Resource-limited ecosystems, such as drylands, can exhibit self-organized spatial patterns. Theory suggests that changes in these patterns can inform about the ecosystem degradation level. While the current theory is expected to work well when following a given site in time, we still lack ways of comparing different field sites using static observations of vegetation spatial structure. Such methods would be crucial to pinpoint the least fragile ecosystem areas from observation of spatial structure. Here, we tackle these limitations using an inverse-modelling approach relying on Approximate Bayesian Computing, which uses a snapshot of the spatial structure of ecosystems characterized by so-called irregular patterns and estimates a distance to its desertification point regardless of whether it is reached via progressive or sudden degradation. The approach allows us to comparably rank sites according to their distance to their desertification point (i.e. their resilience). We validated the approach on simulated landscapes from different models, showing that the approach performs well at ranking all landscapes. We applied the approach to a global dryland dataset and investigated the drivers of the estimated distances to desertification. We emphasized a possible application of our approach by combining these distances to desertification with aridity projections to illustrate the possibility of integrating our approach into the risk assessment of drylands. This made it possible to pinpoint the least resilient sites among those studied, therefore, paving the way for a risk-assessment method for spatially organized ecosystems. Our inverse-modelling approach can be extended to other ecosystems and other types of spatial patterns (e.g. periodic patterns). By allowing linking spatial structure and ecosystem resilience, it offers a promising development toward comparing and ranking sites of self-organized ecosystems from a single snapshot of their spatial structure.