Inferring fluctuating interaction probabilities in ecological networks across environmental change

Knowledge of species interactions unlocks our understanding of how ecological communities respond to climate change or habitat loss, explaining their resilience and robustness. Such knowledge requires inferring the presence, sign, and per capita strength of species interactions, as well as species intrinsic growth rates. While various studies have attempted to infer these parameters in isolation, none have successfully inferred them simultaneously. Here, we solve this grand challenge using an integrative approach combining ecological mechanistic models and statistical inference to simultaneously infer these parameters across time, capturing environmental variation and seasonality. We validate our approach on synthetic data in constant and changing environments, highlighting its ability to detect high-probability weak interactions - the key contribution of our method, and proving our ability to detect environmental changes. Applied to empirical data, it recovers the expectations from biological knowledge and unveils network rewiring. Our approach takes one step further to bridge the gap between mechanistic models and empirical ecology. It advances the understanding of ecological networks and their dynamics, thereby helping to validate existing hypotheses, spark new theories, and help guide ecological management and conservation.