David nous présentera son travail avec Claudia Mendes sur " Application of causal inference to estimate effect of extreme weather on crop yields".
Abstract : Impacts of extreme weather events (EWEs) on crop yields are commonly estimated using yield time series obtained from surveys. As these data do not come from randomized controlled trials, they are subject to confounding factors and may lead to biased estimates when analyzed using standard statistical methods. Recently, causal analysis methods have been proposed to take into account the risks of confounding effects, but they have never been compared in an agronomic context. Here, we used a dataset including 23 years of yield data in France to estimate the impact of two important types of weather events — drought events in summer (DE) and cold wave events in spring (CE) — on the yields of maize and sunflower, two crops potentially sensitive to these adverse
weather conditions. We applied and compared several causal analysis methods, namely Inverse Probability Weighting (IPW), Matching (Match), Standardization (SDZ), and Double Robust (DR) using both linear mixed-effects models and gradient boosting machine learning models.