Characterizing Soil Profile Salinization in Cotton Fields Using Landsat 8 Time-Series Data in Southern Xinjiang, China

Jiaqiang Wang, Bifeng Hu, Weiyang Liu, Defang Luo, Jie Peng2023 · DOI

TL;DR

This study demonstrates the feasibility of using Landsat 8 satellite imagery and the Cubist modeling approach to map soil profile salinization in cotton fields across southern Xinjiang, China. By integrating multi-year vegetation indices with ground-based electrical conductivity measurements, the research provides a cost-effective and scalable method for monitoring this critical issue in arid agricultural regions.

What Problem Does This Paper Address?

Soil salinization is a major threat to agricultural productivity, especially in arid regions like southern Xinjiang, China, which is a significant cotton-producing area. Severe salinization can lead to crop damage, reduced yields, and the degradation of fertile land. While traditional methods for assessing soil salinity are labor-intensive and time-consuming, and existing remote sensing approaches often focus only on surface salinity, there's a need for efficient, regional-scale methods to understand salinity distribution at different soil depths. This is crucial for optimizing irrigation practices and managing salinized farmland effectively.

Methods

The researchers employed a multi-pronged approach:

  • Ground-based Data Collection: Soil apparent conductivity (ECa) was measured using an EM38-MK2 device across 30 cotton fields. Soil samples were collected from these locations at three different depths (0–0.375 m, 0–0.75 m, and 0–1.00 m) to determine actual soil electrical conductivity (ECe).
  • Remote Sensing Data: Landsat 8 satellite images from 2013 to 2019 were acquired. These images were pre-processed for radiometric calibration, atmospheric correction, and clipping.
  • Feature Engineering: Seventeen vegetation spectral indices were calculated from the Landsat 8 data. To mitigate the impact of single-year variations and external factors like weather or pests, "extreme value" synthesis was performed, meaning the annual maximum or minimum composite value for each index was used. This approach aimed to capture more stable signals related to vegetation health and its interaction with soil conditions.
  • Modeling:
    • Random Forest (RF) Model: Initially, an RF model was used to establish a relationship between apparent conductivity (ground truth) and soil electrical conductivity at different depths. This helped in creating a representative soil electrical conductivity map for each pixel.
  • Cubist Model: The core of the study involved using the Cubist model, a machine learning algorithm based on decision trees, to predict soil electrical conductivity. This model was trained using the synthesized vegetation indices (as predictors) and the measured soil electrical conductivity at different depths (as the target variable). The Cubist model's ability to handle complex relationships and select important features was leveraged.

Datasets

  • Landsat 8 OLI Satellite Imagery: Acquired for the Alar Reclamation Area of southern Xinjiang, China, from 2013 to 2019. A total of 172 images were used.
  • Ground-based Soil Data:
    • Apparent Electrical Conductivity (ECa) data collected using an EM38-MK2 device.
  • Soil profile samples collected from 30 cotton fields, analyzed for soil electrical conductivity (ECe) at three depths: 0–0.375 m, 0–0.75 m, and 0–1.00 m. A total of 540 sets of ECa and corresponding soil profile samples were collected.
  • Digital Elevation Model (DEM): A 30 m spatial resolution DEM was also used as a potential modeling factor.

Key Insights and Ideas

  • Multi-year Extreme Vegetation Indices are Key: The study highlights that using the annual maximum or minimum values of vegetation indices, rather than single-year data, significantly improves the robustness of the salinization monitoring model. This approach helps to filter out transient noise and capture more consistent signals related to soil conditions.
  • Cubist Model's Efficacy: The Cubist modeling method proved effective in establishing relationships between vegetation indices and soil electrical conductivity at various depths, demonstrating its suitability for complex environmental modeling.
  • Depth-Dependent Salinization Patterns: The research clearly shows that soil salinization patterns differ significantly with depth. The topsoil (0–0.375 m) in the study area was largely nonsalinized after winter irrigation, while deeper layers (0–0.75 m and 0–1.00 m) exhibited more pronounced salinization, suggesting a trend of salt accumulation in deeper soil profiles.
  • Spatial Distribution of Salinity: The study identified specific areas within the cotton fields that are more severely affected by salinization, particularly in the north, northwest, south, and southeast of the study area. This information is vital for targeted interventions.

Key Results

  • Cubist Model Accuracy: The Cubist models developed for soil profile electrical conductivity at different depths showed high prediction accuracy.
  • For the 0–0.375 m depth, the validation set achieved an R² of 0.80, RMSE of 0.86 dS m⁻¹, and MAE of 0.55 dS m⁻¹.
  • For the 0–0.750 m depth, the validation set achieved an R² of 0.74, RMSE of 1.50 dS m⁻¹, and MAE of 1.08 dS m⁻¹.
  • For the 0–1.000 m depth, the validation set achieved an R² of 0.72, RMSE of 1.35 dS m⁻¹, and MAE of 0.98 dS m⁻¹.
  • Spatial Distribution Findings:
    • At the 0–0.375 m depth, approximately 92.24% of the cotton planting area was classified as nonsalinized.
  • At the 0–0.750 m depth, the nonsalinized area decreased to 40.19%, with mild and moderate salinization increasing significantly.
  • At the 0–1.000 m depth, nonsalinized soil accounted for 35.48%, with mild and moderate salinization further increasing.
  • Indicator Factors: The study identified key vegetation indices and DEM as important factors for modeling soil salinity at different depths, with specific indices like NG, LAI, COSRI, and NR being prominent in the Cubist model's partitioning rules.

Limitations and Future Work

  • Crop Specificity: The developed soil salinization estimation model is specific to cotton fields and may not be directly applicable to other crops due to differences in farming practices, environmental responses, and water requirements.
  • Satellite Data Specificity: The model is currently tailored for Landsat 8 OLI imagery and may require adaptation for other satellite sensors.
  • Future Directions: The authors suggest future research should focus on:
  • Synergistic application of multi-temporal and multi-sensor satellite data for enhanced monitoring.
  • Expanding the application to other staple or cash crops.
  • Further exploring soil salinization monitoring for the advancement of precision and smart agriculture.

Summary updated

BibTeX
@misc{wang2023characterizing,
  author = {Jiaqiang Wang and Bifeng Hu and Weiyang Liu and Defang Luo and Jie Peng},
  title = {{Characterizing Soil Profile Salinization in Cotton Fields Using Landsat 8 Time-Series Data in Southern Xinjiang, China}},
  year = {2023},
  doi = {10.3390/s23157003},
}
Figure