From virtuality-to-reality: testing the effects of occurrence-based temporal settings on species distribution modeling prediction

Background Temporal information is a fundamental yet often underutilized dimension in species distribution modeling (SDM). While the temporal resolution of environmental predictors is constrained by availability, occurrence data are becoming increasingly abundant and temporally rich. This study investigates how different temporal settings of occurrence data—specifically, overlapping versus non-overlapping designs—affect the performance, consistency, and ecological interpretation of SDMs. Methods Using virtual species and bioclimatic predictors across Southeast Asia, we used SDMs under two temporal scenarios: (i) Nonoverlapping setting is defined by the progressive characteristics in movingwindow time series of the occurrences and sample prevalence within each time window, and (ii) Overlapping setting is defined by each subset of the species distribution data includes cumulatively all the available data (occurrences) prior to the given time period and the increment of sample prevalence as the occurrences cumulate. Models were built using the Random Forest algorithm and evaluated across varying species and sample prevalence levels. Performance was assessed using AUC-ROC, AUC-PR and TSS, while temporal stability and spatial coherence were measured using the temporal coefficient of variation (CV) and the species-habitat index (SHI). Results Overlapping models consistently outperformed nonoverlapping counterparts, yielding higher AUC-ROC and AUC-PR and TSS scores, lower CV values, and more stable suitability predictions. The benefits of overlapping designs were especially pronounced for rare species, where data continuity reduced prediction variability. SHI analyses revealed that overlapping models captured increasing habitat extent and suitability over time, with model variance overwhelmingly driven by habitat area. In contrast, non-overlapping models exhibited declining performance and more balanced contributions from suitability and area metrics. Conclusions Our findings highlight the importance of temporal structure in occurrence data for improving SDM performance and ecological realism. Temporally cumulative (overlapping) sampling strategies provide significant advantages in predictive accuracy, temporal stability, and conservation relevance. As long-term biodiversity datasets become more accessible, we advocate for the integration of temporal structuring as a core component in SDM workflows, particularly for rare species and conservation planning under dynamic environmental change.