In this paper, we develop a stochastic model to investigate the variance dynamics and term structure of the US natural gas market. The model features two variance factors, the Samuelson effect, seasonality and linearly analytical pricing formula. With variance swap rates synthesized from options prices spanning more than 20 years, we show that the two-factor variance model has better performance and is preferred to describe the variance term structure. Moreover, we find that the two variance factors are particularly needed to capture humps in variance term curves implied from the market. Finally, we apply the two-factor model to hedging natural gas straddles and obtain improved performance. These results highlight the importance of introducing multiple variance factors to the natural gas market, and may be applied to other similar sectors.