Soil Moisture Inversion in Grassland Ecosystem Using Remote Sensing Considering Different Grazing Intensities and Growing Seasons

Jiahe Cui, Yuchi Wang, Yantao Wu, Zhiyong Li, Hao Li, Bailing Miao, Yongli Wang, Chengzhen Jia, Cunzhu LiangSustainability 2023 · DOI

TL;DR

This study refines the Water Cloud Model (WCM) for soil moisture (SM) estimation in grasslands by incorporating vegetation characteristics like grazing intensity and growing season. The research demonstrates that accounting for these factors improves SM prediction accuracy, highlighting the importance of vegetation biomass and height for accurate remote sensing-based SM retrieval.

What Problem Does This Paper Address?

Accurate monitoring of soil moisture (SM) is crucial for understanding grassland ecosystems, influencing vegetation growth, biomass accumulation, and overall water cycles. While microwave remote sensing, particularly using the Water Cloud Model (WCM), is a powerful tool for SM retrieval, its accuracy can be significantly affected by vegetation cover. Previous studies have explored various vegetation types and seasonal variations, but the specific impact of grazing intensity and detailed vegetation community information on SM inversion within grassland ecosystems remained unclear. This paper aims to bridge this gap by investigating how different grazing gradients and growing seasons influence SM retrieval using the WCM, and how to optimize the model by considering these factors.

Methods

The core of this research involves calibrating and validating the Water Cloud Model (WCM) for soil moisture inversion. The WCM is a well-established model that relates the backscattering coefficient from radar signals to soil moisture, accounting for vegetation's influence. The novelty here lies in the specific application and refinement of the WCM for grassland ecosystems by incorporating detailed vegetation community information.

Key methodological steps include:

  • Field Measurements: Collecting extensive ground truth data on vegetation characteristics (height, biomass, density, volumetric water content) and soil moisture across different grazing gradients (enclosure, light, moderate, heavy grazing) and growing seasons (early, mid-I, mid-II).
  • Remote Sensing Data Acquisition: Utilizing Sentinel-1A (SAR) and Sentinel-2 (optical) satellite data. Sentinel-1A provides the backscattering coefficient, while Sentinel-2 data is used to derive vegetation indices.
  • Vegetation Index Calculation: Calculating Normalized Difference Vegetation Index (NDVI) and two versions of the Normalized Difference Water Index (NDWI₁ and NDWI₂) from Sentinel-2 data. These indices serve as proxies for vegetation water content (VWC) in the WCM.
  • WCM Calibration: Using 80% of the collected field and remote sensing data to invert the WCM parameters (a, b, c, d) for different grazing gradients and growing seasons. This step aims to find the optimal model coefficients that best represent the relationship between radar backscatter, vegetation properties, and soil moisture under these specific conditions.
  • WCM Validation: Employing the remaining 20% of the data to evaluate the accuracy of the calibrated WCM using metrics like R² (coefficient of determination) and RMSE (root mean square error).

Datasets

  • Field Data: Collected from 2019 to 2021 at the Xilinhot National Climate Observatory in Inner Mongolia, China. This includes measurements of vegetation height, biomass, density, volumetric water content, and soil moisture. The data was collected across four grazing gradients (enclosure, light, moderate, heavy) and three growing seasons (early, mid-I, mid-II). A total of 900 valid samples were obtained, with 720 used for calibration and 180 for validation.
  • Remote Sensing Data:
    • Sentinel-1A: C-band SAR data (VV polarization, 10m resolution) acquired from July 2019 to October 2021.
  • Sentinel-2: Level-2A MSI optical imagery (10-60m resolution) acquired during the same period as Sentinel-1A.

Key Insights and Ideas

  • Vegetation Matters, A Lot: The study strongly emphasizes that vegetation characteristics are not just noise in SM retrieval but critical factors that significantly influence model accuracy. Ignoring them leads to suboptimal results.
  • Grazing Gradient is a Key Differentiator: Different levels of grazing intensity lead to distinct vegetation structures (height, biomass, density). This variation directly impacts how radar signals interact with the surface, necessitating tailored WCM parameters for each gradient.
  • Growing Season is Not Uniform: The WCM's performance varies across different stages of the growing season. Mid-growing season I appears to be the most optimal period for accurate SM retrieval, likely due to a balance in vegetation development and scattering properties.
  • Biomass and Height as Biomarkers: Vegetation biomass and height showed a significant correlation with vegetation indices. The study identified optimal ranges for biomass (around 121.1 g/m²) and height (around 18.6 cm) that yield the highest SM prediction accuracy, suggesting these can serve as valuable biomarkers.
  • NDWI₁ Outperforms NDWI₂ (Mostly): Generally, NDWI₁ provided better SM estimation accuracy than NDWI₂, though the specific optimal index could vary by grazing gradient. This highlights the importance of selecting the right vegetation index for the specific ecosystem and conditions.

Key Results

  • Overall Accuracy: The calibrated WCM achieved average R² values of 0.41 for different grazing gradients and 0.38 for different growing seasons during the evaluation phase. The corresponding average RMSE values were 0.024 and 0.030, respectively.
  • Grazing Gradient Performance: The light grazing (LG) gradient showed the highest accuracy (average R² of 0.68), with NDWI₂ being the most effective index for this specific condition (R² = 0.74). For other gradients (NG, MG, HG), NDVI generally provided better results.
  • Growing Season Performance: Mid-growing season I (MGS-I) yielded the highest accuracy (average R² of 0.51), with NDWI₁ showing slightly better performance (R² = 0.52).
  • Biomass and Height Correlation: Biomass was significantly correlated with NDVI (0.82), NDWI₁ (0.57), and NDWI₂ (0.53). NDVI was also significantly correlated with vegetation height (0.47).

Limitations and Future Work

  • Grassland Specificity: The study focused on a specific type of grassland (dominated by Stipa grandis and Leymus chinensis). The findings might not be directly transferable to other grassland types with different vegetation structures.
  • Model Accuracy: While the R² values are comparable to other studies, they are not exceptionally high (around 0.4). The authors acknowledge that the complex spectral characteristics of grassland vegetation can limit model accuracy.
  • Future Work Suggestions:
    • Detailed Classification: Implementing more detailed grassland classification methods could refine SM estimates.
  • Scattering Effect Investigation: Further research is needed to systematically investigate the scattering effect of vegetation with diverse community structures on microwave signals and its impact on SM prediction.
  • Inter-annual Variability: Estimating the robustness of SM prediction capacity across different years of vegetation growth is suggested.
  • Other Vegetation Indices: Exploring other vegetation indices or combinations thereof could potentially improve accuracy.

Summary updated

Abstract

Although vegetation community information such as grazing gradient, biomass, and density have been well characterized in typical grassland communities with Stipa grandis and Leymus chinensis as dominant species, their impact on the soil moisture (SM) inversion is still unclear. This study investigated the characteristics of a grassland vegetation community at different grazing gradients and growing seasons and its impact on SM inversion using remote sensing data. The water cloud model (WCM) was used for SM inversion, and both field and remote sensing data collected from 2019 to 2021 were used for calibration and prediction. The study found that the calibrated WCM achieved prediction results of SM inversion with average R2 values of 0.41 and 0.38 at different grazing gradients and growing seasons, respectively. Vegetation biomass and height were significantly correlated with vegetation indexes, and the highest model prediction accuracy was achieved for biomass and height around 121.1 g/m2 [102.3–139.9] and 18.6 cm [17.3–19.8], respectively. Generally, NDWI1 produced higher SM estimation accuracy than NDWI2. The growing season of vegetation also affects the accuracy of the WCM to retrieve SM, with the highest accuracy achieved in mid-growing season I. Therefore, the developed WCM with optimal height and biomass of vegetation communities can enhance the SM prediction capacity; it thus can be potentially used for SM prediction in typical grasslands.
BibTeX
@inproceedings{cui2023soil,
  author = {Jiahe Cui and Yuchi Wang and Yantao Wu and Zhiyong Li and Hao Li and Bailing Miao and Yongli Wang and Chengzhen Jia and Cunzhu Liang},
  title = {{Soil Moisture Inversion in Grassland Ecosystem Using Remote Sensing Considering Different Grazing Intensities and Growing Seasons}},
  booktitle = {Sustainability},
  year = {2023},
  doi = {10.3390/su15086515},
}
Figure