Soil Moisture Inversion Based on Data Augmentation Method Using Multi-Source Remote Sensing Data

Wang,Zhao,Guo,Yang,Li

DOI: https://doi.org/10.3390/rs15071899

IF: 5

2023-04-01

Remote Sensing

Abstract:Soil moisture is an important land environment characteristic that connects agriculture, ecology, and hydrology. Surface soil moisture (SSM) prediction can be used to plan irrigation, monitor water quality, manage water resources, and estimate agricultural production. Multi-source remote sensing is a crucial tool for assessing SSM in agricultural areas. The field-measured SSM sample data are required in model building and accuracy assessment of SSM inversion using remote sensing data. When the SSM samples are insufficient, the SSM inversion accuracy is severely affected. An SSM inversion method suitable for a small sample size was proposed. The alpha approximation method was employed to expand the measured SSM samples to offer more training data for SSM inversion models. Then, feature parameters were extracted from Sentinel-1 microwave and Sentinel-2 optical remote sensing data, and optimized using three methods, which were Pearson correlation analysis, random forest (RF), and principal component analysis. Then, three common machine learning models suitable for small sample training, which were RF, support vector regression, and genetic algorithm-back propagation neural network, were built to retrieve SSM. Comparison experiments were carried out between various feature optimization methods and machine learning models. The experimental results showed that after sample augmentation, SSM inversion accuracy was enhanced, and the combination of utilizing RF for feature screening and RF for SSM inversion had a higher accuracy, with a coefficient of determination of 0.7256, a root mean square error of 0.0539 cm3/cm3, and a mean absolute error of 0.0422 cm3/cm3, respectively. The proposed method was finally used to invert the regional SSM of the study area. The inversion results indicated that the proposed method had good performance in regional applications with a small sample size.

environmental sciences,imaging science & photographic technology,remote sensing,geosciences, multidisciplinary

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to improve the retrieval accuracy of surface soil moisture (SSM) based on multi - source remote sensing data when the sample size is small. Specifically, the article proposes a method that combines sample expansion, feature optimization, and machine - learning models to address the following challenges:

1. **Insufficient sample size**: When the SSM sample data measured on - site is insufficient, it will affect the accuracy of SSM retrieval. Therefore, a method is needed to increase the sample size, thereby improving the effect of model training.

2. **Feature parameter optimization**: The feature parameters extracted from multi - source remote sensing data may contain redundant information or noise, which will reduce the performance of the model. Therefore, it is necessary to optimize the feature parameters and select the most representative and relevant features for subsequent modeling.

3. **Selection of machine - learning models**: In the case of small samples, select a suitable machine - learning model and optimize its parameters to ensure the generalization ability and prediction accuracy of the model.

### Solutions

To solve the above problems, the author proposes the following steps:

1. **Data expansion**: Use the alpha approximation method to expand the SSM sample data measured on - site to provide more training data.

\[
\sigma^2_0 \approx \left| \frac{\alpha_{pp}(\varepsilon_s, \theta)}{\alpha_{pp}(\varepsilon_s, \theta)} \right|^2
\]

2. **Feature parameter extraction**: Extract feature parameters from Sentinel - 1 SAR data and Sentinel - 2 optical remote sensing data, including polarization feature parameters, vegetation indices, and surface roughness, etc.

3. **Feature optimization**: Use three methods, namely Pearson correlation analysis, random forest (RF), and principal component analysis (PCA), to optimize the extracted feature parameters and select the most appropriate feature subset.

4. **Model construction**: Construct three machine - learning models suitable for small - sample training, namely genetic algorithm - back - propagation neural network (GA - BP), support vector regression (SVR), and random forest (RF), and evaluate their retrieval accuracy.

5. **Accuracy assessment**: Compare the retrieval accuracy of different combinations of feature optimization methods and machine - learning models, and select the optimal combination for regional SSM retrieval.

Through these steps, the author aims to improve the SSM retrieval accuracy in the case of small samples and verify the effectiveness of the proposed method in practical applications. The experimental results show that after sample expansion, the SSM retrieval accuracy has been significantly improved. In particular, when using random forest for feature screening and SSM retrieval, a relatively high accuracy has been achieved, with the coefficient of determination \( R^2 = 0.7256 \), root - mean - square error \( RMSE = 0.0539 \, \text{cm}^3/\text{cm}^3 \), and mean absolute error \( MAE = 0.0422 \, \text{cm}^3/\text{cm}^3 \).