Climate variability, encompassing changes in climate statistics beyond individual weather events, poses a significant challenge to agriculture. Farmers often make crucial decisions based on anticipated climatic conditions, making them vulnerable to unexpected weather patterns. Studies have shown that temperature and precipitation during the growing season can explain a substantial portion (up to 60%) of year-to-year yield fluctuations for major crops globally. In SSA, a region facing severe food security risks by 2050, crop production may be more sensitive to temperature changes than precipitation changes, although this varies regionally. The traditional understanding of static climate-crop yield relationships is becoming obsolete due to amplified climate fluctuations. To enhance agricultural resilience, leveraging readily available climate and weather data to create relevant decision-making variables is crucial for implementing climate-smart agriculture. However, the complex, nonlinear, and threshold-based response of crop yields to weather poses challenges. Existing approaches, including data-driven statistical models and process-based crop growth models, have limitations such as assumptions of time-invariant relationships, limited focus on interannual variability, univariate analysis, and use of coarse spatio-temporal indicators. This study aims to overcome these limitations by exploring the influence of sub-seasonal wet-dry spell patterns on maize yield variance in SSA, utilizing an approach suitable for large spatio-temporal scales and limited data availability—a key asset for developing regions.

Existing research demonstrates that climate variability significantly influences crop yields, with temperature and precipitation accounting for a considerable portion of year-to-year fluctuations. However, the relative importance of temperature versus precipitation varies across regions and crops. In SSA, the impact of climate change on maize yield has been a subject of debate, with some studies suggesting greater sensitivity to temperature changes, while others emphasize the regional variation in response to both temperature and precipitation. The assumption of static relationships between climate and yield is increasingly challenged as global warming and human interventions amplify natural system fluctuations. Efforts to improve agricultural decision-making through climate services and climate-smart agriculture highlight the need for more sophisticated methodologies that consider the complex, nonlinear, and threshold-type responses of crops to weather. Previous studies have limitations regarding their assumptions of time-invariant relationships between climate drivers and crop yields, inadequate attention to interannual climate variability, univariate analysis of climate-yield relationships, and the use of coarse spatio-temporal indicators. This research addresses these limitations by examining the influence of sub-seasonal dry-wet spell patterns on maize yield variability in SSA.

The study utilized a multi-step methodology to analyze the impact of sub-seasonal dry-wet spell patterns on maize yield variability in SSA. Firstly, maize yield across SSA and related trends (1982–2009) were characterized using the Global Dataset of Historical Yield (GDHY). Secondly, changes in mean precipitation and temperature, and the prevalence and distribution of intra-growing season dry-wet spells during the same period (1982–2009) were explored using the Agriculture Modern-Era Retrospective Analysis for Research and Applications (AgMERRA) climate dataset. Thirdly, the study identified the most suitable subset of dry-wet spell patterns to explain maize production variability and its spatial heterogeneity. This involved the computation of the Standardized Precipitation Evapotranspiration Index (SPEI) at a dekadal (10-day) timescale. SPEI values were classified into categories (very dry, dry, moderately dry, moderately wet, wet, very wet) based on quantiles. A random forest (RF) algorithm was used to determine the relative contribution of temperature versus precipitation to the onset of dry spells for each class and developmental stage. A Bayesian network (BN) was employed to assess the likelihood of concurrent dry-wet spells at different crop growth stages and their potential impacts on yields. Trends in mean precipitation, temperature, and dry-wet spell incidence were detected using Sen's slope and the Mann-Kendall test. Finally, multiple linear regressions were performed to explain maize yield variance using two different sets of predictors: changes in mean precipitation and temperature at four crop development stages and changes in the likelihood of combined dry/wet spells across these stages. The study considered the spatial heterogeneity of crop responses and integrated the time specificity of weather impacts on yields and the existence of diverse agroclimatic zones within SSA. The SPEI was computed using the Penman-Monteith equations for evapotranspiration, and the generalized extreme value distribution was used to fit the aggregated effective precipitation time series. A penalty matrix based on the Needleman-Wunsch algorithm was used to assess the similarity of SPEI class sequences to reference sequences (representing different dry-wet spell patterns). A risk matrix framework combined the likelihood and severity of dry-wet spell patterns to generate regional crop risk maps.

The study revealed significant spatial and temporal heterogeneity in maize yield, mean precipitation, temperature, and dry-wet spell incidence across SSA from 1982 to 2009. Maize yields were generally low across the region, with higher interannual variability in semi-arid zones. Positive yield trends were observed in parts of East Africa and West Africa, while negative trends were noted in some areas of East Africa and Southern Africa. Rainfall trends varied across the region and growing season stages. A positive trend was observed in most of Southern Africa and the Sahel, while negative trends were found in some parts of the Gulf of Guinea and Sierra Leone. Mean temperature showed a positive trend across most of SSA. The relative contribution of temperature to the onset of dry spells was higher at the initial growth stage, increasing with event intensity. Changes in mean temperature and precipitation explained 30-35% of maize yield variability, while shifts in dry-wet spell patterns explained 50-60%. Four key dry-wet spell patterns were identified as major drivers of yield variability: prevalence of dry or wet spells across all growth stages and contrasting conditions during the initial stage compared to the rest of the season. Risk associated with prevalent dry spells generally declined across SSA during the reference period, while risk related to prevalent wet spells exhibited regional heterogeneity. Net production losses due to prevalent dry spells were substantial, impacting a significant portion of the study area. The study quantified these losses in terms of energy requirements, highlighting their implications for food security.

This study's findings underscore the limitations of assuming static weather-crop yield relationships. The observed changes in climatic conditions and their impact on maize yields were spatially heterogeneous, reflecting the diversity of agroclimatic zones within SSA. The positive rainfall trend in Southern Africa and the Sahel aligns with observations of regional greening. The negative rainfall trend in some parts of the Gulf of Guinea is consistent with trends towards less frequent but more intense rainfall. The study's finding that dry-wet spell patterns explain a larger proportion of yield variability than mean climatic variables supports the growing recognition of the importance of extreme weather events in influencing crop production. The identification of four key dry-wet spell patterns driving maize yield variability provides valuable insights for targeted interventions aimed at improving agricultural resilience. The observed decline in risk associated with prevalent dry spells suggests potential adaptation and mitigation efforts might be producing positive impacts. However, the regional heterogeneity in the impact of wet spell patterns highlights the need for regionally tailored strategies.

This study demonstrates the critical role of intra-growing season dry-wet spell patterns in driving maize yield variability in SSA. The analysis highlights the inadequacy of using mean climate variables alone for predicting yield and emphasizes the need to incorporate sub-seasonal climate variability. The findings contribute to a better understanding of climate-crop yield relationships in data-scarce regions and can inform climate-smart agriculture strategies. Future research should explore the impact of other factors, refine the methodology by using additional climate datasets and combining it with crop model outputs, and examine the effects of shifting planting dates as an adaptation strategy.

The study acknowledges several limitations. The reliance on fixed crop calendars may not fully capture the impact of adaptive planting date adjustments. The analysis does not incorporate the influence of biotic (e.g., pests) and abiotic (e.g., agricultural inputs, management practices) factors. The quality of available agricultural statistics and climatic forcing data introduce uncertainty, particularly in regions with sparse in-situ observations. The use of linear regression models may not fully capture nonlinear relationships. Future research can address these limitations by incorporating time-varying planting dates, considering other factors influencing yields, improving the quality of input data, and exploring alternative modeling approaches.