Robustness of spatial metrics under controlled information loss in alien plant distributions

Abstract Context Spatial metrics are often used to describe the observed distribution of alien plant species, but it remains unclear how reliable they are when sampling is incomplete or spatially uneven. Objectives We assessed how four spatial metrics respond to decreasing sampling completeness under different patterns of spatial information loss. Methods We used a systematically surveyed regional atlas of alien vascular plants in Lazio (central Italy), comprising 272 taxa across 532 grid cells and progressively reduced the amount of available spatial information under two contrasting scenarios: random removal of grid cells and clustered removal of contiguous spatial areas. We quantified the response of four spatial metrics: area of occupancy (AOO), extent of occurrence (EOO), number of patches (NP), and patch cohesion (COH). Results Spatial metrics responded differently to decreasing sampling completeness. AOO declined gradually and remained relatively robust under both scenarios, whereas EOO and COH were more sensitive to the spatial pattern of information loss. EOO declined more strongly under clustered removal, whereas COH was more strongly reduced under random removal. NP behaved differently from all other metrics, increasing under random removal and decreasing under clustered removal. Metric robustness also differed among invasion-status categories, with invasive species generally maintaining higher EOO and COH values than naturalised and casual taxa. Conclusions Spatial metrics are not interchangeable and differ markedly in their sensitivity to information loss and to the spatial pattern of missing data. In the context of biological invasions, understanding these responses is essential for the reliable interpretation of documented species distributions and for the effective use of spatial metrics in monitoring, management and control strategies. These metrics should be interpreted as descriptors of observed distributions, not as estimates of potential distributions, especially when sampling is incomplete or spatially structured.