A new study published in Big Earth Data develops explainable Random Forest models to predict stream total nitrogen (TN) and total phosphorus (TP) concentrations using upstream catchment characteristics derived from openly accessible environmental datasets. The results show that the models provide robust predictions for TN, while reduced and spatial covariates help mitigate overfitting and improve model interpretability, demonstrating the potential of explainable machine learning for regional nutrient monitoring.

Researchers at the FAMU-FSU College of Engineering and Florida State University’s Resilient Infrastructure and Disaster Response Center have identified the factors that are most likely to create hazardous respiratory conditions inside homes after flooding.

A study led by Prof. Emre Yakşi of the Norwegian University of Science and Technology, who is also a visiting faculty member at Koç University, reveals how the zebrafish forebrain organizes and integrates visual and vibrational information. The findings point to a shared computational principle of sensory processing across vertebrate brains.

Scheduling agile earth observation satellites is a complex optimization challenge. To tackle this, researchers developed a novel co-evolutionary framework where Large Language Models (LLMs) learn from expert-designed algorithms to automatically design superior scheduling strategies. This pioneering approach, which bridges the LLM's knowledge gap in this specialized domain, resulted in an average imaging profit improvement of 8.48% compared to state-of-the-art methods, demonstrating a new path for AI-driven algorithm design.

In a paper published in ENGINEERING Management, researchers from Shanghai University and Tsinghua University propose a cross-city transfer learning framework that enables rapid planning of shared e-scooter parking stations in new cities without requiring local operational data. Using operational data from 25 European cities and multi-source open-space data, the study demonstrates that a multi-dimensional urban similarity matching strategy, integrating socio-economic, POI distribution, and spatial structure features, achieves more stable and generalizable predictions than conventional geographic or economic grouping methods, with the TOP-3 similarity match yielding the best overall performance.

A new review in ENGINEERING Management shows how the inverse Gaussian process has developed into an important probabilistic tool for degradation modeling, remaining useful life prediction, and maintenance decision-making in prognostics and health management. It also explores how artificial intelligence can be combined with interpretable statistical models to support more reliable and scalable intelligent maintenance.