This study evaluated different input features for the local climate zone (LCZ) classification using a random forest (RF) classifier. The input features included spectral reflectance and textural features from Sentinel-2 multi-spectral imagery and polarimetric features from dual-polarized (HH+HV) PALSAR-2 data. The analysis of the feature importance for the RF classifier was measured by Gini and permutation importance. The analysis of the feature contributions to each LCZ class was performed by a feature contribution method based on decision paths in the RF. The results showed that the multi-spectral bands from Sentinel-2 imagery played a dominant role in LCZ classification, especially Band 12 (short-wave infrared-2). The contributions of the PALSAR-2 HV polarization band were higher in land cover LCZ types than in built LCZ types. The combined analysis of feature importance and contribution would provide a reference for the performance of RF classifiers in terms of LCZ mapping.