Multi-indicator deterministic model based on time series of Sentinel-2, to assess the degree of natural succession on the abandoned arable areas

Abstract The article presents the concept of a deterministic model for assessing the degree of natural succession on long-term abandoned land in the agricultural production area of Poland. The model was implemented as a geographic information system tool. It is based on two basic sources of information: cadastral maps, which can suggest if the agricultural land is agriculturally used, and seasonal time series of satellite images. The following working hypothesis was adopted in this study: “based on the data from the Sentinel-2 sensor, it is possible to assess the degree of natural succession on small and medium agricultural plots – by separating at least three classes of land cover, representing: early succession consisting mainly grass and ruderal vegetation, advanced succession represented by complexes of shrubs and young trees, mature succession - which is a transitional phase preceding the afforestation phase, or can already be a functional forest area”. The obtained results confirmed the above hypothesis. In the case of early succession, the classification efficiency was about 94% in the winter period, for advanced succession about 75% in the autumn period, and for mature succession about 78% in the summer period. In the classification process, 8 vegetation indices were examined. In the end, the model algorithms were based on the GNDVI index, whose properties allowed for the best differentiation between the above-mentioned succession classes. NDVI, NDRE, NDVIre1, NDVIre2, NDVIre3 were used as auxiliary indices, which, as shown in the research, can improve the classification accuracy at a higher uncertainty threshold in case of weaker separation of classes with the GNDVI index.