Title: CHAM-net: A Contrastive Hierarchical Adaptive Meta-network for Robust Global Methane Flux Prediction

Abstract: Methane acts as a powerful greenhouse gas, playing a major role in driving global warming. Nevertheless, precise quantification of global methane uptake and release continues to pose a significant challenge, largely because environmental factors interact in complex ways that shift across different spatial and temporal dimensions. Previous data-driven approaches have frequently neglected the inherent spatiotemporal heterogeneity found in ecosystems, thereby failing to account for unique site-specific traits and evolutionary trends over multiple years. To overcome these limitations, we introduce CHAM-net (Contrastive Hierarchical Adaptive Meta-network), an innovative framework designed to explicitly learn from historical contexts to model site-specific dynamics. This architecture utilizes a hierarchical encoder-decoder structure; the encoder extracts distinct site characteristics from past data, which then dynamically conditions the decoder to produce the final forecast. Our experiments reveal that CHAM-net consistently surpasses all baseline methods on both observational and simulation datasets regarding methane consumption and emission. The model achieved nRMSE scores as low as 0.43 and 0.88, with R2 values reaching up to 0.97 and 0.68, respectively, for emission prediction tasks.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC