This dataset integrates high-frequency distributed acoustic sensing (DAS) with synchronized agronomic and meteorological records to characterize the hydromechanical response of agricultural soils. The data were collected in March 2023 at Harper Adams University, UK, following a mechanical tillage event. By analyzing ambient seismic noise, the dataset tracks relative seismic velocity changes (dv/ v) as a proxy for soil moisture dynamics and structural changes across various traffic and tillage management systems. The results provide a mechanistic framework for understanding how tillage impairs soil moisture retention and drought resilience.

Description of the data and file structure

The dataset is organized into a large-scale raw seismic file and a processed directory containing environmental and physical metadata.

1. Raw Data: farmDAS_concat_100hz.hdf5

It mainly contains a 2D HDF5 array representing the raw strain-rate measurements. This file is intended for researchers who wish to apply alternative seismic interferometry to the raw wavefield.

• Sub-objects of HDF5 dataset:
  • Key "data": dimension is {50, 12960000}, representing the DAS data saved for 129600 seconds with 100 Hz sampling rate, and at 50 locations.
  • Key "dt": sampling rate (single scalar value)
  • Key "timestamp": a list of 2408 strings of the time stamp representing the starting time of every minute.

2. Processed Data: processed.zip

This archive contains the interpreted data products required to reproduce the study’s figures and models.

A. Autocorrelations (ACFs: autocorr_15_60Hz_3chs500pts.hdf5):

• Terminology: ACFs represent the "fingerprint" of the subsurface. By comparing these over time (Passive Seismic Interferometry), we measure dv/v (relative velocity change). A decrease in dv/v generally indicates soil wetting or loosening.

• Data definition: minutely averaged ACFs and downsampled in time by a factor of 5. Today 482 samples in clock time and 500 points in lag time.

• Window: Correlation lag times from -5 to +5 seconds.

• Sub-objects of HDF5 dataset [key, dimension {clock time every 5 minutes, lag time every 0.01 s}]: 

  corr_all_time_ch18,       {482, 500}

  corr_all_time_ch33,       {482, 500}

  corr_all_time_ch44,       {482, 500}

  corr_all_time_stretched1_ch18,  {482, 500}

  corr_all_time_stretched1_ch33,  {482, 500}

  corr_all_time_stretched1_ch44,  {482, 500}

  corr_all_time_stretched2_ch18,  {482, 500}

  corr_all_time_stretched2_ch33,  {482, 500}

  corr_all_time_stretched2_ch44,  {482, 500}

  corr_all_time_stretched3_ch18,  {482, 350}

  corr_all_time_stretched3_ch33,  {482, 350}

  corr_all_time_stretched3_ch44,  {482, 350}

B. dv/v (final_peaks_deRatio_3iterations.h5):

• Data definition: dv/v measured by stretching ACFs, and normalized by measuring the ratio between each dv/v to the reference.

• Sub-objects of HDF5 dataset:

  key "deratio_dvv": dimension is {50, 482}, representing dv/v measured at 482 clock time points and 50 locations.

C. Rain proxy by power spectral density (integrated_psd.hdf5):

• Data definition: minutely integrated power spectral density (PSD) of 80-140 Hz of the raw data spectrum.

• Sub-objects of HDF5 dataset:

  • Key "file_list": 2408 strings representing the filenames of the raw minute-long DAS data.
  • Key "freq Dataset": 129 float numbers representing the discrete frequencies between 80-140 Hz.
  • Key "int_PSD": dimension is {51, 2408}, representing integrated PSDs at 2408 minutes and 51 locations (including an unused last end point).

D. Weather Records 1 (met_HarperAdams_merged.csv):

• Key "Date": day/month/year
• Key "Time": HHMM
• Key "10cm Soil Temperature": soil temperature at 10 cm depth, in degrees Celsius (°C)
• Key "30cm Soil Temperature": soil temperature at 30 cm depth, in degrees Celsius (°C)
• Key "100cm Soil Temperature": soil temperature at 100 cm depth, in degrees Celsius (°C)
• Key "Rainfall Total since 0900": cumulative gauges with daily manual calibration, in mm.
• Key "Humidity": air relative humidity (normalized to 100%)

E. Weather Records 2 (met_newport_700m.csv):

• Key "Time": hour:minute, in every 30 minutes
• Key "Temperature": air temperature in degrees Celsius (°C)
• Key "Wind": wind direction (S: south, N: north, E: east, W: west)
• Key "Wind Speed": wind speed in m/s
• Key "Condition": weather condition

F. Soil Properties (soil_data_interp.csv):

• Key "tillage_depth": depth of tillage at each plot (0-25 cm)
• Key "tire_pressure": traffic tire pressure (70-150 kPa)
• Key "porosity_10cm": soil porosity at 10 cm depth derived from the bulk density measurements
• Key "porosity_20cm": soil porosity at 20 cm depth derived from the bulk density measurements
• Key "porosity_30cm": soil porosity at 30 cm depth derived from the bulk density measurements

Sharing/Access information

Data was derived from the following sources:

• Weather history for Newport, Shropshire: https://www.timeanddate.com/weather/@7296017/historic
• Harper Adams University Meteorological Station (Internal Records).

Code/Software

The processing workflow for this dataset is primarily Python-based.

Core Software:

• NoisePy4DAS (v1.0): Used for computing the time-lapse cross-correlations and autocorrelations.
• Python (v3.8+): Including packages h5py for HDF5 manipulation, pandas/numpy for meteorological data analysis, and obspy for seismic data handling.

Workflow:

The workflow selects one out of every five 1-minute raw segments to reduce data volume, applies one-bit normalization to the waveforms to remove transient noise (like vehicle pass-bys), and computes the ACFs in the frequency domain before transforming them back to the time domain for phase-shift analysis.

The code for reproducing the analysis can be accessed through the Denolle Lab GitHub repository: https://github.com/Denolle-Lab/FarmDAS