Dynamic Monitoring of Surface Soil Moisture Fluctuations Using Synthetic Aperture Radar and Data- Driven Algorithms

doi:10.21203/rs.3.rs-4958324/v1

Download PDF

Research Article

Dynamic Monitoring of Surface Soil Moisture Fluctuations Using Synthetic Aperture Radar and Data- Driven Algorithms

https://doi.org/10.21203/rs.3.rs-4958324/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

The primary goal of the study is to employ SAR data and data driven approaches to model Surface Soil Moisture (SSM) for cultivable bare fields. Three experimental test plots were selected which are basically cultivable but due water deficiency the plots are left bare. Samples for surface soil moisture, soil surface roughness and bulk density are collected from test plots in grid sampling manner in parallel with SAR data pass over study area. Sentinel-1A data is pre-processed and each field sampling grid backscattering energy values are obtained. Surface roughness, dielectric constant and backscattered energy were used as input features to model SSM using RF, SVR and BPANN. We observed that BPANN outperformed SVR and RF by accurately predicting soil moisture with RMSE = 0.077, bias = 0.013, and R = 0.94. This study sheds light on small scale agricultural lands which are deficient of water to support crop growth.

Surface Soil Moisture

SAR

SVR

ANN

Surface Soil Moisture (SSM) represents the amount of water in the top 5 to 15 cm of the soil layer (Adab et al., 2020a; Wang & Qu, 2009). Even though it only makes up just a small fraction of the world's water reserves, water in this thin film of soil account for hydrological, biochemical, physiological, agricultural, and other earth activities (Lunt et al., 2005). Assessment of SSM can be done by point measurements and remote sensing methods (Owe et al., 2001). The Point measurement techniques include gravimetric, neutron moisture meter, time domain reflectometry, capacitance, FDR, tensiometer, and hygrometric methods; out of these methods, gravimetric traditional technique is accurate and but it has limited coverage, slow and time-consuming. (Dobriyal et al., 2012; Rasheed et al., 2022).

A different approach to model, monitor and measure SSM at different scales using recurring methods has been offered by technological advances in the area of Satellite sensing (Li et al., 2021). Various remote sensing methods, including optical, thermal, and microwave ones, have demonstrated their ability to extract soil moisture from the earth's surfaces (Mitran et al., 2020) but the onboard microwave remote sensing equipment demonstrated an exceptional sensitivity and capabilities to track soil dielectric characteristics (Barrett et al., 2009; Mao et al., 2023). Satellites equipped with microwave sensors can provide accurate surface Soil Moisture Content (SMC) estimations that are crucial for monitoring geo-hazards like drought and flood prediction as well as investigating socio-economic activity.

Despite of different satellite sensing-based measurement methodologies, Synthetic Aperture Radar (SAR) data have shown enormous promise in giving better estimates of soil moisture at both the global and regional scale. (Kseneman et al., 2011; Ma et al., 2020; Pierdicca et al., 2013). Backscatter coefficient when applied to bare soil, is strongly correlated with soil moisture through dielectric constant (Altese et al., 1996; Das & Paul, 2015; Ezzahar et al., 2020). As a result, numerous backscattering models with differing levels of complexity, accuracy, and validity have developed in order to extract SSM estimates over bare soil. In previous studies many empirical, semi empirical, physical models like Dubois model, Oh model, IEM etc., have been used for the SSM retrieval. Outlier data, nonlinearity, heteroscedasticity, and multicollinearity are some of the statistical assumptions that must be made for these classical approaches, which may restrict their usefulness (Adab et al., 2020b).

Thus, to recover SSM from bare and vegetated soil surfaces, parameter-free and better-understood Machine Learning (ML) modelling techniques are required. To generate reliable soil moisture estimations and steer clear of the aforementioned problems, it is believed that an efficient data-driven model can relate the dependent and independent variable to the obtain desired output and is not computationally demanding is required (Ahmad et al., 2010; Shen et al., 2018). In this context, Machine learning (ML) techniques have demonstrated their adaptability in a variety of conditions by utilizing optical and SAR data, combining remotely sensed information with field information, and also by exploiting the data obtained from a UAV platform (Ali et al., 2015). In the past two decades, ML approaches have gained enormous popularity among researchers. They have the potential to get over the drawbacks of the models previously mentioned for retrieving SSM using radar backscattering. Several ML methods are utilised in last few years to retrieve surface soil moisture at catchment and global scale such as Support Vector Regression (SVR), Random Forest (RF), Artificial Neural Network (ANN), Convolutional Neural Network Regressor (CNNR), etc. Very limited studies were observed in testing ML methods at plot or regional scale. Hence, there is a need to check ML methods to retrieve surface soil moisture in bare agricultural experimental plots. (Kumar et al., 2018; Srivastava et al., 2013)

The objective of this study is to develop SSM model based on ML method and check efficacy of ML methods in plot or field scale. This study will be benefitted to the marginal farmers/stakeholders to utilise the water effectively to sustain the growth of crops in bare fields.

The experimental test plot is cultivable bare field in the Mysuru, Karnataka, India (Plot 1, 2, & 3), due to insufficient water to grow crop in the field. This region falls under semi-arid climatic conditions with average annual rainfall of 700-800mm. All of the research area's plots are situated on the flat Indian Deccan Plateau. Salient features of study area plots have been given in Table 1.

Table 1

Salient features of study area
Features	Plot 1	Plot 2	Plot 3
Soil type	Sandy loam & silty clay	Sandy loam and silty clay	Sandy soil
Area(acres)	1.43 acres	0.53 acres	1.38 acres
Irrigation Pattern	Irrigation by groundwater	Irrigation by groundwater	Rain-fed

The Mysuru region is in altitude around 760 m above mean sea level. Figure 1 shows the Location of study area and Fig. 2 shows Google Earth image of study area and the sample grid. Groundwater is utilised to irrigate agricultural areas in this area. To minimize the irrigation dependency on groundwater in this area, soil moisture researches are need of the time.

In the current investigation, two distinct types of data are used. Remote sensing data comprises Sentinel-1A, Sentinel-2A, SRTM DEM, and Google Earth imagery. Field data provides information about surface soil moisture and roughness.

3.1 Field data

Field soil data was collected from 21 March 2022 to 1 June 2022 and 9 April 2023 to 21 April 2023 from Plot 1, 2 and 3 respectively in coordination with satellite over the study area at 12 day frequency. Field photographs during soil data collection and acquisition dates is shown in Fig. 3. At precisely 6:30 a.m. ± ½ hours, when Sentinel-1A SAR crosses over the investigation's plot, soil data were taken from every square. Each research square yielded two sets of soil samples. To assess the gravimetric moisture content, a set of samples (0–5 cm) was gathered and marked in zip-lock bags. In order to assess soil texture and modify soil moisture content to gravimetric to volumetric values, another set of samples was taken using the core cutter to ascertain bulk density. (IS 2720-4, 1985). A total of 36 surface soil samples from plot-3, 24 from plot-2, and 56 from plot-1 were taken in each satellite visit over study area. To test SSM in bare fields, 552 field samples were collected from all three plots during the research period. The soil's surface roughness is assessed using a roller chain technique. Surface roughness was assessed at each sampling grid every day as the satellite passed.

Table 2

Acquisition dates of field and Sentinel-1A data
Data acquisition dates of Plot 1 & 2	Data acquisition dates of Plot 3
21/03/2022; 02/04/2022; 26/04/2022; 08/05/2022; 20/05/2022; 01/06/2022	09/04/2023; 21/04/2023

3.2 Rainfall

The weekly rainfall data was obtained from Karnataka State Natural Disaster Monitoring Centre (KSNDMC) website (https://www.ksndmc.org/) and rainfall corresponding to data collected dates are given in the Fig. 4.

Figure 4, notifies that no rainfall recorded over the test plots during the sampling dates except on 21-03-2022, whereas the rainfall occurred after the data collection. Hence the rainfall did not alter the SSM in the given date.

3.3 Sentinel-1A

The Sentinel-1 mission is constellation of 2 helio-synchronous satellites (1A & 1B) that operate in C-band SAR imaging to acquire earth information in any weather and light condition. Sentinel-1 is intended to perform tasks in a configured, conflict-free mode, collecting high-resolution data of the planet's landmasses, coastlines, and shipping lanes in addition to small images of the whole ocean. This guarantees the integrity of archived data libraries created for tasks based on big time series, as well as the dependability of service necessary for operational services. The Sentinel-1A data for this investigation were obtained from the Alaska Land Facility (ASF) server. Table 3 summarises the Sentinel-1A data.

Table 3

Characteristics of Sentinel-1A data
Frequency Band	Resolution	Revisit time	Polarization	Orbit direction	Swath(km)
C	10*10 m	12 days	VV and VH	Descending	100–150

3.4 Sentinel-2A

Sentinel-2A is a European Space Agency (ESA) imaging mission with a large swath, high resolution, and multi-spectral capabilities. Sentinel-2A L1C data is acquired via Alaska Land Facility (ASF) server. Sentinel-2A works with spatial resolution of 10*10m, swath width of 290km and repitivity of 10 days. To ascertain the impact of vegetation, particularly shrubs, on radar backscattered energy, the NDVI values of test plots are produced using the Sentinel-2A data.

3.5 SRTM data

All plots' SRTM DEMs were procured from USGS Earth Explorer website and the resolution of the DEMs is 30 x 30m. The SRTM data is used to examine the SAR constraints.

3.6 Google Earth data

With the use of GIS data, satellite images, and other image sources, Google Earth is an online application that allows users to view the world in three dimensions. The field data collection is planned using information from Google Earth.

The research work starts with selection of suitable test plots. Figure 5 displays the flowchart of the work conducted. It describes the details of data collected, image pre-processing, field data analysis and machine learning algorithms used for modelling SSM.

4.1 Field data

In the section 3.1, field data collection is discussed through this samples soil moisture is calculated according IS code. The Root Mean Square (RMS) height of the soil surface roughness was measured at the site using a roller chain method (Saleh, 1994). This is the quickest and most practical method of determining surface roughness. It is based on the observation that when a chain of a particular length L₁ is put on the surface, horizontal length reduces as Soil Surface Roughness (SSR) increases. SSR for each test plot is calculated using Eq. (1)

SSR = (1 – L₂/L₁)*100 (1)

Surface Soil Roughness is measured in field in direction 90⁰ to ridges provided the aggregate-caused roughness (random roughness). The roller chain's L₁ length and the chain's L₂ linear distance owing to roughness are the same thing. The statistical parameter, RMS (vertical variation), was used to explain SSR, which is calculated using the Eq. (2)

RMS= $\:\sqrt{\frac{1}{n-1}\left[{\left({z}_{i}-z\right)}^{2}\right]}$ (2)

where,

z_i = height of the soil surface

z = mean height of the soil surface

n = number of samples taken into analysis.

4.2 Field Data Analysis

To understand the relationship between backscattered energy (σ°) and surface soil moisture without any vegetation effects, three experimental bare field plots are selected. In a single satellite visit, 56 samples from plot 1, 24 samples from plot 2, and 36 samples from plot 3 were collected. Figure 6 (a), (b) & (c) displays the backscattered energy and soil moisture variations over time within the plots.

The box plot in Figs. 6 (a) and (b) shows the spatiotemporal fluctuations of soil moisture in plots 1 and 2. Due to rainfall event on 20/04/2022 and 06/05/2022, the surface soil moisture on 21/03/2022 and 08/05/2022 has showed increase soil moisture levels. Because of no rainfall and irrigation activity on 02/04/2022, 26/04/2022, and 01/06/2022, the surface soil moisture is low. The spatiotemporal fluctuation of soil moisture in Plot 3 is shown in Fig. 7(c) box plot, as there was no rainfall as well irrigation during the acquisition dates, which has resulted in less moisture content compared to other soil moisture values.

4.3 Image Processing

Sentinel-1A data was obtained for the aforementioned date (Table 2), and it was pre-processed with the Sentinel Application Platform (SNAP) software. The procured images were processed using functions like data import, orbit file application, radiometric calibration, multi-looking, speckle filter, geometric terrain correction, and data conversion (linear to dB).

Data import translates Ground Range Detected (GRD) data into software-compatible data, and to speed up processing, a subset of the research region is chosen. When a precise orbit is applied by SNAP, it enables precise satellite location and the pixel values are converted to radiometrically calibrated SAR backscatter through the calibration procedure. The modified Lee speckle filter is applied to eliminate unwanted noise (speckle) with window size of 7*7, (Filipponi, 2019). Speckle does not get conveyed in following steps when a technique like this is used early in the SAR data processing process. Then the Range doppler terrain correction make the geometric representation of the image as true to reality as possible (Loew & Mauser, 2007). In this study SRTM DEM used for terrain correction. The last step of pre-processing is conversion of backscattering coefficient values to decibels (dB) values using data conversion.

4.4 NDVI Calculation

Sentinel-2A image was procured from Copernicus open access hub website. Sentinel-2A image was used to calculate NDVI for the study area by band math operation in QGIS. The soil and vegetation reflectivity in the red and near-infrared bands were used to calculate the NDVI values of field as shown in Eq. (3)

NDVI= $\:\frac{B8-B4}{B8+B4}$ (3)

Spatiotemporal variation of NDVI across the plots are shown in the Fig. 7 (a), (b) & (c) for plot 1, 2 &3 respectively. The NDVI values are less than 0.4 hence the experimental plots are less affected by vegetation. But there are some outliers because of presence of tree and shrubs beside the plots.

4.5 Model development using Machine Learning algorithms

Modelling SSM involves a multi-step approach, leveraging the strengths of Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Network (ANN) models to achieve more accurate predictions. The process commences with data preprocessing and feature engineering, where the soil moisture data is carefully prepared and relevant features are selected. SVR is a potent machine learning method that can handle data that can be separated into linear and non-linear categories (Trafalis & Gilbert, 2007a). In order to increase accuracy and prevent overfitting, RF is an ensemble learning technique that creates numerous decision-trees during the training process and combines predictions. (Mehta & Patnaik, 2021). ANN composed of interconnected nodes, organised into layers with each connection between the nodes having an associated weight. These are some of the significant ML algorithms that are chosen due to their diverse modelling techniques and potential effectiveness in capturing complex soil moisture dynamics. After the selection, each model undergoes rigorous training using the pre-processed data, learning distinct patterns and relationships between soil moisture and environmental factors. Each model learns to capture unique patterns and relationships between soil moisture and environmental features (Liu et al., 2022). Once trained, these models provide individual predictions, which are then aggregated to generate a more reliable and accurate estimation of soil moisture levels (Carranza et al., 2021). Figure 8 shows flow chart steps involved in developing soil moisture retrieval model using ML algorithms.

56 surface soil samples from Plot 1, 24 surface soil samples from Plot 2, and 36 surface soil samples from Plot 3 were obtained during a single satellite visit. For the purpose of assessing SSM in bare soil, a total of 552 field soil samples were obtained from the field. Here, the SVR, RF, and ANN models were trained on 386 samples and tested on 166 samples.

i. Support Vector Regression

SVM model has demonstrated its reliability for tasks involving classification and regression (Kumar et al., 2015). The primary focus is on class borders, and a non-linear transformation is used to a kernel function to map the input space created by independent variables (Shafizadeh-Moghadam et al., 2017). The most often utilized kernel functions are linear, polynomial and radial basis function kernel. Maximizing the margin allows for a more generalized solution with less overfitting. The model can handle big data spaces and training datasets (Cervantes et al., 2020). SVR is used to define the true-valued result function for the aforementioned independent input variables.

ii. Artificial Neural Network (ANN) Regression Model:

In a Back Propagation Artificial Neural Network (BPANN), input, hidden, and output layers comprise the basic processing units (neuron) (Neaupane & Achet, 2004a). The backscattered values and the surface soil moisture content were the input and output for the ANN models, respectively. Each node of the hidden layer receives information from the input level in the forward direction along with synaptic weights. At every node, these weighted inputs are added. Each hidden layer calculates the output values based on these weighted sums using either linear or non-linear sigmoidal transfer functions (Sunar Erbek et al., 2004; Gupta et al., 2017).

iii. Random Forest

Random forest is an ensemble learning approach that constructs several Decision Trees (DT) during the training process and combine the prediction to improve the accuracy (Guo et al., 2022). The process of building a DT starts entirely as boot strap sample as root node. During training recursive splitting will be employed to find best features and split points, which maximizes homogeneity and minimizes MSE (Grubinger et al., 2014) To optimize the model performance we have to adjust hyper parameters such as number of trees, tree depth, minimum samples per leaf, which prevents over fitting and enhances generalization (Leiva et al., 2019).

Datasets from all three experimental plots are being considered for analysis. The effective parameters taken into account to calculate SSM based on susceptibility of backscattered energy$\:({\sigma\:}^{0}$) are discussed in this section. Additionally, it describes the modelling of SSM using RF, SVR, BPANN.

5.1 Effect of vegetation on $\:{\varvec{\sigma\:}}^{0}$

Figure 9 demonstrates that the relationship between NDVI and backscattered energy is less strong, with R² = 0.07 for $\:{\sigma\:}_{vv}^{0}$ and R² = 0.20 for$\:\:{\sigma\:}_{vh}^{0}$. NDVI is not showing any relation with backscattered energy and NDVI values are within the range i.e., NDVI < 0.4. Hence NDVI is not considered as effective parameter for development of model. (Zhang et al., 2018).

5.2 Effect of Soil Moisture on$\:\:{\varvec{\sigma\:}}^{0}$

The relation between VV polarized and VH polarized backscattered energy and surface soil moisture is shown in Fig. 10 for data that were gathered at 12-day intervals. With an R² of 0.43 for VV polarization and 0.42 for VH polarization, backscattered energy of processed radar data is strongly dependent on SSM. The findings demonstrate that the soil moisture measured in the field has the highest correlation with the soil backscattering coefficient for VV polarisation. This is due to the fact that cross-polarization has lower penetrability than co-polarization and is more vulnerable to obstruction from vegetation (Rawat et al., 2018; Gururaj et al., 2021).

5.3 Effect of soil surface roughness on $\:{\varvec{\sigma\:}}^{0}$

The specifics of SSR measurement are explained in the preceding paragraph. SSR in the study area was determined, and the correlation (Fig. 11) demonstrates that backscattered values is somewhat reliant on SSR, with R² values of 0.14 and 0.1 for VV and VH polarization, respectively. There isn't much fluctuation in the surface soil roughness over the data collecting period because the plot is bare ground (not ploughed). Hence the effect of roughness on backscattered energy is low and neglected.

5.4 Effect of dielectric constant on $\:{\varvec{\sigma\:}}^{0}$

The dielectric constants of each sample were determined using field data such as soil moisture content and texture (more especially, the proportion of sand, silt, and clay). For a certain microwave frequency, $\:{\sigma\:}^{0}\:$can be determined using Eq. (4) according to (Hallikainen et al., 1985).

$$\:\begin{array}{c}\epsilon\:=\left({a}_{0}+{a}_{1}s+{a}_{2}c\right)+\left({b}_{0}+{b}_{1}s+{b}_{2}c\right){M}_{v}+\\\:\left({c}_{0}+\:{c}_{1}s+{c}_{2}c\right){M}_{v}^{2}\end{array}$$

Where, ε = dielectric constant

s = percentage of sand by weight

c = percentage of clay by weight

$\:{\:a}_{i}$ , $\:{b}_{i}$, and $\:{c}_{i}$ = frequency-dependent coefficients adopted

$\:{M}_{v}$ = volumetric soil moisture content.

Figure 12 shows that there is a positive relation between dielectric constant and backscattered energy with R² = 0.27 for $\:{\sigma\:}_{vh}^{0}$ and R² = 0.31 for $\:{\sigma\:}_{vv}^{0}$.

5.5 Retrieval of bare soil moisture using RFR, SVR and ANNR mode

The trial and error method, which is the most commonly used approach for obtaining the optimized model's parameters, is employed in the current work to optimize models.

To get ideal parameters, each method's parameters are adjusted. The radial basis function kernel of the SVM model is optimised by the adjustment of relevant parameters derived from the cost, epsilon, and gamma lists. The two input parameters that the RF model was tweaked over were mtry and ntree. The scatter plots between observed and retrieved soil moisture in bare field is shown in the Fig. 13, 14, and 15 using SVR, RF and BPANN models respectively.

Table 4

Accuracy assessment of surface soil moisture model
Methods	RMSE(m³m^− 3)	Bias (m³ m^− 3)	R	MAE
ANN	0.077	0.0137	0.94	0.0572
SVR	3.944	-0.197	0.94	2.847
RF	3.671	-0.695	0.95	2.653
RMSE = Root mean square error
R² = Coefficient of determination
MAE = Mean absolute error

From above accuracy assessment Table 4, It is evident that ANN outperformed SVR, RF with RMSE = 0.077, bias = 0.013 and R = 0.94 for the estimation of soil moisture in bare fields for the combined plot.

The Taylors plot, which was used to compare the model quality, is shown in Fig. 16. Using observed data, the Taylors plot's red rectangle along the x-axis is referred to as the reference point. The study's additional circle, triangular, and rectangular points were created using estimated data from RF, ANN and SVR models. Perfect match between the observed and estimated data is represented by points that are closer to the reference point. It also shows a propensity for data to be over- or under-estimated. Overestimation will occur if the estimated values' standard deviation is greater than the observed values' standard deviation, and vice versa. With the objective to estimate the surface soil moisture, ANN was shown to be more accurate since it was closer to the reference point.

ANN's superior performance over SVR and RF in soil moisture estimation stems from its capacity to model intricate non-linear relationships and capture feature interactions effectively. Its flexible architecture enables adaptation to the complexity of the problem, while representation learning aids in learning abstract features from the data (Gupta et al., 2017). Proper data preprocessing and regularization prevent overfitting and enhance generalization. With large training datasets, ANN leverages ample data to discern patterns, yielding improved results. Moreover, hyper-parameter tuning optimizes the model, and ensembling combines its strengths with other models, making ANN a powerful choice for accurate soil moisture estimation with the given input features (Singh & Gaurav, 2023).

In this study, assessment of SSM was calculated for bare land test plots. Soil samples were collected, and lab testing was executed. Sentinel-1A used to extract σ°_VV & σ°_VH. Effective parameters are selected for modelling SSM using various ML algorithms. SSR, dielectric constant, σ°_VV & σ°_VH, determined as effective parameters used as input features for the ML models (SVR, RF, ANN), whilst in-situ soil moisture is provided as the target variable.

The VH polarization (R² = 0.2) showed positive relationship with vegetation than VV polarization (R² = 0.07)
The VV polarization exhibited a greater relationship with SSM than VH polarization due to its greater penetration strength, with R² = 0.30 for σ°_VV and R² = 0.01 for σ°_VH
SSM retrieval using ML algorithms, ANN performed well by surpassing RF and SVR, with RMSE = 0.077, bias = 0.013 and R = 0.94
RF & SVR, compared to ANN having greater RMSE and positive bias suggesting a tendency for overestimation and indicates that it is less accurate in predicting SSM values.

This study focuses on monitoring SSM and managing irrigation water over small-scale agricultural lands and sets a guidelines in application of Machine Learning applications for SSM retrieval using SAR based studies in small-scale agricultural fields.

Conflict of interest:

On behalf of all authors, the corresponding author states that there is no conflict of interest.

Funding:

No funding received.

Author Contribution

Hrushikesh is the post graduate student who conducted the work under the guidance of Punithraj G. Abhsihek A Pathak helped in preparation of the maniscript.

Data availability:

Data can be shared upon request.

Adab, H., Morbidelli, R., Saltalippi, C., Moradian, M., & Ghalhari, G. A. F. (2020a). Machine learning to estimate surface soil moisture from remote sensing data. Water (Switzerland), 12(11), 1–28. https://doi.org/10.3390/w12113223
Adab, H., Morbidelli, R., Saltalippi, C., Moradian, M., & Ghalhari, G. A. F. (2020b). Machine learning to estimate surface soil moisture from remote sensing data. Water (Switzerland), 12(11), 1–28. https://doi.org/10.3390/w12113223
Ahmad, S., Kalra, A., & Stephen, H. (2010). Estimating soil moisture using remote sensing data: A machine learning approach. Advances in Water Resources, 33(1), 69–80. https://doi.org/10.1016/j.advwatres.2009.10.008
Ali, I., Greifeneder, F., Stamenkovic, J., Neumann, M., & Notarnicola, C. (2015). Review of machine learning approaches for biomass and soil moisture retrievals from remote sensing data. In Remote Sensing (Vol. 7, Issue 12, pp. 16398–16421). MDPI AG. https://doi.org/10.3390/rs71215841
Altese, E., Bolognani, O., Mancini, M., & Troch, P. A. (1996). Retrieving soil moisture over bare soil from ERS 1 synthetic aperture radar data: Sensitivity analysis based on a theoretical surface scattering model and field data. Water Resources Research, 32(3), 653–661. https://doi.org/10.1029/95WR03638
Babaeian, E., Paheding, S., Siddique, N., Devabhaktuni, V. K., & Tuller, M. (2021). Estimation of root zone soil moisture from ground and remotely sensed soil information with multisensor data fusion and automated machine learning. Remote Sensing of Environment, 260. https://doi.org/10.1016/j.rse.2021.112434
Barrett, B. W., Dwyer, E., & Whelan, P. (2009). Soil moisture retrieval from active spaceborne microwave observations: An evaluation of current techniques. Remote Sensing, 1(3), 210–242. https://doi.org/10.3390/rs1030210
Bulcock, H. H., & Jewitt, G. P. W. (2010). Spatial mapping of leaf area index using hyperspectral remote sensing for hydrological applications with a particular focus on canopy interception. In Hydrol. Earth Syst. Sci (Vol. 14). www.hydrol-earth-syst-sci.net/14/383/2010/
Carranza, C., Nolet, C., Pezij, M., & van der Ploeg, M. (2021). Root zone soil moisture estimation with Random Forest. Journal of Hydrology, 593, 125840. https://doi.org/10.1016/J.JHYDROL.2020.125840
Cervantes, J., Garcia-Lamont, F., Rodríguez-Mazahua, L., & Lopez, A. (2020). A comprehensive survey on support vector machine classification: Applications, challenges and trends. Neurocomputing, 408, 189–215. https://doi.org/10.1016/j.neucom.2019.10.118
Chamasemani, F. F., & Singh, Y. P. (2011). Multi-class Support Vector Machine (SVM) classifiers - An application in hypothyroid detection and classification. Proceedings - 2011 6th International Conference on Bio-Inspired Computing: Theories and Applications, BIC-TA 2011, 351–356. https://doi.org/10.1109/BIC-TA.2011.51
Chaudhary, S. K., Srivastava, P. K., Kumar Gupta, D., Kumar, P., Prasad, R., Pandey, D. K., Das, A. K., & Gupta, M. (2021). Machine learning algorithms for soil moisture estimation using Sentinel-1: Model development and implementation.
Das, K., & Paul, P. K. (2015). Present status of soil moisture estimation by microwave remote sensing. Cogent Geoscience, 1(1), 1084669. https://doi.org/10.1080/23312041.2015.1084669
Dobriyal, P., Qureshi, A., Badola, R., & Hussain, S. A. (2012). A review of the methods available for estimating soil moisture and its implications for water resource management. In Journal of Hydrology (Vols. 458–459, pp. 110–117). https://doi.org/10.1016/j.jhydrol.2012.06.021
Edokossi, K., Calabia, A., Jin, S., & Molina, I. (2020). GNSS-reflectometry and remote sensing of soil moisture: A review of measurement techniques, methods, and applications. In Remote Sensing (Vol. 12, Issue 4). MDPI AG. https://doi.org/10.3390/rs12040614
Ezzahar, J., Ouaadi, N., Zribi, M., Elfarkh, J., Aouade, G., Khabba, S., Er-Raki, S., Chehbouni, A., & Jarlan, L. (2020). Evaluation of backscattering models and support vector machine for the retrieval of bare soil moisture from sentinel-1 data. Remote Sensing, 12(1). https://doi.org/10.3390/RS12010072
Filipponi, F. (2019). Sentinel-1 GRD Preprocessing Workflow. 11. https://doi.org/10.3390/ecrs-3-06201
Grubinger, T., Zeileis, A., & Pfeiffer, K. P. (2014). evtree: Evolutionary Learning of Globally Optimal Classification and Regression Trees in R. Journal of Statistical Software, 61(1), 1–29. https://doi.org/10.18637/JSS.V061.I01
Guo, R., Fu, D., & Sollazzo, G. (2022). An ensemble learning model for asphalt pavement performance prediction based on gradient boosting decision tree. International Journal of Pavement Engineering, 23(10), 3633–3646. https://doi.org/10.1080/10298436.2021.1910825
Gupta, D. K., Prasad, R., Kumar, P., & Vishwakarma, A. K. (2017). Soil moisture retrieval using ground based bistatic scatterometer data at X-band. Advances in Space Research, 59(4), 996–1007. https://doi.org/10.1016/j.asr.2016.11.032
Gururaj, P., Umesh, P., & Shetty, A. (2021). Modeling of surface soil moisture using C-band SAR data over bare fields in the tropical semi-arid region of India. Applied Geomatics, 13(4), 555–564. https://doi.org/10.1007/s12518-021-00370-7
Hallikainen, M. T., Ulaby, F. T., Dobson, M. C., El-Rayes, M. A., & Wu, L.-K. (1985). Microwave Dielectric Behavior of Wet Soil-Part 1: Empirical Models and Experimental Observations. In IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (Issue 1).
Jiang, Z., Huete, A. R., Chen, J., Chen, Y., Li, J., Yan, G., & Zhang, X. (2006). Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction. Remote Sensing of Environment, 101(3), 366–378. https://doi.org/10.1016/j.rse.2006.01.003
Kseneman, M., Gleich, D., & Čučej, Ž. (2011). Soil moisture estimation using high-resolution spotlight terraSAR-X data. IEEE Geoscience and Remote Sensing Letters, 8(4), 686–690. https://doi.org/10.1109/LGRS.2010.2099641
Kumar, P., Gupta, D. K., Mishra, V. N., & Prasad, R. (2015). Comparison of support vector machine, artificial neural network, and spectral angle mapper algorithms for crop classification using LISS IV data. International Journal of Remote Sensing, 36(6), 1604–1617. https://doi.org/10.1080/2150704X.2015.1019015
Kumar, P., Prasad, R., Gupta, D. K., Mishra, V. N., Vishwakarma, A. K., Yadav, V. P., Bala, R., Choudhary, A., & Avtar, R. (2018). Estimation of winter wheat crop growth parameters using time series Sentinel-1A SAR data. Geocarto International, 33(9), 942–956. https://doi.org/10.1080/10106049.2017.1316781
Leiva, R. G., Anta, A. F., Mancuso, V., & Casari, P. (2019). A novel hyperparameter-free approach to decision tree construction that avoids overfitting by design. IEEE Access, 7, 99978–99987. https://doi.org/10.1109/ACCESS.2019.2930235
Li, Z. L., Leng, P., Zhou, C., Chen, K. S., Zhou, F. C., & Shang, G. F. (2021). Soil moisture retrieval from remote sensing measurements: Current knowledge and directions for the future. In Earth-Science Reviews (Vol. 218). Elsevier B.V. https://doi.org/10.1016/j.earscirev.2021.103673
Lillicrap, T. P., Cownden, D., Tweed, D. B., & Akerman, C. J. (2016). Random synaptic feedback weights support error backpropagation for deep learning. Nature Communications, 7. https://doi.org/10.1038/ncomms13276
Liu, J., Rahmani, F., Lawson, K., & Shen, C. (2022). A Multiscale Deep Learning Model for Soil Moisture Integrating Satellite and In Situ Data. Geophysical Research Letters, 49(7), e2021GL096847. https://doi.org/10.1029/2021GL096847
Liu, Y. Y., Dorigo, W. A., Parinussa, R. M., De Jeu, R. A. M., Wagner, W., McCabe, M. F., Evans, J. P., & Van Dijk, A. I. J. M. (2012). Trend-preserving blending of passive and active microwave soil moisture retrievals. Remote Sensing of Environment, 123, 280–297. https://doi.org/10.1016/j.rse.2012.03.014
Loew, A., & Mauser, W. (2007). Generation of geometrically and radiometrically terrain corrected SAR image products. Remote Sensing of Environment, 106(3), 337–349. https://doi.org/10.1016/j.rse.2006.09.002
Lunt, I. A., Hubbard, S. S., & Rubin, Y. (2005). Soil moisture content estimation using ground-penetrating radar reflection data. Journal of Hydrology, 307(1–4), 254–269. https://doi.org/10.1016/j.jhydrol.2004.10.014
Ma, C., Li, X., & McCabe, M. F. (2020). Retrieval of high-resolution soil moisture through combination of Sentinel-1 and Sentinel-2 data. Remote Sensing, 12(14). https://doi.org/10.3390/rs12142303
Mao, K., Wang, H., Shi, J., Heggy, E., Wu, S., Bateni, S. M., & Du, G. (2023). A General Paradigm for Retrieving Soil Moisture and Surface Temperature from Passive Microwave Remote Sensing Data Based on Artificial Intelligence. Remote Sensing, 15(7). https://doi.org/10.3390/rs15071793
Mehta, S., & Patnaik, K. S. (2021). Improved prediction of software defects using ensemble machine learning techniques. Neural Computing and Applications, 33(16), 10551–10562. https://doi.org/10.1007/S00521-021-05811-3/METRICS
Mitran, T., Meena, R. S., & Chakraborty, A. (2020). Geospatial Technologies for Crops and Soils. In Geospatial Technologies for Crops and Soils. Springer Singapore. https://doi.org/10.1007/978-981-15-6864-0
Neaupane, K. M., & Achet, S. H. (2004a). Use of backpropagation neural network for landslide monitoring: A case study in the higher Himalaya. Engineering Geology, 74(3–4), 213–226. https://doi.org/10.1016/j.enggeo.2004.03.010
Neaupane, K. M., & Achet, S. H. (2004b). Use of backpropagation neural network for landslide monitoring: A case study in the higher Himalaya. Engineering Geology, 74(3–4), 213–226. https://doi.org/10.1016/j.enggeo.2004.03.010
Owe, M., De Jeu, R., & Walker, J. (2001). A Methodology for Surface Soil Moisture and Vegetation Optical Depth Retrieval Using the Microwave Polarization Difference Index. In IEEE Transactions On Geoscience And Remote Sensing (Vol. 39, Issue 8).
Peng, J., Loew, A., Merlin, O., & Verhoest, N. E. C. (2017). A review of spatial downscaling of satellite remotely sensed soil moisture. Reviews of Geophysics, 55(2), 341–366. https://doi.org/10.1002/2016RG000543
Petropoulos, G. P., Ireland, G., & Barrett, B. (2015). Surface soil moisture retrievals from remote sensing: Current status, products & future trends. In Physics and Chemistry of the Earth (Vols. 83–84, pp. 36–56). Elsevier Ltd. https://doi.org/10.1016/j.pce.2015.02.009
Pierdicca, N., Pulvirenti, L., Bignami, C., & Ticconi, F. (2013). Monitoring soil moisture in an agricultural test site using SAR data: Design and test of a pre-operational procedure. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6(3), 1199–1210. https://doi.org/10.1109/JSTARS.2012.2237162
Rasheed, M. W., Tang, J., Sarwar, A., Shah, S., Saddique, N., Khan, M. U., Imran Khan, M., Nawaz, S., Shamshiri, R. R., Aziz, M., & Sultan, M. (2022). Soil Moisture Measuring Techniques and Factors Affecting the Moisture Dynamics: A Comprehensive Review. In Sustainability (Switzerland) (Vol. 14, Issue 18). MDPI. https://doi.org/10.3390/su141811538
Rawat, K. S., Sehgal, V. K., Pradhan, S., & Ray, S. S. (2018). Semi-empirical model for retrieval of soil moisture using RISAT-1 C-Band SAR data over a sub-tropical semi-arid area of Rewari district, Haryana (India). Journal of Earth System Science, 127(2). https://doi.org/10.1007/s12040-018-0919-2
Robinson, D. A., Campbell, C. S., Hopmans, J. W., Hornbuckle, B. K., Jones, S. B., Knight, R., Ogden, F., Selker, J., & Wendroth, O. (2008). Soil Moisture Measurement for Ecological and Hydrological Watershed‐Scale Observatories: A Review. Vadose Zone Journal, 7(1), 358–389. https://doi.org/10.2136/vzj2007.0143
Saleh, A. (n.d.). DIVISION S-6-NOTES MEASURING AND PREDICTING RIDGE-ORIENTATION EFFECT ON SOIL SURFACE ROUGHNESS.
Shafizadeh-Moghadam, H., Tayyebi, A., Ahmadlou, M., Delavar, M. R., & Hasanlou, M. (2017). Integration of genetic algorithm and multiple kernel support vector regression for modeling urban growth. Computers, Environment and Urban Systems, 65, 28–40. https://doi.org/10.1016/j.compenvurbsys.2017.04.011
Shen, C., Laloy, E., Elshorbagy, A., Albert, A., Bales, J., Chang, F. J., Ganguly, S., Hsu, K. L., Kifer, D., Fang, Z., Fang, K., Li, D., Li, X., & Tsai, W. P. (2018). HESS Opinions: Incubating deep-learning-powered hydrologic science advances as a community. Hydrology and Earth System Sciences, 22(11), 5639–5656. https://doi.org/10.5194/hess-22-5639-2018
Singh, A., & Gaurav, K. (2023). Deep learning and data fusion to estimate surface soil moisture from multi-sensor satellite images. Scientific Reports, 13(1). https://doi.org/10.1038/s41598-023-28939-9
Singh, A., Gaurav, K., Sonkar, G. K., & Lee, C. C. (2023). Strategies to Measure Soil Moisture Using Traditional Methods, Automated Sensors, Remote Sensing, and Machine Learning Techniques: Review, Bibliometric Analysis, Applications, Research Findings, and Future Directions. IEEE Access, 11, 13605–13635. https://doi.org/10.1109/ACCESS.2023.3243635
Srivastava, P. K., Han, D., Ramirez, M. R., & Islam, T. (2013). Machine Learning Techniques for Downscaling SMOS Satellite Soil Moisture Using MODIS Land Surface Temperature for Hydrological Application. Water Resources Management, 27(8), 3127–3144. https://doi.org/10.1007/s11269-013-0337-9
Sunar Erbek, F., Özkan, C., & Taberner, M. (2004). Comparison of maximum likelihood classification method with supervised artificial neural network algorithms for land use activities. International Journal of Remote Sensing, 25(9), 1733–1748. https://doi.org/10.1080/0143116031000150077
Trafalis, T. B., & Gilbert, R. C. (2007). Robust support vector machines for classification and computational issues. 22(1), 187–198. https://doi.org/10.1080/10556780600883791
Wang, L., & Qu, J. J. (2009). Satellite remote sensing applications for surface soil moisture monitoring: A review. In Frontiers of Earth Science in China (Vol. 3, Issue 2, pp. 237–247). Higher Education Press. https://doi.org/10.1007/s11707-009-0023-7
Zhang, H., Chang, J., Zhang, L., Wang, Y., Li, Y., & Wang, X. (2018). NDVI dynamic changes and their relationship with meteorological factors and soil moisture. Environmental Earth Sciences, 77(16). https://doi.org/10.1007/s12665-018-7759-x.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
28 Sep, 2024
Reviews received at journal
28 Sep, 2024
Reviews received at journal
17 Sep, 2024
Reviews received at journal
15 Sep, 2024
Reviewers agreed at journal
07 Sep, 2024
Reviewers agreed at journal
07 Sep, 2024
Reviewers agreed at journal
05 Sep, 2024
Reviewers invited by journal
05 Sep, 2024
Editor assigned by journal
26 Aug, 2024
Submission checks completed at journal
26 Aug, 2024
First submitted to journal
22 Aug, 2024

You are reading this latest preprint version

Dynamic Monitoring of Surface Soil Moisture Fluctuations Using Synthetic Aperture Radar and Data- Driven Algorithms

Status:

Version 1

Abstract

Figures

1. INTRODUCTION

2. STUDY AREA

3. DATA PRODUCTS

3.1 Field data

3.2 Rainfall

3.3 Sentinel-1A

3.4 Sentinel-2A

3.5 SRTM data

3.6 Google Earth data

4. METHODOLOGY

4.1 Field data

4.2 Field Data Analysis

4.3 Image Processing

4.4 NDVI Calculation

4.5 Model development using Machine Learning algorithms

5. Results and Discussion

5.1 Effect of vegetation on \(\:{\varvec{\sigma\:}}^{0}\)

5.2 Effect of Soil Moisture on\(\:\:{\varvec{\sigma\:}}^{0}\)

5.3 Effect of soil surface roughness on \(\:{\varvec{\sigma\:}}^{0}\)

5.4 Effect of dielectric constant on \(\:{\varvec{\sigma\:}}^{0}\)

5.5 Retrieval of bare soil moisture using RFR, SVR and ANNR mode

6. Conclusion

Declarations

Conflict of interest:

Funding:

Author Contribution

Data availability:

References

Additional Declarations

Status:

Version 1