Skillful subseasonal-to-seasonal (S2S) forecasts rely heavily on windows of opportunity, often afforded by teleconnected sources of variability throughout the Earth system. Due to its longer decorrelation scale, the stratosphere is highlighted as one such source of variability. Recently, the surface impacts of stratospheric variability on S2S timescales have been viewed not only through the mean flow diagnostics but also through the distinct geometry and behavior of the stratospheric polar vortex in the wintertime (November through March). This analysis will assess the impact of stratospheric polar vortex variability on the troposphere through its interactions with climate modes, primarily the North Atlantic Oscillation (NAO).  In this research, lower stratospheric potential vorticity (PV) that aids in capturing troposphere-stratosphere coupling processes is used to investigate stratospheric sources of S2S prediction skill. The analysis uses ERA-5 reanalysis data to assess stratospheric teleconnections to the NAO index as described by geopotential height (GPH) anomalies and their contributions to S2S predictions of 2-meter wintertime temperature anomalies across the Northern Hemisphere. Using eXplainable Artificial Intelligence (XAI) and Machine Learning, this research builds dynamical-statistical probabilistic forecasts of 2-meter temperature anomaly on the S2S scale over Eurasia through a feed-forward Artificial Neural Network (ANN) framework. The merged ANN model is constructed with data that includes 14-day lead lower stratospheric PV at 100hPa and synoptic scale North Atlantic GPH anomalies at 500hPa. The model output is interpreted through an XAI method called Layerwise Relevance Propagation (LRP), which is used to examine how each input feature influences the 2-meter temperature predictions and highlights their relative importance. Requisite spatial maps of the input features discretely connect physical processes from the stratosphere to the NAO and highlight potential windows of opportunity for confident and correct predictions of  2-meter temperature outcomes on the S2S timescale.