I received a grant from the Marie Skłodowska-Curie Actions Postdoctoral Fellowships (MSCA-PF) program: between August 2024 and August of 2026, I will be at the University of Muenster, Germany, working on a project named PRISM: PReservation and RecognItion of Spatial patterns using Machine learning . The project’s primary goal is to develop and compare methods for validating and including spatial patterns in machine learning.

The spatial kinetic Ising model is a simple model of spatial patterns that can be used to simulate the evolution of spatial patterns over time. Its two main parameters are B and J, which control the external pressure and the local autocorrelation tendency, respectively. Both of them have a strong effect on the results of the spatial kinetic Ising model. Thus, the question is how to find the best values of these parameters for a given situation.

A two-dimensional Ising model is an idealized physical system that consists of a lattice of binary variables ( magnetic spins ) that can be in one of two states: up or down. Each spin’s state is influenced by its neighbors: the more neighbors in the same state, the more likely the spin will be in the same state.

Spatial signatures represent spatial patterns of land cover in a given area. Thus, they can be used to search for areas with similar spatial patterns to a query region or to quantify changes in spatial patterns. The approaches above are implemented as lsp_search() and lsp_compare() functions of the motif R package, respectively. At the same time, it is possible to create other, more customized workflows.

The spatial signatures of categorical rasters are a set of numbers that describe the spatial patterns of the provided variables. Next, they allow for further operations such as searching, comparing, or clustering. Less known is that they can also be used to extract information about the composition and configuration of spatial patterns. This blog post shows how to do it using the motif R package.

This is a second blog post in a series about the supercells package. You can read the first one at “supercells: universal superpixels algorithm for applications to geospatial data”. The main idea of supercells is to create groupings of adjacent cells that share common characteristics.

Segmentation is a process of partitioning space into smaller segments. For example, imagine looking at your family photo and trying to distinct individual people. Similarly, we can look at a satellite image (in RGB colors) with the goal of delineating where are the buildings, fields, roads, etc. In geography, segmentation can also be associated with regionalization.

In the past, I wrote a post on how to calculate landscape-level metrics for local landscapes. There, I showed how to divide the categorical input map into a number of smaller areas and calculate selected landscape metrics for each area.

TLTR: This is a last blog post in a series about motif - an R package aimed for pattern-based spatial analysis. It sums up previous posts, but also underlines potential considerations when working with spatial patterns. Finally, it lists underexplored topics and future ideas related to pattern-based spatial analysis.

TLTR: Clustering similar spatial patterns requires one or more raster datasets for the same area. Input data is divided into many sub-areas, and spatial signatures are derived for each sub-area. Next, distances between signatures for each sub-area are calculated and stored in a distance matrix. The distance matrix can be used to create clusters of similar spatial patterns.

TLTR: Quantifying changes of spatial patterns requires two datasets for the same variable in the same area. Both datasets are divided into many sub-areas, and spatial signatures are derived for each sub-area for each dataset. Next, distances for each pair of areas are calculated. Sub-areas with the largest distances represent the largest change.