Sentinel-1 Soil Moisture Estimation

Overview

This repository outlines a step-by-step workflow for estimating soil moisture at the field scale using Sentinel-1 Ground Range Detected (GRD) Synthetic Aperture Radar (SAR) data. The approach combines the Dual-Polarization Radar Vegetation Index (DpRVI_c) with a change detection-based algorithm (CDA) for better accuracy during cropping period.

For a detailed explanation of the methodology, please refer to the research paper by Bhogapurapu et al. (2022).

Key Findings

Model Performance: Correlation coefficient (r) = 0.75, RMSE = 0.095 m³ m^-3.

---

Workflow

1. Preparing Sentinel-1 Data Using Google Earth Engine

This section guides you through preprocessing Sentinel-1 GRD SAR scenes (VV/VH, 10m resolution) with Google Earth Engine. The process involves:

• Calculating the Dual-Polarization Radar Vegetation Index (DpRVI_c) to reduce vegetation influence on SAR backscatter intensity.
• Preparing the backscatter intensity in vertical-vertical polarization (VV_dB).

For your convenience:

• Access the Google Earth Engine code for Step 1

2. Dry and Wet Reference Calibration

In this step, you will calculate:

• Dry Reference Backscatter Intensity (σ°_dry)
• Backscatter Change (Δσ)
• Maximum Backscatter Change (Δσ_max) by preforming pixel based regression analysis between the backscatter change (Δσ) and the Dual-Polarization Radar Vegetation Index (DpRVI_c)

This step is implemented using an R-script, which processes the GeoTIFF stack output from Step 1.

3. Soil Moisture Estimation and Validation

The second R-script handles the estimation of relative soil moisture (Θ), calculated as the ratio of backscatter change to the maximum backscatter change (Δσ).

This relative soil moisture (Θ) is then normalized to field measurements using the field capacity and wilting point values to derive volumetric soil moisture (Θ_v).

In the final, temporal plot was generated by overlaying the measured in-situ soil moisture data with the Sentinel-1–based soil moisture estimates at selected sample points.

The results show strong model performance:

• Correlation coefficient (r): 0.75
• RMSE: 0.095 m³/m³

4. Soil Moisture Mapping

In the final step, a soil moisture map was generated using QGIS. This map, saved as a single GeoTIFF file, represents soil moisture distribution on a selected date, in this case it was 2019-03-11.

The visualization highlights spatial variation in soil moisture, specifically comparing the soil moisture between No-tillage vs Conventional agricultural practices.