It was observed that the difference in the maximum and minimum NDVI values at a time close to harvest (mxNDVI and mnNDVI, respectively), referred to as the haNDVI index (harvest amplitude of NDVI), correlates with agricultural soil quality and the share of sowings. The NDVI becomes saturated when the values of the Leaf Area Index (LAI) significantly exceed one so spatial variation in haNDVI is mainly due to the minimum post-harvest NDVI (mnNDVI). To explain the variability of mnNDVI values three hypotheses were formulated: i) impact of crop selection, ii) field size impact, and iii) impact of soil. To determine which of these hypotheses had the highest impact on the variation in the mnNDVI, the developed machine learning models of this indicator were subjected to a test removing individual explanatory variables from them. Removing a variable does not cause a significant increase in model error if a variable does not contribute useful information to the model. This test showed that the mnNDVI index depends almost exclusively on the crop indicator which was the median of mnNDVI for crops, not directly from soil variables such as the agricultural quality of soil or soil moisture. According to this, the hypothesis of direct impact of soil was rejected. The explanation for the observed correlation of haNDVI with soil quality is the agricultural practice of choosing crops with low mnNDVI (cereals, rapeseed) at better soil conditions and crops with high mnNDVI (fodder crops, grassland) for worse soil conditions.