Hasan (2025) Soil Moisture SK4

Identification

• Journal: Mendeley Data
• Year: 2025
• Date: 2025-11-10
• Authors: Hasan, MD Ahad
• DOI: 10.17632/2ndb7297ff

Research Groups

• Acadia University
• International Soil Moisture Network (ISMN)
• Canada’s Real-Time In-Situ Soil Monitoring for Agriculture (RISMA) network

Short Summary

This paper presents the Soil Moisture SK4 dataset, an 8-year, high-resolution in-situ record of multi-sensor soil moisture, temperature, and collocated meteorological variables from Saskatchewan, Canada, along with a detailed preprocessing workflow for its use in soil moisture prediction research.

Objective

• To describe and make publicly available the Soil Moisture SK4 dataset, a comprehensive 8-year (2014-2021) in-situ record of soil moisture, temperature, and meteorological data from a Canadian agricultural site.
• To detail a transparent data preprocessing and feature engineering workflow designed to prepare this raw sensor data for advanced machine learning models, specifically recurrent neural networks.

Study Configuration

• Spatial Scale: Single in-situ station (SK4) within the RISMA network in Saskatchewan, Canada.
• Temporal Scale: 8 years (2014–2021) of data recorded at 15-minute intervals, with aggregated features at hourly, 24-hour, and 3-day resolutions.

Methodology and Data

• Models used: No models are applied in this data descriptor; however, the data is prepared for use with Recurrent Neural Networks with Temporal Attention.
• Data sources:
  • In-situ sensors: Stevens HydraProbe II frequency-domain reflectometry sensors for soil moisture and temperature at six depths (0-0.05 m, 0.05 m, 0.20 m, 0.50 m, 1.00 m, 1.50 m).
  • Collocated meteorology: Measurements from a nearby weather station including air temperature, precipitation, relative humidity, wind speed, and wind direction.
  • Network: Data collected as part of the International Soil Moisture Network (ISMN) and Canada’s Real-Time In-Situ Soil Monitoring for Agriculture (RISMA) network.
  • Preprocessing: Sensor fusion (averaging co-located probes), antecedent aggregation (24-hour cumulative rainfall, 24-hour means for air temperature, wind speed, relative humidity, 3-day rolling mean of soil moisture), seasonal and diurnal cyclical encodings (day of year, hour of day), physical state features (is_frozen when shallow soil temperature < 0 °C), wind direction decomposition, linear interpolation for missing values, and Min–Max scaling (0–1) for normalization.

Main Results

• A comprehensive 8-year (2014-2021) dataset of in-situ soil moisture and temperature at six distinct depths (0-0.05 m, 0.05 m, 0.20 m, 0.50 m, 1.00 m, 1.50 m) and collocated meteorological variables (air temperature, precipitation, relative humidity, wind speed, wind direction) is provided.
• The dataset features a high temporal resolution of 15 minutes, offering detailed insights into soil and atmospheric dynamics.
• A robust and transparent data preprocessing and feature engineering workflow is detailed, including sensor fusion, antecedent aggregation, cyclical encoding of time, and derivation of physical state features, specifically designed to optimize the data for machine learning applications like recurrent neural networks.

Contributions

• Provision of a unique, high-resolution, multi-depth, multi-sensor in-situ soil moisture and meteorological dataset (SK4) from a Canadian agricultural site, spanning 8 years, made openly accessible.
• Detailed and transparent workflow for preprocessing and feature engineering of raw sensor data, specifically tailored for recurrent neural network-based soil moisture prediction, serving as a reproducible methodology for similar studies.
• Enrichment of the International Soil Moisture Network (ISMN) and Real-Time In-Situ Soil Monitoring for Agriculture (RISMA) with a valuable, openly accessible dataset for research and model development.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Hasan2025Soil,
  author = {Hasan, MD Ahad},
  title = {Soil Moisture SK4},
  journal = {Mendeley Data},
  year = {2025},
  doi = {10.17632/2ndb7297ff},
  url = {https://doi.org/10.17632/2ndb7297ff}
}