Daily 100 m near-surface soil moisture prediction from in-situ data upscaled to Landsat footprint in the Yanco agricultural region during 2016-2021

This collection is structured to support reproducible research for "Spatial soil moisture prediction from in-situ data upscaled to Landsat footprint: Assessing area of applicability of machine learning models" (Yu et al., 2025). It provides all necessary input data, trained models, and soil moisture (SM) data extrapolated from 28 OzNet in-situ sites across a primary study area (100 km × 100 km) and an extended area (300 km × 300 km) in southeastern Australia (i.e., the Yanco agricultural region) during 2016-2021. The study period spans a cross-validation period (2016-2019) and an independent test period (2020-2021). The spatial resolution of SM prediction is 100 m and the temporal frequency is daily. A key focus is the characterisation of Area of Applicability (AOA) for Random Forests (RF) and eXtreme Gradient Boosting (XGB) models, delineating where predictions are statistically reliable. The collection includes multiple independent validation datasets from field campaigns, different in-situ networks, and SMAP L2 retrievals for further evaluations.

本数据集旨在支撑论文《基于升尺度至Landsat影像幅宽的原位观测数据开展空间土壤湿度预测：评估机器学习模型的适用域》（Yu等，2025）的可复现研究。数据集涵盖全部所需输入数据、训练完成的机器学习模型，以及2016至2021年间从澳大利亚东南部（即扬科农业区）28个OzNet原位观测站点推演得到的土壤湿度（soil moisture, SM）数据，覆盖核心研究区（100 km × 100 km）与扩展研究区（300 km × 300 km）。研究时段分为交叉验证期（2016-2019）与独立测试期（2020-2021）两个阶段。土壤湿度预测的空间分辨率为100米，时间分辨率为逐日尺度。本数据集的核心研究目标为刻画随机森林（Random Forests, RF）与极限梯度提升（eXtreme Gradient Boosting, XGB）模型的适用域（Area of Applicability, AOA），明确预测结果具备统计可靠性的空间范围。此外，数据集还包含多组独立验证数据集，涵盖野外实测数据集、不同原位观测网络数据以及SMAP L2反演产品，可用于后续模型性能评估。