From Linear Clustering to Deep Learning: Assessing Weather Regimes’ Impacts on Winter Extreme Temperatures over Northwestern Africa with a Focus on Morocco.

This work investigates the connection between large-scale atmospheric dynamics in the North Atlantic and winter temperature variability by analyzing the contribution of weather regimes to the occurrence of daytime and nighttime cold and warm events in Northwest Africa, focusing on Morocco. 

Weather regimes are first identified using a conventional circulation-based framework relying on k-means clustering of geopotential height anomalies. The sensitivity of the inferred circulation–temperature relationships to the choice of regime identification method is then investigated by comparing classical geopotential-based regimes with classifications incorporating jet-stream information and with non-linear regimes derived from variational autoencoders. This analysis is intended to evaluate the robustness and impact relevance of weather regimes for winter temperature extremes in Morocco.

For daytime temperatures, warm winter days are generally associated with a Greenland Anticyclone (NAO−) configuration across most of Morocco, while the zonal regime (NAO+) exhibits a marked inland–coastal contrast, with warmer conditions inland. In contrast, Blocking (BL) and Atlantic Ridge (AR) regimes are more likely to lead to cold daytime events. The AR regime, in particular, shows a dominant influence, accounting for more than 80% of cold daytime events, especially in northern and coastal regions. For nighttime temperatures, the AR regime clearly favors cold outbreaks over the entire country, whereas NAO− conditions strongly enhance the occurrence of warm winter nights. These relationships can be physically interpreted in terms of large-scale warm and cold air mass advection from the Atlantic, with an additional contribution from local radiative warming or cooling under anticyclonic and cyclonic conditions. The intersections and differences between the above-mentioned methods are also analyzed in terms of correlations with the four extremes in addition to weather regime structure.