Prediction of Agricultural Land Cover Change with the Use of ANNs and Open Source Data

Land cover changes occur continuously across the world. Both human activities and natural processes constantly modify the appearance of a given area. Predicting the direction of such changes can not only help prepare the population mentally for them, but also provide tangible benefits. This is particularly important for agricultural areas, whose location is crucial to the efficiency of agricultural holdings. In this study, the authors proposed a methodology for predicting land cover changes. For this purpose, artificial neural networks (ANNs), implemented using MOLUSCE, a QGIS plugin, were employed. The prediction was performed for a selected area in Poland based on historical land cover data from 2019 to 2022. The model was validated against the observed land cover map of 2025 and used to predict land cover for 2028. The validation yielded an overall accuracy of 95.48%; however, comparison with a persistence (no-change) baseline and change-specific metrics showed that this accuracy results mainly from the stability of land cover over the three-year horizon. The prediction for 2028 indicates an increase in agricultural land by 6.55 km2 (+2.0%), mainly at the expense of forests (−5.78 km2, −2.9%), with no further expansion of artificial surfaces. This reflects the dynamics of the 2019–2022 calibration period and differs from the urban expansion observed in 2022–2025. The results demonstrate both the potential and the limitations of the open-source workflow and lead to recommendations for its application in other areas.