Temporal Memory Mechanisms and Biome-Specific Drivers of Ecosystem Carbon Flux: Insights From Explainable Deep...

Area of Science:

• Ecology
• Environmental Science
• Data Science

Background:

• Ecosystem carbon dynamics involve intricate temporal dependencies and environmental interactions, poorly understood across biomes.
• Accurate quantification of these processes is crucial for climate modeling.

Purpose of the Study:

• To develop and apply an explainable AI framework to uncover temporal dependencies in ecosystem carbon flux.
• To identify distinct temporal memory patterns and biome-specific environmental drivers of carbon exchange.

Main Methods:

• Developed an LSTM-Attention framework integrating LSTMs, attention mechanisms, and gradient-based attribution.
• Analyzed eddy covariance flux data from 71 North American sites across eight biomes.
• Employed ablation experiments to validate the role of temporal memory.

Main Results:

• Identified three temporal memory patterns: short-term (grasslands), long-term (forests, wetlands), and U-shaped dual memory (croplands, evergreen forests).
• Discovered biome-specific drivers (e.g., soil moisture in wetlands, radiation in evergreen forests) and critical nonlinear interactions.
• Confirmed that temporal memory is integral to accurate carbon flux modeling and driver identification.

Conclusions:

• Temporal memory plays a fundamental role in regulating ecosystem carbon flux responses to environmental drivers.
• Mechanistic insights into biome-specific temporal controls improve understanding of terrestrial carbon-climate feedbacks.
• This framework enhances predictions of ecosystem functioning under global change scenarios.