.png){kind=icon}
.png){kind=icon}
Return
A new approach to soil moisture estimation based on correlation between SAR and multispectral indices and seasonal pattern of soil moisture dynamics: a case study of Lake Urmia Basin
.png){kind=icon}
PRE
AI
S
K
K
G
S
DOI：10.26833/ijeg.1734366.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Radar satellite imagery has been widely used to obtain soil moisture (SM) estimates of high accuracy. Accurate information on surface soil moisture content under scalable conditions is important for hydrological and climatological applications. the aim of this study, the integration of multi-sensor satellite data to investigate the importance of features in various products and estimate soil moisture, is conducted using two machine learning models, Random Forest (RF) and Support Vector Regression (SVR), in the Lake Urmia basin. we used Sentinel-1 C-band Synthetic Aperture Radar (SAR) data, Sentinel-2 and Landsat-8 optical-thermal imagery, soil property maps (SoilGrids), and climate variables (FLDAS). at first analysis of correlation and regression was done seasonally for the four-year period to examine the importance and effectiveness of their use in estimating soil moisture. Then, the implementation of the model was done in two stages, using all the features (22) and 10 that were determined based on the performance of the model. The results show that soil organic carbon (SOC250) and radar indices governed winter and spring moisture dynamics, whereas vegetation indices dominated summer and autumn predictions, reflecting vegetation-climate-soil interactions. The Rand Forest model using all features had the highest accuracy of 0.88 in spring and the lowest in summer, with an accuracy of 0.80, while the SVR model had the lowest accuracy in summer (50.539) and the highest accuracy (0.628) in autumn, with the SVR results using the 10 most important features increasing by 0.1 R2 across all the variables. This increase in accuracy was observed in the RF model from 0.1 to 0.3, with the highest increase in accuracy in the summer. RF outperforms the SVR model in all evaluation metrics, including Mean Squared Error (MSE), R-2, and Mean Absolute Error (MAE), for both feature sets and in all seasons. The Normalized Vegetation Structural Difference Index (NVSDI) and Normalized Radar Vegetation Difference Index (NRVDI) indices have helped improve model accuracy by providing more combined and detailed information about specific soil and vegetation characteristics. normalized difference vegetation index (NDVI) has a specific focus on vegetation, while NVSDI and NRVDI provide more comprehensive and detailed environmental information. These findings demonstrate the potential of multi-sensor data integration seasonally, for soil moisture estimation, providing and critical insight for hydrological modeling, monitoring environmental and agricultural.
Keywords:
Soil Moisture
SAR Indices
Correlation Analysis
Machine Learning
lake Urmia basin
Journal
I
IF:
2.5
Papers:
160
Citations:
348
Organization
Cited Papers