Microscale nonrandomness revisited: An overlooked predictor of ecological change

Abrupt ecological transitions have driven the development of early warning signal metrics that track how ecological systems respond as external conditions gradually change and systems approach transitions between states. However, their generality and applicability remain uncertain, particularly for short-term ecological dynamics that can occur without directional external change. To address this gap, we propose a general framework based on the hypothesis that microscale nonrandomness, quantified from frequency distributions or spatial patterns, predicts subsequent short-term rates of change in populations and interactions, as well as current macroscale ecosystem states. We applied the framework to four datasets: time series from an incubation experiment of phytoplankton-parasitic chytrid interactions; spatiotemporal field observations of leaf beetles on willow trees and aphids in a crop system; and spatial snapshot data linking soil surface characteristics with aboveground and soil ecosystem properties in a forest vegetations-soil system. We quantified microscale nonrandomness using Morisita I_δ from frequency distributions and Moran's I and SADIE from spatial patterns. Across all systems, models incorporating microscale nonrandomness metrics consistently showed greater cross-validated predictive performance than those based on macroscale variables or spatial averages alone. Although their relative performance varied among systems, frequency-based metrics performed as well as or better than spatially explicit metrics in some cases. The consistent improvement across contrasting systems supports the broad applicability of this framework, including under limited data conditions. This framework broadens ecological forecasting to short-term, continuous dynamics beyond those associated with directional environmental change.