A fundamental challenge in climate science is decomposing the concurrent drivers of weather extremes in observations. Achieving this can provide insights into the drivers of individual extreme events as well as into possible future changes in extreme event frequencies under greenhouse forcing. In the present work, we exploit recent results from dynamical systems theory to study the co-variation and recurrence statistics of different atmospheric variables. Specifically, we present a methodology to quantify the recurrences of bivariate variables and the coupling between distinct univariate variables in terms of their joint recurrences. The coupling is defined by a parameter which varies according to the chosen variables, season, and domain and can be understood in terms of the underlying physics of the atmosphere. For suitably chosen variables, this approach enables to decompose the different drivers of weather extremes. Here, we compute the above metrics for near-surface temperature and sea level pressure, and use them to study warm or cold days over North America. We first identify states where temperature is strongly or weakly coupled to the large-scale atmospheric circulation, and then elucidate the interplay between coupling and the occurrence of temperature extremes.