In the recent years, global climate change has induced evergrowing loss of sea ice in the Arctic. As the sea ice disappears, albedo diminishes and the sea surface is more likely to be warmed by incoming solar radiation. With the right wind conditions, this extra heat may also be advected towards the shore and thus influence coastal atmospheric temperatures. Thus, knowing how coastal atmospheric temperature is related to offshore conditions is paramount to help predict inshore effects. To study this relation, an exploratory approach using machine learning algorithms is proposed. Based on a combination of daily in situ (i.e. wind velocity, sea level pressure) and remotely sensed (i.e. sea surface temperature, sea ice concentration) data, a series of predicting features are constructed for the years 1981-2010. Two implementations of support vector regression (SVR), one with a linear kernel and the other with a combination of gaussian and histogram intersection kernels are then applied. Results of the SVR indicate that prediction root mean squared errors of less than 5°C are routinely achievable. Prediction errors are also found to be the smallest in summer months and/or at lower latitudes. Finally, the relative importance (ranking) of features appears to be highly variable, depending both on the location and the period of the year.