A land pattern change represents a globally significant trend with implications for the environment, climate, and societal well-being. While various methods have been developed to predict land change, our understanding of the underlying change processes remains inadequate. To address this issue, we investigate the suitability of the 2D kinetic Ising model (IM), an idealized model from statistical mechanics, for simulating land change dynamics. We test the IM on a variety of patterns, each with different focus land type. Specifically, we investigate four sites characterized by distinct patterns, presumably driven by different physical processes. Each site is observed on eight occasions between 2001 and 2019. Given the observed pattern at the time $t_i$ we find two parameters of the IM such that the model-evolved land pattern at $t_{i+1}$ resembles the observed land pattern at that time. The data supports simulating seven such transitions per site.

Our findings indicate that the IM produces approximate matches to the observed patterns in terms of layout, composition, texture, and patch size distributions. Notably, the IM simulations even achieve a high degree of cell-scale pattern accuracy in two of the sites. Nevertheless, the IM has certain limitations, including its inability to model linear features, account for the formation of new large patches, and handle pattern shifts.

The raw data was downoladed from the National land Cover Dataset and processed to 250x250 binary rasters where cell label 1 indicates the focus land cover and -1 indicates other land covers.  The data consists of time series of binary landscape patterns for four sites in the US and corresponding simulated patterns. All rasters are txt files with integer entries.