Land use changes due to natural and human‐related factors, which include wildfires and crop abandonment, are among the most important drivers of soil degradation and demand regular monitoring. Proximal soil sensing in visible–near infrared–shortwave infrared spectral regions could offer a solution. However, to become operational, optimal combination of data and technique has to be defined. Thus, the purpose of this study was (a) to predict the soil organic matter (SOM) content and soil texture in areas of wildfire burns and crop abandonment in Aragón Province, Northern Spain, from their laboratory reflectance spectra using novel correlated components regression with a step‐down variable selection algorithm (CCR‐SD) and (b) to compare the CCR‐SD and the partial least squares regression (PLSR) methods. The results obtained by the tested methods were similar. CCR‐SD models showed high predictive capacity with coefficients of determination (R2) in the range of 0.80–0.86 and 0.70–0.87 for calibration and validation data sets, respectively, and the highest R2 value was attained in the SOM estimation. Moreover, the CCR‐SD models stand out for the superior accuracy–parsimony relationship: the number of predictors varied from 16 (silt models) to 49 (SOM models). On average, the CCR‐SD calibrations needed less than a half of the predictors employed in PLSR models. This research confirmed that CCR‐SD can be used for monitoring SOM content and texture of soils from visible–near infrared–shortwave infrared spectra in the study area and, probably, in other areas of land use/land cover change and that CCR‐SD can create highly parsimonious models that achieve results comparable with the commonly used PLSR method.