The impact of abnormal temperatures on crop yields in Italy: a functional quantile regression approach

TL;DR

The paper addresses how intra-season temperature and precipitation anomalies affect crop yields in Italy by introducing a scalar-on-function regression framework that treats weather covariates as functions over the growing season. Using provincial data for maize and soft wheat from 1952–2023, it employs functional principal component analysis to reduce dimensionality and estimates time-varying effects via OLS and functional quantile regression, enabling analysis of distributional tails. Key findings show that high temperatures reduce maize yields mainly in June–August, with milder effects in April and October, while soft wheat is negatively impacted by heat in late March to early April; precipitation benefits early wheat yields but harms later stages. The results provide precise timing for climate adaptation, such as targeted irrigation, and demonstrate the method’s strong predictive performance and suitability for tail-risk assessment in climate-sensitive agriculture.

Abstract

In this study, we apply functional regression analysis to identify the specific within-season periods during which temperature and precipitation anomalies most affect crop yields. Using provincial data for Italy from 1952 to 2023, we analyze two major cereals, maize and soft wheat, and quantify how abnormal weather conditions influence yields across the growing cycle. Unlike traditional statistical yield models, which assume additive temperature effects over the season, our approach is capable of capturing the timing and functional shape of weather impacts. In particular, the results show that above-average temperatures reduce maize yields primarily between June and August, while exerting a mild positive effect in April and October. For soft wheat, unusually high temperatures negatively affect yields from late March to early April. Precipitation also exerts season-dependent effects, improving wheat yields early in the season but reducing them later on. These findings highlight the importance of accounting for intra-seasonal weather patterns to provide insights for climate change adaptation strategies, including the timely adjustment of key crop management inputs.

Figures (5)

• Figure 1: Figures illustrating the yield series (kg/ha) for each province, which are presented over the period from 1952 to 2023. Specifically, panel (a) depicts the temporal evolution of yields for maize, while panel (b) shows the corresponding evolution for wheat.
• Figure 2: The figure presents the spatial distribution of yield values (kg/ha) across provinces, highlighting regional variations in productivity. Specifically, panel (a) displays maize yields for the year 1990, while panel (b) shows the corresponding maize yields for 2021. Provinces depicted in white on each panel do not have available production data for that particular year.
• Figure 3: The figure presents the spatial distribution of yield values (kg/ha) across provinces, highlighting regional variations in productivity. Specifically, panel (a) displays wheat yields for the year 1990, while panel (b) shows the corresponding wheat yields for 2021. Provinces depicted in white on each panel do not have available production data for that particular year.
• Figure 4: Estimates of the temperature functional coefficient with 95% confidence bands. The standard OLS regression coefficient is shown in blue, while the quantile regression coefficients for $\tau = 0.1$ are shown in red and those for $\tau = 0.9$ are shown in green. As regards the quantile regression estimators alone, the estimate using simple QR is represented by a solid line, while that relating to the QA estimator is represented by a dashed line. For the QA approach the neighbourhood considered is $\mathcal{N}(\tau) = \{\tau\pm 0.025,\tau\pm 0.05\}$.
• Figure 5: Estimates of the functional coefficients of temperature (panel (a)) and precipitation (panel (b)) with 95% confidence bands. The standard OLS regression coefficient is shown in blue, while the quantile regression coefficients for $\tau = 0.1$ are shown in red and those for $\tau = 0.9$ are shown in green. As regards the quantile regression estimators alone, the estimate using simple QR is represented by a solid line, while that relating to the QA estimator is represented by a dashed line. For the QA approach the neighbourhood considered is $\mathcal{N}(\tau) = \{\tau\pm 0.025,\tau\pm 0.05\}$.