THEFAM: An AI-Enhanced Framework for Assessing Thermal Ecological Flow Time Series in River Networks

Emulating marine ecosystem dynamics

Marine ecosystem models such as Atlantis have valuable insight into multispecies interactions, spatial dynamics and fisheries management, but their high computational cost limits are rapid scenario analysis and real-time decision making. This study presented a data-driven ecosystem emulator for the Northeast U.S. Atlantis model using automated machine learning approaches. A large-scale dataset (~3.4 million records) spanning 1964-2020 was contracted, integrating biomass of species functional groups, spatial polygons temporal indices and environments variables such as temperature and salinity. The emulator framework employed automated machine learning techniques, including Random Forest and Extra tress regression, with model selection and hyperparameter optimization performance using Automatic Machine Learning strategies. In addition, Autokeras was utilized to explore neural network architecture in an automated manner, enabling data-driven model allocation, lagged variables to capture ecological inertia and time-aware transformations. Model performance was evaluated using out-of-sample temporal validation, recursive back testing and ecological plausibility assessments. Result demonstrated string predictive performance. with species-level R2 values frequently exceeding 0.90 and overall model accuracy approaching 94%. The emulator achieved high computational efficiency, with end-to-end prediction completed in under few seconds, substantially reducing runtime compared to Atlantis simulations. This work established a scalable and efficient AutoML-driven alternative to process -based ecosystem, model, enabling rapid biomass estimation and supporting data-driven fisheries management and ecosystem analysis.

Forecasting ecological trajectories from ecological dynamic regimes to improve resilience analysis

The ecological dynamic regime (EDR) framework was recently proposed as an alternative to equilibrium‐based approaches for assessing ecological resilience in empirical systems, explicitly incorporating dynamic regimes as a reference for assessing the system's deviation during disturbances. Yet the lack of predictive capacity of the EDR framework limits its applications, especially when long‐term data are unavailable or the disturbed system is not well represented by frequently observed dynamics. Here, we extend the EDR framework by introducing an algorithm (PETRA‐EDR: Predicted Ecological TRAjectories in Ecological Dynamic Regimes ) and a metric (MPD: Mean Predicted Deviation ) to forecast ecological dynamics and estimate prediction accuracy. Our method employs multivariate analyses and can be applied to any ecological system characterized by a set of state variables (e.g. species abundances, functional traits). We conducted a simulation study to evaluate the performance of our method and illustrated its application using empirical data from Canadian boreal forests. Our results demonstrate the method's ability to forecast ecological dynamics and the effectiveness of distance‐weighting functions in improving predictions while mitigating the effects of insufficient sampling, observation noise and hidden variables. Finally, we discuss the assumptions of our method and its applications for assessing ecological resilience to pulse disturbances and detecting regime shifts from a multidimensional, dynamic perspective.

LLM-based Evaluation Policy Extraction for Ecological Modeling

Evaluating ecological time series is critical for benchmarking model performance in many important applications, including predicting greenhouse gas fluxes, tracking soil moisture and soil–atmosphere interactions, and monitoring hydrological cycles. Traditional numerical metrics (e.g., R-squared, root mean square error) have been widely used to quantify the similarity between modeled and observed ecosystem variables, but they often fail to capture domain-specific temporal patterns critical to ecological processes. As a result, these methods often need to be accompanied by expert visual inspection, which requires substantial human labor and limits the applicability to large-scale evaluation. To address these challenges, we propose a novel framework that integrates metric learning with large language model (LLM)-based natural language policy extraction to develop interpretable evaluation criteria.The proposed method processes pairwise annotations and implements a policy optimization mechanism to generate and combine different assessment metrics. The results obtained from datasets that cover crop GPP and CO2 flux, streamflow discharge, and site-level CH4 flux have confirmed the effectiveness of the proposed method in capturing target assessment preferences, including both synthetically generated and expert-annotated model comparisons. The proposed framework bridges the gap between numerical metrics and expert knowledge while providing interpretable evaluation policies that accommodate the diverse needs of different ecosystem modeling studies. Implementation can be found in https://github.com/qxc101/APEF_sub.git.

Machine learning-based multi-scale dynamics of terrestrial carbon fluxes and their environmental drivers along the U.S. East Coast

MacroSimply: An Intuitive Rule-Based Classifier to Predict Biotic Status in Supporting Ecological Restoration

We address the challenging problem of predicting the biological quality of a water body based on a set of environmental and management-related drivers. While modelling approaches for predicting water quality from hydrological and pollution-load variables are well established, analogous tools for predicting biological status remain less consolidated. Both components are nevertheless required within the Water Framework Directive to assess ecological status (ES) and support restoration planning. This paper concentrates on the latter challenge by presenting an experience developed over four heavily impacted rivers in Regione Lombardia (Northern Italy). Rutinary environmental, hydromorphological and biological data were systematized to develop a predictive framework linking management-related pressures to the macroinvertebrate component of ecological status. A rule-based classifier, denominated MacroSimply, was developed and compared with alternative statistical and machine learning approaches: logistic regression, a classification tree (CHAID), and a multilayer perceptron neural network. The tested models exhibited different strengths and weaknesses: Logistic regression provided a good balance between predictive performance and interpretability; the classification tree generated transparent threshold-based decision rules; and the neural network captured potentially complex non-linear relationships, albeit with reduced interpretability. MacroSimply achieved predictive performance comparable to the alternative approaches, and performing even higher reliability, while maintaining full transparency of the structure and a direct connection between predictors and management actions. The proposed framework was subsequently implemented in a spreadsheet-based tool to be easily used in restoration planning exercises. Although the obtained model should not be considered the definitive modelling solution even for our particular case, the results suggest that the adopted rule-based approach represents a working, valuable option preferrable to more complex data-driven methods when interpretability, reproducibility and practical applicability are key requirements for environmental management.

The rENM Framework: A Modular System for Reconstructing and Analyzing Long-Term Ecological Niche Dynamics

Retrospective ecological niche modeling (rENM) combines historical species occurrence records with historical environmental data to reconstruct the spatio-temporal dynamics of species-environment relationships under changing conditions. Despite growing recognition that those relationships can be nonstationary, time-series approaches to ecological niche modeling remain uncommon, and the tools to support them at scale are limited. Here, we describe the rENM Framework, an experimental, open-source suite of R packages that automates a complete rENM workflow spanning data preparation, ensemble time-series construction, trend analysis, AI interpretation, and report generation. The framework integrates eBird occurrence records with environmental variables derived from NASA’s MERRA-2 reanalysis across a 45-year study period (1980–2024) and executes a complete analysis for any species with eBird data through a single function call. By treating climatic suitability as a dynamic ecological response surface rather than a static baseline, the framework produces the following analytical products that complement conventional ecological niche modeling approaches: suitability time series, long-term trend and acceleration maps, centroid displacement estimates, bioclimatic velocity metrics, variable contribution trajectories, and hotspot analyses identifying areas of accelerating suitability decline. We illustrate the framework’s outputs with a representative run for Cassin’s Sparrow (Peucaea cassinii), a grassland species of conservation concern in the arid southwestern United States and the focal species throughout our development work. The framework’s automated, unsupervised pipeline makes systematic application across large numbers of species tractable, with direct implications for conservation assessments, such as State Wildlife Action Plans, where species-specific analytical capacity is often limited by available resources. The rENM Framework is openly available on GitHub and archived on Zenodo.