Modelling developmental stage-specific climate effects and extremes on wheat yield under global change

Wheat is a key staple crop in temperate regions where projected increases in temperature variability, drought frequency and extreme weather events pose a significant threat to yield stability. Understanding how weather variability, climate extremes and, importantly, their timing in relation to crop development affect yield is essential for modelling agricultural systems in the context of climate change. Increasing climate variability and compound stress events challenge static or season-averaged approaches that fail to resolve developmentally sensitive stress periods.

However, it is often difficult to disentangle these effects across phenological stages due to incomplete or inconsistent phenological observations, particularly in long-term, multi-site datasets. This study presents a proxy-based modelling framework that quantifies yield–environment relationships while explicitly accounting for sensitivities specific to developmental phases.

We analyzed a multi-year historical dataset spanning multiple locations, combining wheat yield records with weather and environmental data extracted from the Spartacus dataset by GeoSphere Austria. Our primary objectives were to: (i) quantify the influence of weather and environmental variables on wheat yield; (ii) identify the most relevant stress factors within these environments; and (iii) assess how stress impacts vary across developmental phases. As phenological scoring data were incomplete across years and locations, we used Growing Degree Days (GDD) as a biologically motivated proxy to reconstruct crop developmental timing.

Based on the reconstructed developmental phases, raw weather and soil data were transformed into explanatory envirotyping variables at multiple temporal scales, including annual aggregates, phase-specific windows and daily to weekly resolutions. Explicit stress indicators were derived for thermal, hydrological and precipitation extremes, including compound and duration-based events, described using both frequency and intensity metrics. These were described using both frequency and intensity metrics. Such temporally explicit stress characterization is essential in the context of global change, given the projected increases in heatwaves, drought duration and compound extremes, which are expected to amplify phase-specific yield sensitivities.

Yield responses were modelled using a combination of linear models and machine learning approaches to capture non-linear effects and interactions. This framework enables the identification of critical developmental windows, the quantification of stress sensitivities and the assessment of environmental similarity and transferability across sites and years. Overall, this scalable envirotyping framework enables the identification of critical developmental windows and time-specific environmental drivers of yield variation, supporting more robust crop modelling and breeding decisions under increasing climate variability.