Generating Daily Soil Moisture at 16 m Spatial Resolution Using a Spatiotemporal Fusion Model and Modified Perpendicular Drought Index

Xin Lu, Hongli Zhao, Yanyan Huang, Shuangmei Liu, Zelong Ma, Yunzhong Jiang, Wei Zhang, Chuan Zhao2022 · DOI

TL;DR

This paper introduces a novel method to generate daily soil moisture (SM) data at a high 16-meter spatial resolution by fusing a spatiotemporal fusion model (ESTARFM) with a modified perpendicular drought index (MPDI). This approach overcomes the limitations of existing low-resolution SM products, offering more detailed and accurate SM information crucial for applications like precision agriculture.

What Problem Does This Paper Address?

Soil moisture (SM) is a critical variable for understanding land surface processes and the global water cycle. While remote sensing offers a way to monitor SM globally, existing satellite-derived SM products, such as the ESA CCI SM, suffer from low spatial resolutions (typically >3 km). This coarse resolution limits their utility for applications requiring detailed spatial information, such as precision agriculture, localized water resource management, and detailed crop monitoring. Existing methods to downscale SM data often rely on land surface temperature (LST), which itself has limited spatial and temporal resolution, or face challenges in maintaining the temporal resolution of the original data. This paper aims to bridge this gap by providing daily SM data at a significantly finer spatial resolution (16 m) while preserving temporal accuracy.

Methods

The proposed method involves a three-step process:

  1. High-Resolution Surface Reflectance Generation: The Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM) is employed to fuse lower-resolution MODIS (500m) and higher-resolution GF6 (16m) surface reflectance data. This fusion generates a daily surface reflectance dataset at 16m resolution, effectively overcoming the "spatiotemporal contradiction" where high spatial resolution data often has low temporal resolution.
  2. Spatial Scale Conversion Factor (SSCF) Calculation: A Modified Perpendicular Drought Index (MPDI) is calculated using the fused 16m surface reflectance data. This MPDI is then used, along with an aggregated MPDI (at 25km resolution), to derive a Spatial Scale Conversion Factor (SSCF) dataset at 16m resolution. The MPDI is chosen for its strong correlation with SM and its ability to be derived from high-resolution surface reflectance.
  3. Spatially Downscaled SM Generation: The original low-resolution (25km) ESA CCI SM product is downscaled to 16m resolution using the derived SSCF dataset. The core idea is that the SSCF acts as a bridge, transferring the high spatial detail from the reflectance-derived MPDI to the coarser SM product.

Datasets

  • High-Resolution Optical Data: Two 16m GF6 images (May 2nd and June 3rd, 2019).
  • Low-Resolution Optical Data: Nine 500m MOD09GA images (May 1st, 2nd, 20th, 21st, 22nd, 23rd, 28th, June 1st, 11th, 2019).
  • Low-Resolution Soil Moisture Data: Nine 25km ESA CCI SM images (corresponding dates to MOD09GA).
  • In-Situ Soil Moisture Data: 151 ground-based soil moisture measurements collected at 10cm depth on four dates (May 1st, 21st, June 1st, 11th, 2019) from eight selected sites.

The study area comprised five cities in southern Hebei Province, China.

Key Insights and Ideas

  • MPDI as a Proxy for SM: The study leverages the strong correlation between the Modified Perpendicular Drought Index (MPDI) and soil moisture. This is a key insight, as MPDI can be derived from readily available high-resolution optical data, offering a more detailed spatial representation than traditional LST-based methods.
  • ESTARFM for Bridging Temporal Gaps: Using ESTARFM to generate daily high-resolution surface reflectance is crucial. It effectively addresses the temporal resolution limitations of high-resolution optical sensors like GF6, allowing for daily SM downscaling.
  • SSCF for Spatial Disaggregation: The SSCF, derived from MPDI, acts as a sophisticated disaggregation factor. It allows the spatial patterns observed in the high-resolution reflectance data to inform the distribution of soil moisture within the coarser CCI pixels.
  • Improved Spatial Detail: The downscaled 16m SM data visually and quantitatively show significantly finer spatial detail compared to the original 25km CCI SM, enabling the identification of smaller-scale variations in soil moisture.

Key Results

  • Improved Accuracy: The downscaled 16m SM data demonstrated higher accuracy when compared to in-situ measurements. Correlation coefficients (R) ranged from 0.45 to 0.67 for the downscaled data, compared to 0.33 to 0.54 for the original CCI SM. Root Mean Square Errors (RMSE) were also lower, ranging from 0.023 to 0.031 cm³/cm³ for the downscaled data versus 0.027 to 0.032 cm³/cm³ for the original CCI SM.
  • Enhanced Spatial Resolution: Visual comparisons (Figure 5) clearly show that the downscaled SM data captures finer spatial details and textures, better reflecting ground objects and variations compared to the coarse CCI SM.
  • Maintained Temporal Resolution: The time-series analysis (Figure 6) indicates that the downscaled SM data better reflects the dynamic changes in soil moisture over time, with more accurate dynamic changes and smaller average absolute differences (AAD) compared to the original CCI SM.

Limitations and Future Work

  • Spatial Resolution Ratio: The authors note that the R values were slightly lower than some previous studies, which they hypothesize might be related to the spatial resolution ratio between the original and downscaled data. As this ratio increases, the relevance between downscaled data and in-situ data might decrease.
  • Data Availability: The method relies on the availability of both high-resolution optical data (like GF6) and lower-resolution SM products (like CCI). Future work could explore applying this method with other sensor combinations.
  • Generalizability: While validated in a specific region (southern Hebei Province), further validation across diverse geographical and climatic zones would be beneficial to assess the method's broader applicability.
  • Cloud Cover: Although ESTARFM helps mitigate temporal gaps, persistent cloud cover can still impact the availability of high-quality optical data, potentially affecting the daily generation of surface reflectance.

The paper concludes that the proposed method offers a practical and effective way to generate high-resolution daily soil moisture data, which is crucial for various agricultural and environmental applications.

Summary updated

BibTeX
@misc{lu2022generating,
  author = {Xin Lu and Hongli Zhao and Yanyan Huang and Shuangmei Liu and Zelong Ma and Yunzhong Jiang and Wei Zhang and Chuan Zhao},
  title = {{Generating Daily Soil Moisture at 16 m Spatial Resolution Using a Spatiotemporal Fusion Model and Modified Perpendicular Drought Index}},
  year = {2022},
  doi = {10.3390/s22145366},
}
Figure