Towards Fine-Tuning-Based Site Calibration for Knowledge-Guided Machine Learning: A Summary of Results

TL;DR

This work tackles accurate, scalable prediction of agroecosystem land emissions under spatial variability by introducing FTBSC-KGML, a framework that combines global pretraining on multi-state data with site-level fine-tuning and site-specific calibration. By embedding physics-guided constraints and a two-phase transfer mechanism, the approach leverages cross-site knowledge while adapting to local covariate distributions, addressing data sparsity. Empirical results across Illinois, Iowa, and Indiana show that global pretraining plus local calibration yields lower validation MSE than purely global or site-only methods, with particularly strong gains in data-limited regions and robustness to hyperparameter changes. The methodology provides a practical, interpretable route to spatially aware KGML for reliable land-emission estimation in heterogeneous agroecosystems, extending prior SDSA-KGML work.

Abstract

Accurate and cost-effective quantification of the agroecosystem carbon cycle at decision-relevant scales is essential for climate mitigation and sustainable agriculture. However, both transfer learning and the exploitation of spatial variability in this field are challenging, as they involve heterogeneous data and complex cross-scale dependencies. Conventional approaches often rely on location-independent parameterizations and independent training, underutilizing transfer learning and spatial heterogeneity in the inputs, and limiting their applicability in regions with substantial variability. We propose FTBSC-KGML (Fine-Tuning-Based Site Calibration-Knowledge-Guided Machine Learning), a pretraining- and fine-tuning-based, spatial-variability-aware, and knowledge-guided machine learning framework that augments KGML-ag with a pretraining-fine-tuning process and site-specific parameters. Using a pretraining-fine-tuning process with remote-sensing GPP, climate, and soil covariates collected across multiple midwestern sites, FTBSC-KGML estimates land emissions while leveraging transfer learning and spatial heterogeneity. A key component is a spatial-heterogeneity-aware transfer-learning scheme, which is a globally pretrained model that is fine-tuned at each state or site to learn place-aware representations, thereby improving local accuracy under limited data without sacrificing interpretability. Empirically, FTBSC-KGML achieves lower validation error and greater consistency in explanatory power than a purely global model, thereby better capturing spatial variability across states. This work extends the prior SDSA-KGML framework.

Figures (6)

• Figure 1: Input and output of KGML-ag Liu2024_KGML_Carbon.
• Figure 2: Validation MSE loss evolution across the five-step KGML-ag training process (batch size = 8, learning rate = 0.001).
• Figure 3: Overview of the proposed FTBSC-KGML framework.
• Figure 4: Heatmap of validation MSE across different state-level training and testing combinations.
• Figure 5: Validation MSE by state comparing with global pretraining (global $\rightarrow$ state fine-tuning; maroon) vs. without pretraining (state-only training from scratch; blue). Pretraining yields consistently lower MSE in IA/IL/IN.