Recently, extremely large-scale multiple-input multiple-output (XL-MIMO) is considered to be one of the potential technologies to satisfy the high-data demand of the next-generation mobile telecommunication system. However, existing literatures cannot accurately estimate the near-field channel state information for wideband XL-MIMO due to the lack of a thorough account of inherent spatial non-stationarity and spatial wideband effect, which leads to severe data-rate performance degradation in the practical implementations. To tackle the above challenges, we investigate the near-field non-stationary channel estimation (CE) issue for the wideband XL-MIMO systems. To be specific, we firstly refine the conventional near-field wideband channel model by carefully considering the joint effect of the spatial wideband effect and the spatial non-stationarity. On the basis of the refined channel model, we develop the optimal sub-array division principle. Then, a subarray-wise polar domain simultaneous orthogonal matching pursuit (SP-SOMP) algorithm for CE of wideband XL-MIMO is proposed. The simulated results have indicated that proposed SP-SOMP algorithm can achieve a higher normalized mean square error (NMSE) performance than traditional CE schemes under this consideration of the spatial wideband effect and spatial non-stationarity.