Abstract
<jats:p>Evidence of global warming is clearly reflected in extreme daily maximum temperature events, particularly when record-breaking temperatures occur. In the Iberian Peninsula, previous studies showed that the frequency of such records exhibits a non-stationary behaviour, with an upward trend and strong spatial variability. In this work, we build a statistical model to explore and interpret the spatial heterogeneity and spatio-temporal structure of this phenomenon.Daily maximum temperature (Tx) series from 36 meteorological stations across Spain covering the period 1960-2023 were obtained from the European Climate Assessment & Dataset. Predictor variables were obtained from ERA5 reanalysis data, consisting of geopotentials at 300, 500, and 700 hPa at 12:00 UTC, covering a 1º x 1º grid spanning [45º N, 10º W, 35º S, 5º E]. The analysis was restricted to the summer season (JJA).A previously developed modelling framework was applied to obtain station-specific logistic regression models, as well as global models. The response variable was defined as a binary indicator of extreme heat event (EHE) occurrence. For each station s, the event threshold was defined as the 95th percentile of Tx in the reference period 1981-2010, computed over summer days only. Formally, the threshold is given by us = Q0.95 (Tx,t,l t ∈ [1981, 2010], l ∈ [1, 92]) where Tx,t,l denotes the daily maximum temperature at station s on day l of year t. An EHE is then defined through the indicator Ix,t,l = 1 if Tx,t,l > us , and 0 otherwise, with value 1 indicating the occurrence of an EHE.The modelling strategy was carried out in three steps: (1) stepwise logistic regression was performed independently at each station to identify relevant predictors; (2) the most frequently selected and influential variables across stations were used to construct a global model; and (3) three extended models were developed by incorporating interactions with geodesic, climatic, and spatial covariates, followed again by stepwise selection. The first 51 years of the period were used for building the models and the final 13 years were reserved for validation. Due to class imbalance, model performance was evaluated using the AUC measure.The best results were obtained from the global model including climatic interactions, which reached an AUC of 0.89 with k = 34 parameters and was therefore selected as the winning model. The individual geopotential terms of this model were analysed to better understand the climatic characteristics associated with EHEs. It was also used to simulate EHEs over the validation period. The simulated EHE sequences were compared with the observations in order to evaluate the model’s ability to reproduce consecutive-day heatwave dynamics. In addition, the model’s assigned probabilities of EHE occurrence were assessed during selected heatwave episodes in the validation period.</jats:p>