Transferable neural operators to estimate sub-surface soil moisture code

This Zenodo record provides the Python code used in the manuscript to train and evaluate Fourier Neural Operator (FNO) models and to assess transfer learning performance across sites/datasets. The record includes three scripts: Tranfer_FNO.py: trains baseline FNO models from scratch for each dataset/site and saves trained weights and prediction outputs. Transfer_FNOv1.py: performs zero-shot transfer by applying pretrained source-site FNO models directly to target sites without fine-tuning, and evaluates performance over multiple target-data fractions. Transfer_FNOv2.py : performs transfer learning with fine-tuning, evaluating progressive fine-tuning fractions (including a 0% baseline) using a fixed target test split. Data availability: The datasets required to reproduce the results are archived at Zenodo: https://zenodo.org/records/16106763.For reviewers: Please use the dataset-access link provided in the manuscript Open Research section (restricted access during peer review) The scripts save model checkpoints and CSV outputs (predictions and summary metrics) in the working directory (or user-defined output folders). Please refer to README.TXT in this record for environment requirements, folder structure, and step-by-step run instructions.