Improved digital soil mapping with multitemporal remotely sensed satellite data fusion: A case study in Iran

Area of Science:

• Soil Science
• Remote Sensing
• Geospatial Analysis

Background:

• Digital soil mapping (DSM) is crucial for environmental applications, often relying on static covariates from digital elevation models.
• Traditional DSM methods may not fully capture soil property variability over time.
• Incorporating temporal data from satellite imagery offers potential for improved soil property prediction.

Purpose of the Study:

• To evaluate the performance of digital soil mapping (DSM) using various covariate types: terrain derivatives (static), single-date satellite indices (limited dynamic), multitemporal satellite indices (dynamic), and fused covariates.
• To assess the impact of these covariates on predicting soil properties (organic carbon, sand content, calcium carbonate equivalent) and estimating prediction uncertainty.
• To compare the effectiveness of Cubist and Random Forest (RF) models in DSM with different covariate strategies.

Main Methods:

• Digital soil mapping (DSM) was employed to predict soil organic carbon (OC), sand content, and calcium carbonate equivalent (CCE).
• Covariates included terrain derivatives (static), single-date satellite indices, and multitemporal satellite indices (dynamic).
• Covariate fusion combined static and dynamic data; Cubist and Random Forest (RF) models were utilized for prediction and uncertainty estimation.

Main Results:

• The inclusion of single-date and multitemporal satellite indices significantly improved soil property predictions compared to terrain indices alone.
• Models using satellite data showed substantial increases in R-squared (e.g., 126% for OC in Cubist) and decreases in RMSE (e.g., 34% for OC in Cubist).
• Multitemporal satellite data fusion enhanced prediction accuracy and reduced uncertainty in soil property estimation and mapping.

Conclusions:

• Multitemporal satellite data fusion offers a significant advantage over static terrain indices for digital soil mapping.
• Dynamic covariates derived from satellite imagery substantially improve the prediction of key soil properties.
• This approach holds great potential for advancing soil modeling and mapping applications by reducing uncertainty and increasing accuracy.