Assessment of the relations between crop yield variability and the onset and intensity of the West African monsoon

Abstract

<p>In rain-fed systems, efficient and timely crop planning is crucial to maximize crop production, adapt to climate variability, and increase the sustainability and resilience of the production systems. Smallholder farmers plan and anticipate possible interventions during the season based on the actual onset of the monsoon. However, their knowledge to define and predict the monsoon onset is limited to traditional methods whose predictive skill decreases significantly with a recent increase in both temperature and rainfall variability in the region. Therefore, defining the start of the monsoon accurately is a priority for improving crop production in rain-fed systems. Since the 1970s, researchers have produced more than 18 definitions—from local to regional scale—to define the start of the monsoon in the Sahel region which makes it difficult for one to find a suitable definition for a specific application. The present study compared and analyzed the West African Monsoon (WAM) onset according to Raman’s, Stern’s, Yamada’s, and Liebman’s definitions using station data from 13 locations in Senegal i.e. Dakar, Louga, Matam, St. Louis, Thies, Diourbel, Fatick, Kaffrine, Kaolack, Kedougou, Kolda, Tambacounda, and Ziguinchor from 1981 to 2020. To this end, we applied machine learning algorithms—<em>K-means clustering and Decision Tree</em>—to cluster the Sea Surface Temperature anomalies (SSTa) obtained from different regions of the Mediterranean and the Atlantic Ocean. We then used the clusters in the decision tree model to predict the onset and intensity of seasonal rainfall in the study locations according to the four definitions. Subsequently, we applied the set of the generated onset dates according to the four definitions as sowing dates in simulations of maize growth and yields using the  Agricultural Production Systems sIMulator (APSIM). Our analysis showed a statistically significant difference between the onset dates defined by the four definitions. Raman’s and Stern’s definitions delayed the monsoon onset at least two to four weeks after 1st June while Yamada’s and Liebman’s definitions delayed the onset one to two weeks after 1st June. Moreover, the amounts of seasonal rainfall in the season defined by Raman’s and Stern’s definitions were on average lower and more variable compared to those defined by Yamada’s and Liebman’s definitions. Similarly, we found statistically significant differences between the means of simulated maize yields in the four sets of sowing dates used. The highest yields with the lowest interannual variability were found in Yamada followed by Liebman’s sowing dates. The other sets of sowing dates had very low yields and higher variability compared to Yamada’s and Liebman’s sowing dates. We found the SSTa from the Southern Atlantic Ocean, Mediterranean Sea, and Tropical Atlantic Ocean regions as good predictors of both onset dates and intensity of the monsoon. The accuracy ranged from 50% to 80% depending on the location. </p>