Bot., 57, 21–34, doi:10.1663/0013-0001(2003)057[0021:DMOMLI]2.0.CO;2.
Perales, H. R., Benz B. F. , and Brush S. B. , 2005: Maize diversity and ethnolinguistic diversity in Chiapas, Mexico. Proc. Natl. Acad. Sci. USA, 102, 949–954, doi:10.1073/pnas.0408701102.
Pressoir, G., and Berthaud J. , 2004a: Patterns of population structure in maize landraces from the Central Valley of Oaxaca in Mexico. Heredity, 92, 88–94, doi:10.1038/sj.hdy.6800387.
Pressoir, G., and Berthaud J. , 2004b: Population structure and strong divergent selection shape phenotypic diversification in maize landraces. Heredity, 92, 95–101, doi:10.1038/sj.hdy.6800388.
R Core Team, 2013: R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. R Foundation for Statistical Computing. [Available online at http://www.R-project.org/.]
Segele, Z. T., and Lamb P. J. , 2005: Characterization and variability of Kiremt rainy season over Ethiopia. Meteor. Atmos. Phys., 89, 153–180, doi:10.1007/s00703-005-0127-x.
Sgrò, C. M., Lowe A. J. , and Hoffmann A. A. , 2011: Building evolutionary resilience for conserving biodiversity under climate change. Evol. Appl., 4, 326–337, doi:10.1111/j.1752-4571.2010.00157.x.
Tonhasca, A., and Byrne D. N. , 1994: The effects of crop diversification on herbivorous insects: A meta-analysis approach. Ecol. Entomol., 19, 239–244, doi:10.1111/j.1365-2311.1994.tb00415.x.
Vandermeer, J. H., 1989: The Ecology of Intercropping. Cambridge University Press, 237 pp.
Van Etten, J., 2006: Seeds, hands and lands: Maize genetic resources in highland Guatemala in space and time. Ph.D, Wageningen University, 132 pp.
Waha, K., Müller C. , Bondeau A. , Dietrich J. P. , Kurukulasuriya P. , Heinke J. , and Lotze-Campen H. , 2013: Adaptation to climate change through the choice of cropping system. Global Environ. Change, 23, 130–143, doi:10.1016/j.gloenvcha.2012.11.001.
Wood, D., and Leene J. M. , 1997: The conservation of agrobiodiversity on-farm: Questioning the emerging paradigm. Biodiversity Conserv., 6, 109–129, doi:10.1023/A:1018331800939.
Zimmerer, K. S., 2003: Geographies of seed networks for food plants (Potato, Ulluco) and approaches to agrobiodiversity conservation in the Andean countries. Soc. Nat. Resour., 16, 583–601, doi:10.1080/08941920309185.
