Improving spectral estimation of soil inorganic carbon in urban and suburban areas by coupling continuous wavelet transform with geographical stratification

•Absorption peak around 2310 nm was identified as the most important for SIC.•An urban SSL with 3492 samples was developed for SIC prediction.•CWT scale 1 showed the optimal performance in predicting SIC.•Performance of CNN was superior to other common machine learning models.•Stratifying SSL by LULC in more homogeneous classes achieved better SIC prediction.