Towards Fine-Tuning-Based Site Calibration for Knowledge-Guided Machine Learning

Ruolei Zeng, Arun Sharma, Shuai An, Mingzhou Yang, Shengya Zhang, Licheng Liu, David Mulla, Shashi Shekhar

Abstract:

Accurate and cost-effective quantification of the agroecosys-
tem carbon cycle at decision-relevant scales is essential for
climate mitigation and sustainable agriculture. However, both
transfer learning and spatial variability exploitation in this
field are challenging, as they involve heterogeneous data and
complex cross-scale dependencies. Conventional approaches
often rely on location-independent parameterizations and in-
dependent training, underutilizing transfer learning and spa-
tial heterogeneity in the inputs, and limiting their applicabil-
ity in regions with strong 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 ma-
chine 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 mid-
western sites, FTBSC-KGML estimates land emissions while
leveraging transfer learning and spatial heterogeneity. A key
component is a spatial heterogeneity-aware transfer-learning
scheme: a globally pretrained model that is fine-tuned per
state/site to learn place-aware representations, improving local
accuracy under limited data without sacrificing interpretability.
Empirically, FTBSC-KGML achieves lower validation error
and more consistent explanatory power than a purely global
model, better capturing spatial variability across states. This
work extends the prior SDSA-KGML framework.