Machine learning algorithms for soil moisture estimation using Sentinel-1: Model development and implementation

Abstract

The present study provided the first-time comprehensive evaluation of 12 advanced statistical and machine learning (ML) algorithms for the Soil Moisture (SM) estimation from dual polarimetric Sentinel-1 radar backscatter. The ML algorithms namely support vector machine (SVM) with linear, polynomial, radial and sigmoid kernel, random forest (RF), multi-layer perceptron (MLP), radial basis function (RBF), Wang and Mendel’s (WM), subtractive clustering (SBC), adaptive neuro fuzzy inference system (ANFIS), hybrid fuzzy interference system (HyFIS), and dynamic evolving neural fuzzy inference system (DENFIS) were used. Extensive field samplings were performed for collection of in-situ SM data and other parameters from the selected sites for seven different dates and at two different locations (Varanasi and Guntur District, India), concurrent to Sentinel-1 overpasses. The backscattering coefficients were considered as input variables and SM as output variable for the training, validation and testing of the ML algorithms. The site at Varanasi was used for the training, validation and testing of the models. On the other hand, the Guntur site was used as an independent site for checking the model performance, before finalizing the algorithms. The performances of different trained algorithms were evaluated in terms of correlation coefficient (r), root mean square error (RMSE) (in m³/m³) and bias (in m³/m³). The study identified the RF, SBC and ANFIS as the top three best performing models with comparable and promising SM estimation. In order to test the robustness of these best models (RF, SBC and ANFIS), further performance analysis was performed to the independent datasets of the Varanasi and Guntur test sites, which indicates that the performance of these three models were consistent and SBC can be recommended as the best among all for SM estimation.

Introduction

Soil moisture (SM) plays an important role for the proper growth of crops and agricultural productions (Petropoulos et al., 2015, Srivastava et al., 2015). It is an important variable required to understand hydrological cycle and land surface fluxes (Srivastava et al., 2019, Srivastava et al., 2014). Monitoring of SM at a regular time interval over cropped surfaces are important for the effective irrigation management and development (Srivastava et al., 2013, Suman et al., 2020). Bazzi et al. (2019) also recognized the potential use of SM for detection of heavy rainfall in the south of the France by correlating SM and rainfall data. Another study also investigate the relative SM indicators for reclamation of wetlands sites in previously mined oil sands in Alberta, Canada (Zakharov et al., 2020). Radar can be used for SM estimation by using the backscatter measurements at different polarization. From the research, it is evident that the microwave response at low-frequency (P to L-band) are sensitive towards the SM content over bare surface as well as vegetated surfaces, because of its higher penetration capability (Shi et al., 1997). However, very few synthetic-aperture radar (SAR) systems onboard remote sensing satellites e.g., Japanese Earth Resources Satellite 1 (JERS-1), Advanced Land Observation Satellite 1 (ALOS-1) and ALOS-2 are operating at low frequency. The high-frequency C and X bands i.e., Sentinel-1, Radar Imaging Satellite 1 (RISAT-1), RADARSAT-1 & 2, COnstellation of small Satellites for the Mediterranean basin Observation- SkyMed (COSMO-SkyMed) and TerraSAR-X operated SAR systems also provide a significantly good results in the accurate retrieval of SM. These high-frequency operated SAR systems were frequently used in various studies for the mapping and retrieval of SM (Kumar et al., 2019, Paloscia et al., 2013, Prasad et al., 2009) and crop monitoring (Kumar et al., 2018, Navarro et al., 2016).

Various empirical, semi empirical and physical based models have been developed and successfully employed to retrieve the SM over bare soil surfaces (Baghdadi et al., 2017, Dave et al., 2021). Some physical models namely physical optical (PO), geometric optics (GO), small perturbation model (SPM), and advanced integrated equation model (AIEM) are useful for the retrieval of SM up to a limited range of surface roughness & surface characteristic. Attema & Ulaby (1978) introduced a radiative transfer theory based semi empirical model namely water cloud model (WCM) for the retrieval of crop covered SM and vegetation parameters and it is later extended by various authors (Ulaby et al., 1984). However, the semi empirical and physical based modelling approaches is very complex due to the less understanding about radiative transfer-based microwave response for the bare and vegetated soil surfaces. These models also have larger number of parameters. Therefore, there is a need of parameter free and less complex machine learning (ML) modelling approaches for the retrieval of SM from bare and vegetated soil surfaces.

The ML techniques are very popular among the scientist community during past two decades, which may overcome the limitations of above discussed models in retrieval of SM using radar backscattering (Gupta et al., 2015, Gupta et al., 2017, Kumar et al., 2019, Srivastava, 2017, Srivastava et al., 2013). Kumar et al. (2019) retrieved the wheat, barley and corn underneath SM, using dual polarimetric Sentinel-1A microwave satellite data by support vector machine (SVM), random forest (RF) and artificial neural network (ANN) models. They showed that the performance of the ANN model is found lower in comparison to SVM and RF models. Gupta et al., (2017) compared the Back Propagation ANN, Radial Basis Function (RBF) neural network, Generalized Regression (GR) neural network for the SM estimation using the bistatic scatterometer data. They suggested that the ML algorithms may provide some promising results. Chai et al., (2010) used the ANN for the retrieval of SM using microwave data with the spatial variability information of SM. Notarnicola et al., (2008) reported the comparison between neural networks and Bayesian algorithms for the retrieval of SM using scatterometer and radiometer data for a variety of agricultural field. They found that the neural network approach is better than Bayesian, when algorithm is trained with more parameters. Liu et al. (2017) proposed a new SM retrieval approach based on ultra-wide echoes and adaptive neuro fuzzy inference system (ANFIS) algorithms.

A recent study by Greifeneder et al., (2021) explores the possibility of ML based approach and google earth engine for real time cloud based mapping SM at high spatial resolution (50 m) by integrating data from the Landsat-8 optical and thermal images, Sentinel-1 SAR images, and modelled data. Training and independent validation dataset were taken from International Soil Moisture Network. Liu et al. (2021a) presented an approach to retrieve SM over farmland with the combination of SAR and optical data from Sentinel-1 and Sentinel-2, respectively. To establish the relationship between the various features and SM, two ML algorithms, viz. Support vector regression (SVR) and GR neural network models were used. They also used a convolutional neural network regressor (CNNR) to extract deep features from remote sensing data. They conclude that the CNNR model with optimal feature combination can promisingly increase the SM retrieval accuracy. Liu et al. (2021b) used physical models to estimate SM in bare and vegetation covered soil surfaces using dual-polarized Sentinel-1A backscattering coefficients (VV and VH). WCM model was used to remove the vegetation effect from the radar backscattering coefficients. Modified SM monitoring index (MSMMI) and modified perpendicular drought index (MPDI) from optical source, i.e., Sentinel-2A data was also used to estimate SM. And then, they used some ML models to integrate the optical and SAR data ability for improved SM estimation. These models were the GR, SVR, RF regression, and deep neural network (DNN). They concluded that the integration improves the SM estimation accuracy.

Most of the researchers have used some common ML algorithms, however, the performance of several other ML algorithms like Wang and Mendel’s (WM), Subtractive clustering (SBC), Hybrid neural fuzzy inference system (HyFIS), Dynamic evolving neural fuzzy inference system (DENFIS) were not yet tested for the retrieval of SM using SAR data. Thus, a rigorous analysis of various ML algorithms for SM retrieval is needed to understand their performance in different scenarios and to check the robustness of each model for high-resolution SM estimation. In the purview of the abovementioned problems, the present study focused on (1) Evaluation of different statistical and ML algorithms for SM estimation using Sentinel-1A data (2) Rigorous optimization of the models with respect to different model parameters (3) Validation of the model at the different spatial and temporal scales for large scale implementation.