Predictive value of soil moisture and concurrent variables in the multivariate modelling of cereal yields in water-limited environments.

Rainfed cereal yields show high variability depending on the varying conditions of concurrent factors during the crop year. Among them, hydrometeorological factors such as maximum temperature, rainfall, and notably, soil moisture, strongly affect crop production, but the greatest source of uncertainty on yield estimates stems from their interaction. This is of special interest in water-limited regions where climate change is expected to affect more intensely, but also in others where water is increasingly limited. Despite the highly non-linear nature of the interactions, simple statistic models such as multilinear regression accurately explore a notable proportion of the variability of cereal yields. To describe the impacts behind interactions, we perform stepwise multilinear regression of meteorological factors derived from E-OBSv23 database and soil moisture from ERA5-Land against annual wheat and barley yields for the period 1981–2019 in the main cereal regions of Spain. The multivariate approach characterizes the temporal shifts of factors' influence. Beyond the temporal shifts on the synchrony of the factors, some of them tend to co-dominate the impact during the critical period of crop development, with soil moisture exceeding all others in relevance. Multivariate analysis fosters discussion about the impact of the choice of variables on the model fit, as well as on the pertinence of monthly and annual scales for explorative and predictive purposes. Monthly models perform particularly well during the critical period of growth and reproduction of crops and consistently better than univariate estimates. The annual model built using the data of the months of maximum impact of key variables outperforms the model at a monthly scale, which underlines the decisive role of the critical period. Similarly, results highlight the worth of parsimony in modelling. Soil moisture stands out as the principal concurrent variable to improve yield estimates from environmental data, which governs yields of rainfed water-limited croplands. • Soil moisture stands out from environmental factors on influence on cereal yields. • Stepwise multilinear regression of less colinear factors largely explains yield variability. • Using key variables in the critical period of crop sensitivity improves model fit. • Soil moisture notably improves model estimates both at monthly and yearly scales. [ABSTRACT FROM AUTHOR]