From diagnosis to prediction: A cross-scale intelligence framework redefines how we assess aquatic ecosystem health

Aquatic ecosystems worldwide are facing unprecedented degradation driven by climate change, dam construction, and intensive human activity. Conventional assessment methods—relying on periodic biological sampling—have proven largely descriptive and retrospective, often failing to detect ecosystem collapse until catastrophic events such as mass algal blooms occur.

In a new review article published in Water & Ecology, a research team led by Yong Liu from Peking University calls for a fundamental transformation in how we monitor and manage freshwater ecosystem health.

“Traditional indicators tell us a system is 'ill' but cannot diagnose the specific 'cause', nor can they predict when a tipping point might be reached,” says Liu. “We need to move from post-hoc diagnosis to predictive early warning.”

The authors trace how international frameworks—from the U.S. National Aquatic Resource Surveys to the EU Water Framework Directive and China's “Technical Guideline for Aquatic Ecosystem health Assessment”（GB/T 43476-2023）have expanded assessment coverage, yet bottlenecks remain: baseline drift under climate change, decoupling between chemical and biological recovery, and inability to detect non-linear regime shifts.

To address these gaps, the team proposes a gene-to-landscape framework spanning four interconnected scales:

• At the micro-scale, metagenomic analyses detect cellular stress before macroscopic symptoms appear.
• At the mid-scale, environmental DNA tracks genetic diversity.”
• At the mini-scale, explainable machine learning disentangles multiple stressors.
• At the macro-scale, AI-coupled remote sensing enables continuous, basin-wide surveillance.

“The key is vertical integration—molecular signals propagate upward, while landscape stability provides top-down predictive constraints,” says Liu.

The framework's practical value is already emerging. In China's South-to-North Water Diversion Project, deep-sequencing of cyanobacterial metagenomes identified genomic architecture as a risk indicator—larger streamlined genomes (>3 Mbp) showed significantly higher toxin potential, enabling preemptive intervention. In Lake Hongze, remote sensing revealed that subtle water-level fluctuations govern vegetation distribution and carbon fixation, insights impossible through discrete sampling alone.

“These cases show cross-scale intelligence can reveal hidden mechanisms and provide actionable early warnings,” says Liu. “It redefines assessment from a static snapshot to a dynamic, predictive process.”

The authors acknowledge implementation challenges, including data integration complexity and high technical costs. “Phased implementation starting with priority basins, coupled with expanded monitoring networks and process-informed AI models, offers the most realistic pathway forward,” adds Liu.

###

Contact the authors:

Yong Liu

-College of Environmental Sciences and Engineering, State Environmental Protection Key Laboratory of All Material Fluxes in River Ecosystems, Peking University, Beijing 100871, China

-Institute of Tibetan Plateau, Peking University, Beijing 100871, China

yongliu@pku.edu.cn

The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 200 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).