Surface Soil Moisture Estimation Using a Neural Network Model in Bare Land and Vegetated Areas

Luo, Dayou; Wen, Xingping; He, Ping

doi:https://doi.org/10.1155/2023/5887177

Journal of Spectroscopy

On this page

Abstract Introduction Study Area Results Discussion Conclusion Data Availability Conflicts of Interest Acknowledgments References Copyright Related Articles

Research Article | Open Access

Volume 2023 | Article ID 5887177 | https://doi.org/10.1155/2023/5887177

Surface Soil Moisture Estimation Using a Neural Network Model in Bare Land and Vegetated Areas

Dayou Luo,^1,2Xingping Wen ,^1,2and Ping He^1,3

Academic Editor: Wee Chew

Received29 May 2022

Revised18 Nov 2022

Accepted14 Dec 2022

Published04 Jan 2023

Abstract

Most of the approaches to retrieve surface soil moisture (SSM) by optical and thermal infrared (TIR) spectroscopies are purposed to calculate various characteristic bands/indices and then to establish the regression relationship between them in combination with the measurement data. However, due to the combined impact of many factors, the regression relationship often shows nonlinearity. Moreover, the relationship between the single temporal image and the measured data are not transplantable in time and space, which makes it difficult to construct a more general model for the remote sensing (RS) estimation of SSM. In order to solve this problem, the back propagation (BP) neural network (NN) with an excellent nonlinear mapping ability is introduced to determine the relationship between the characteristic band/index and the measurement data. In the BPNN model, the optical and TIR RS data in different periods were taken as the input parameters, and the in situ soil moisture data were treated as the output parameter. There are 12 schemes designed to retrieve SSM. The key findings of study were as follows: (1) the BPNN model could retrieve SSM with a high accuracy that indicates the correlation coefficient between the estimated and measured soil moisture as 0.9001 and (2) the SSM retrieval model based on the BPNN can be applied to estimate the SSM with different spatial resolution values.

1. Introduction

As an important parameter in meteorology, hydrology, and agricultural science, soil water content (SWC) has a significant impact on land surface evaporation, vegetation photosynthesis, and water displacement [1, 2]. It acts as the hub connecting the groundwater-surface water-atmospheric water cycle system [3, 4]. Therefore, the accurate monitoring of soil moisture (SM) is essential for the prediction of climate change [5, 6]. Though direct measurement is the most accurate method to monitor SM, in situ measurement data are ineffective in showing the spatial distribution of SM, nor can it be used to monitor the spatial and temporal changes in SM on a large scale [7, 8].

The development of a remote sensing (RS) data algorithm provides an effective solution to the large-scale monitoring of surface soil moisture (SSM) [9, 10]. Various RS methods have been developed in different regions of the electromagnetic spectrum. These methods may be classified into three major groups: microwave (MW), thermal infrared (TIR), and optical [11, 12]. Microwave is advantageous due to its stronger penetration ability, longer wavelength, and the robustness to adverse weather conditions. Therefore, MW technology shows a greater potential in monitoring global SM dynamics [13]. However, due to the relatively rough resolution of SM products provided by Soil Moisture Active Passive (SMAP) mission and Soil Moisture and Ocean Salinity (SMOS) mission, they are not suitable for small-scale regional applications. Apart from that, a challenge presented by using MW to monitor SSM is the need to input various physical parameters, some of which are still not well defined at present [14]. For example, surface roughness, which is easily changeable, must be measured in the field, which is time-consuming.

Compared with MW SM products, optical/TIR RS data usually have a higher spatial resolution, which can well meet the needs of SM data in various fields [15]. Optical methods apply the visible, near infrared (NIR), and shortwave infrared (SWIR) data for SM retrieval. Many scholars have used different vegetation indices to monitor SM and dryness [16]. However, there is a certain lag in vegetation growth for the degree of surface drought. Some other studies used spectral feature space to construct SSM retrieval models. For example, the perpendicular drought index (PDI) [17] is derived from the red and near infrared (NIR) band reflectance triangular space. This method needs to determine the soil line artificially to realize the model. However, the distribution of soil lines highly depends on some parameters, such as soil minerals, iron oxide content, particle size distribution, and organic matter [18]. Although TIR methods are developed based on physical principles, the methods are usually empirical and depend on local meteorological conditions, such as humidity, temperature, and wind speed, and thus, the measurement results vary temporally and with land cover types [19, 20].

The combination of optical and TIR RS data mostly results in better accuracy for SM retrieval. To some extent, land surface temperature (LST) can be taken as an indicator of the change in SM and vegetation. Therefore, the retrieval model based on LST and vegetation index (VI) can be used to estimate SM [21]. The temperature vegetation dryness index (TVDI) is parameterized based on the LST-VI space [22]. Over the past two decades, this simple model and those improved models in relation to it have already been applied to SSM estimation [23–26]. Despite this, the spatial distribution of LST-VI remains possible not to conform to triangle or trapezoid, thus affecting the accuracy of SM estimation. The spatial shape is treated as a triangle or trapezoid in the context of human intervention, which can lead to uncertainty in the calculation results.

A major method of estimating SWC based on optical and thermal infrared data is to retrieve the SM index or drought index using satellite data [27–30], so as to establish the regression relationship between the index and the actual SWC using the measurement data [31]. However, the variation of SWC can be affected by various factors (e.g., soil texture, surface roughness, topography, temperature, and vegetation coverage), which leads to the nonlinear regression relationship. Moreover, the relationship between the single temporal image and the measurement data is not transferable in time and space, which makes it challenging to establish a more general model of RS estimation for SM.

Compared with those traditional methods, the neural network (NN) model demonstrates various advantages, including the flexibility in combining information from different sources and mapping the input-output relationship of any data without needing an empirical formula [32]. The statistical relationship between satellite observation and SSM can be simulated on the basis of satellite data and measured SM estimation without explicitly establishing any physical connection. The excellent nonlinear mapping ability of the NN model makes it capable to minimize uncertainty and improve the accuracy of SM retrieval from RS data [33–35].

Some scholars have constructed an NN model based on microwave data for estimating soil moisture [36, 37]. There is also a synergistic use of microwave and optical data to retrieve soil moisture using NN models [38, 39]. In these studies, NN models estimated soil moisture with a high degree of accuracy. Some studies have used the normalized difference vegetation index calculated by optical data and land surface temperature calculated from thermal infrared data as parameters into the neural network model to estimate soil moisture [40]. Optical and thermal infrared data have an exceptional potential for estimating soil moisture. However, there are few studies on SM estimation using optical and TIR data with the NN model.

The primary purpose of this paper is to build a highly universal SM retrieval model, which can overcome the influence of soil texture, vegetation cover, terrain, and other factors. As one variety of the ANN, the back propagation (BP) neural network (NN) is applied to construct the SM prediction model. The characteristic band/index with high resolution and multitemporal data is taken as the input data, while the measured SM is treated as the output data. Finally, the multiple spatial and temporal Landsat-8 and MODIS data were used to estimate SSM. The model constructed in this paper is universal and can be applied to estimate SM in both bare land and vegetated areas.

2. Study Area and Data Source

2.1. Study Area

As shown in Figure 1, the study area is located in the middle of the Tibetan Plateau. The map of China is downloaded from website https://bzdt.ch.mnr.gov.cn/index.html. The climate in the Naqu area is relatively warm from May to September each year, and this period refers to the season of vigorous vegetation growth. During this period, the land cover is dominated by plateau meadow. The other period includes the snow and soil freezing period, during which most areas are covered by snow and frozen soil. As impacted by low biomass, air quality, and humidity, the MW signal attenuation is weak, so this area acts as an ideal place to verify satellite and model simulation of SM products [41].

2.2. In Situ Measurements

The measured SM data used in this study originated from the SM observation network in the Naqu area. There were 57 stations available in the network, with SM and temperature data at four observation depths (i.e., 5, 10, 20, and 40 cm), and the sampling interval was 30 min. The mentioned measurements provided considerable valuable experimental data for SM retrieval in the Tibetan Plateau [42]. In the present study, the measured SM data from 2014 to 2016 were collected to assess the performance of the SSM retrieval model. The dataset is provided by the National Tibetan Plateau Data Center.

2.3. Remote Sensing Data

Landsat 8 carries Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). The OLI provides nine optical bands, and the TIRS provides two thermal bands. Twelve Landsat-8 images from 17 May, 24 May, 18 June, and 27 July, 2014; 20 May, 23 July, 30 July, 31 August, 9 September, and 16 September, 2015; 25 July and 17 August, 2016, were acquired.

3. Research Method

By using the BPNN, the coastal, blue, green, red, NIR, shortwave infrared (SWIR) band, and LST were taken as the input data, the measured soil volumetric moisture was taken as the output data, and the SM retrieval model is established. The LST data were calculated using the TIRS1 and TIRS2 bands of Landsat-8. In addition, the observed SM data are used to test the results of model estimation and evaluate retrieval accuracy.

3.1. BPNN Model

The BPNN refers to a type of the multilayer feed-forward network with a hidden layer. In the practical application of ANNs, the BP network has been extensively used in data compression, pattern recognition, as well as function approximation. The BP network consists of an input layer, hidden layer, and output layer. The BPNN comprises an input layer, a hidden layer, and an output layer. The structure with a single hidden layer of the BPNN is shown in Figure 2. All neurons were configured according to the model shown in Figure 3.

The training operation of the BPNN consists of signal forward propagation and error back propagation. Forward propagation means that the input signal was transmitted from the input layer via the hidden layer to the output layer. If the output layer yields the desired output (the actual output is consistent with the expected output), the learning algorithm ends; otherwise, it turns to back propagation. Back propagation aims to reduce the error by calculating the error according to the original connection path direction and adjusting the weight and threshold of each layer node based on the gradient descent method. After repeated learning and training, the network parameters (weight and threshold) are determined for the minimum error and the training is terminated. At this point, the trained NN is capable to process the input information of similar samples and to output the nonlinear converted information with the minimum error.

3.2. Input Parameter Scheme

The approach to monitoring SM with optical data is mainly based on the spatial shape of pixel distribution in Landsat-8 bands to estimate SM. The SWIR 1 and SWIR 2 bands of Landsat-8 data (1650 nm corresponding to band 6, 2210 nm corresponding to band 7) are sensitive to SWC, especially at 2210 nm [43]. In this study, the SWIR 2 is taken as the input parameters of the SSM estimation model.

There are twelve schemes that were constructed to analyze the effect of various parameters on the accuracy of SSM estimation (Table 1). Scheme 1 sets all optical bands as input parameters. Based on the formula expressing SM in Tasselled Cap transformation, the band used in the formula was taken as the input parameter of scheme 2. According to the previous models applied to calculate SM through optical data, it can be found out that there were only a small number of bands in red, NIR, and SWIR bands as used in these models, so that schemes 3–6 were designed. On the basis of schemes 1–6, the LST data were introduced, respectively, with schemes 7–12 set.

4. Results

4.1. SSM Estimation Results and Validation

In the experiment, a total of 252 sample data were extracted from the study area. In addition, the BPNN was set below: the training sample was 189 and the test sample was 63. In order to ensure the stability of the BPNN model, we trained the BPNN model 100 times to obtain the correlation coefficient of estimating and measuring soil moisture. Finally, the average value of 100 training results is selected as the estimation accuracy of the model. The in situ measured soil volumetric moisture was used as a reference to calculate the coefficient ®, root mean square error (RMSE), unbiased root mean square error (ubRMSE), and bias (Bias) of the estimated SSM. Figure 4 shows the scatter distribution of soil volumetric moisture estimated and measured by 12 schemes. Table 2 shows the accuracy of the estimated SSM in each scheme.

With only the optical bands as the input parameters of the SM estimation model (schemes 1–6), the estimated SM shows correlation with the measured value to a certain extent, and the correlation coefficient reaches up to 0.8682, which renders the accuracy of retrieval poor. In the scheme of estimating SM with optical data, all bands were treated as input to obtain the most accurate SM (scheme 1). In scheme 2, the r between the estimated and measured SM is 0.8163. After removing the coastal band, the estimation accuracy of SM does not change significantly. As for the input, only the red, NIR, and SWIR bands are retained (scheme 3). Whereas, the estimation accuracy of SM remains comparable to that of scheme 1, indicating that the red, NIR, and SWIR bands play a major role in SM estimation.

In schemes 7–12, the input parameters were set as optical bands and LST, and the estimated SSM was closer to the observation. The SM estimated by scheme 7 had the highest accuracy. It showed a RMSE of 0.0017 and r of 0.9001. The normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) were calculated by red, NIR, and SWIR bands. According to the LST-VI SM estimation model, schemes 10–11 of red, NIR, SWIR, and LST are designed as input parameters. Scheme 11 shows a higher estimation accuracy of SSM than that of scheme 10, which suggests that NDWI is more suitable to represent VI for SSM estimation than NDVI. With red, SWIR band, and LST as input parameters (scheme 12), the SM estimation accuracy was higher than in schemes 10 and 11.

When using the BPNN model to retrieve soil moisture, the soil moisture with the highest accuracy can be estimated by selecting scheme 7. In case of lack of TIR data, scheme 1 is preferred to retrieve soil moisture.

4.2. Land Surface Feature Influence on SSM Estimation Results

Zhang et al. [44] used the random forest model, which had land surface feature parameters and in situ soil moisture data to predict SSM. It would be good to test the same experiment on a different environment to check the consistency of the results. To analyse the influence of land surface feature factors on SSM estimation results, the elevation, slope, aspect, and vegetation coverage (normalized difference vegetation index, NDVI) factors were added into BPNN model. The soil texture of all observation stations in the study area is the same type, so this factor was not considered. On the basis of scheme 7, each influence factor was added, respectively. The estimated soil moisture were grouped according to the value of the influence factors. Also, the r between each group of data and the corresponding measured soil moisture were discussed (Table 3).

With the increase of elevation and slope, the r between estimated and measured SSM gradually decreases. The r of the north aspect was obviously lower than that of other aspects. The terrain of Naqu study area fluctuates violently, and it was easy to have shadow in areas with large elevation, steep slope, and north aspect. In the shadow areas, the surface reflectance is very low, and the satellite data in this area will greatly reduce the accuracy of estimated SSM. It is worth noting that the mountain shadow in the southern hemisphere appears on the south aspect. In areas with different NDVI values, the r values of estimated and measured SSM were similar. Vegetation coverage will not significantly affect SSM estimated by the BPNN model.

BPNN is not affected by the underlying surface condition and can be applied to areas with different vegetation coverage. However, when using optical satellite data as input parameters, the BPNN model is more suitable for the area with flat terrain and gentle slope. Also, BPNN cannot’ construct the empirical formula to express the relationship between soil moisture and various parameters.

4.3. Application of the SSM Estimation Model

Since the grid size of Landsat data was 30 m × 30 m, the small-scale observation network of the Naqu SM observation network was taken as the retrieval area of SSM. Figure 5 shows the surface soil moisture map retrieved from scheme 7 on May 20, 2015. The soil volumetric moisture of the 9 stations estimated in the area ranged from 0.122 to 0.357 cm³/cm³ in accordance with the in situ data (0.145–0.395 cm³/cm³). The scatter plot between measured and estimated soil volumetric moisture is shown in Figure 5©. It showed a RMSE of 0.0429 and r of 0.8392. The result indicates that calibration of the BPNN model with in situ data generally leads to reasonable SM estimates.

(a)

(b)

(c)

5. Discussion

The PDI and TVDI models were used to retrieve soil moisture, respectively. The PDI and TVDI models were used to estimate SSM in the same area of Figure 5. The red and NIR bands shall be applied to calculate PDI (corresponding to scheme 4), while the red, NIR, and LST shall be applied to calculate TVDI (corresponding to scheme 10). The scatter plot of SM estimated and measured by PDI and TVDI is shown in Figure 6. The r between the estimated and measured SM are 0.4786 and 0.5047, respectively. Comparing the SSM calculated by PDI and TVDI, the accuracy of SM calculated by the BPNN has significantly improved.

(a)

(b)

We also brought MODIS data into the BPNN model to estimate soil moisture. There is a slight difference between the band range of MODIS and Landsat-8 data (Table 4), and there is no coastal band in MODIS data. The LST data were obtained from MOD11A1 products. There are ten schemes (schemes 2–6 and schemes 7–12 in Table 1) to estimate SM using the BPNN.

We collected 742 MODIS data, 594 data were used as training samples, and the remaining 148 data were used as test samples. The retrieved SSM had the highest accuracy when the input parameters were green, blue, red, NIR, SWIR band, and LST data. The scatter plot between measured and estimated soil volumetric moisture is shown in Figure 7. It showed a RMSE of 0.0573 and r of 0.8164. The SSM estimation accuracy is similar to Landsat-8 data (scheme 8).

It is worth noting that the more the input sample data, the higher the accuracy of SSM retrieved by the BPNN. However, this method is not applicable in cloud and shadow-covered areas. Since the input parameters in the model are all derived from RS raster data, the images with different spatial resolutions can be used for the SSM estimation model based on the BPNN. Sentinel-2 is regarded as another ideal RS data due to its high temporal and spatial resolution.

The BPNN model is not limited by RS data. The RS images such as Landsat-8, MODIS, and Sentinel-2 can be used as input parameters to retrieve soil moisture. Since the BPNN model directly uses the surface spectral reflectance data to retrieve soil moisture, the model is not limited by regional conditions. The BPNN model can simultaneously retrieve soil moisture of bare soil and vegetated areas. However, when optical and TIR data are input into the NN model, it is necessary to exclude raster data that are obscured by clouds or affected by the atmosphere.

6. Conclusion

Combined with optical-thermal infrared RS data and measured soil volumetric moisture data, a retrieval model of SSM is established on the basis of the BPNN. It can be found that the BPNN is suitable for establishing the relationship between the characteristic band/index and the measurement data, thus improving the accuracy of the SM estimation model. However, the accuracy of SSM estimated only by optical data is poor. By introducing TIR data, the retrieval accuracy of SSM can be improved significantly. It is worth noting that when the optical satellite data are used as the input parameters, the BPNN model is more suitable for the area with flat terrain and gentle slope.

When the quality of RS image varies, it is possible that the spatial distribution of the LST-VI model is unable to form a complete triangle or trapezoid, thus resulting in a significant difference between the estimated SM and the measured value. The model proposed in this study addresses the abovementioned problems while estimating SSM in case of a small sample size. The BPNN model provides a powerful tool for estimating SSM in complex/heterogeneous media (e.g., rough vegetated surfaces). In this paper, the soil moisture retrieved by Landsat-8 and MODIS data has high accuracy. The SSM retrieval model based on the BPNN can be applied to estimate the SSM with different spatial resolution.

Data Availability

The data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Acknowledgments

The authors would like to thank for the soil moisture observation data provided by National Tibetan Plateau Data Center and the Landsat-8 data provided by the USGS.

References

E. A. B. Eltahir, “A soil moisture-rainfall feedback mechanism, 1. Theory and observations,” Water Resources Research, vol. 34, no. 4, pp. 765–776, 1998.
View at: Publisher Site | Google Scholar
S. I. Seneviratne, T. Corti, E. L. Davin et al., “Investigating soil moisture-climate interactions in a changing climate: a review,” Earth-Science Reviews, vol. 99, no. 3-4, pp. 125–161, 2010.
View at: Publisher Site | Google Scholar
P. Rahimzadeh-Bajgiran, A. A. Berg, C. Champagne, and K. Omasa, “Estimation of soil moisture using optical/thermal infrared remote sensing in the Canadian Prairies,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 83, pp. 94–103, 2013.
View at: Publisher Site | Google Scholar
I. M. Ferguson, J. L. Jefferson, R. M. Maxwell, and S. J. Kollet, “Effects of root water uptake formulation on simulated water and energy budgets at local and basin scales,” Environmental Earth Sciences, vol. 75, no. 4, p. 316, 2016.
View at: Publisher Site | Google Scholar
P. Falloon, C. D. Jones, M. Ades, and K. Paul, “Direct soil moisture controls of future global soil carbon changes: an important source of uncertainty,” Global Biogeochemical Cycles, vol. 25, no. 3, p. 3010, 2011.
View at: Publisher Site | Google Scholar
W. Dorigo and R. De Jeu, “Satellite soil moisture for advancing our understanding of earth system processes and climate change,” International Journal of Applied Earth Observation and Geoinformation, vol. 48, pp. 1–4, 2016.
View at: Publisher Site | Google Scholar
M. Hosseini and M. R. Saradjian, “Soil moisture estimation in a vegetated area using combination of AIRSAR and landsat5-TM images,” Journal of the Indian Society of Remote Sensing, vol. 42, no. 4, pp. 719–726, 2014.
View at: Publisher Site | Google Scholar
D. Baldwin, S. Manfreda, H. Lin, and E. A. Smithwick, “Estimating root zone soil moisture across the Eastern United States with passive microwave satellite data and a simple hydrologic model,” Remote Sensing, vol. 11, no. 17, p. 2013, 2019.
View at: Publisher Site | Google Scholar
S. Elbialy, A. Mahmoud, B. Pradhan, and M. Buchroithner, “Application of spaceborne synthetic aperture radar data for extraction of soil moisture and its use in hydrological modelling at Gottleuba Catchment, Saxony, Germany,” Journal of Flood Risk Management, vol. 7, no. 2, pp. 159–175, 2014.
View at: Publisher Site | Google Scholar
S. Koley and C. Jeganathan, “Estimation and evaluation of high spatial resolution surface soil moisture using multi-sensor multi-resolution approach,” Geoderma, vol. 378, no. 15, Article ID 114618, 2020.
View at: Publisher Site | Google Scholar
W. W. Verstraeten, F. Veroustraete, and J. Feyen, “Assessment of evapotranspiration and soil moisture content across different scales of observation,” Sensors, vol. 8, no. 1, pp. 70–117, 2008.
View at: Publisher Site | Google Scholar
A. A. Hassaballa, O. F. Althuwaynee, and B. Pradhan, “Extraction of soil moisture from RADARSAT-1 and its role in the formation of the 6 december 2008 landslide at bukit antarabangsa, kuala lumpur,” Arabian Journal of Geosciences, vol. 7, no. 7, pp. 2831–2840, 2014.
View at: Publisher Site | Google Scholar
A. Tabatabaeenejad, M. Burgin, X. Duan, and M. Moghaddam, “P-band radar retrieval of subsurface soil moisture profifile as a second-order polynomial: fifirst AirMOSS results,” IEEE T Geosci Remote, vol. 53, no. 2, pp. 645–658, 2015.
View at: Publisher Site | Google Scholar
M. Sadeghi, S. B. Jones, and W. D. Philpot, “A linear physically-based model for remote sensing of soil moisture using short wave infrared bands,” Remote Sensing of Environment, vol. 164, pp. 66–76, 2015.
View at: Publisher Site | Google Scholar
H. H. Nguyen, S. Cho, and M. Choi, “Spatial soil moisture estimation in agro-pastoral transitional zone based on synergistic use of SAR and optical-thermal satellite images,” Agricultural and Forest Meteorology, vol. 312, no. 7, Article ID 108719, 2022.
View at: Publisher Site | Google Scholar
A. J. Peters, E. A. Walter-Shea, L. Ji, A. Vina, M. Hayes, and M. D. Svoboda, “Drought monitoring with NDVI-based standardized vegetation index,” Photogrammetric Engineering & Remote Sensing, vol. 68, no. 1, pp. 71–75, 2002.
View at: Google Scholar
A. Ghulam, Q. Qin, and Z. Zhan, “Designing of the perpendicular drought index,” Environmental Geology, vol. 52, no. 6, pp. 1045–1052, 2007a.
View at: Publisher Site | Google Scholar
M. Amani, S. Parsian, S. M. MirMazloumi, and O. Aieneh, “Two new soil moisture indices based on the NIR-red triangle space of Landsat-8 data,” International Journal of Applied Earth Observation and Geoinformation, vol. 50, pp. 176–186, 2016.
View at: Publisher Site | Google Scholar
R. Nemani, L. Pierce, S. Running, and S. Goward, “Developing satellite-derived estimates of surface moisture status,” Journal of Applied Meteorology, vol. 32, no. 3, pp. 548–557, 1993.
View at: Publisher Site | Google Scholar
K. Czajkowski, S. Goward, S. J. Stadler, and A. Walz, “Thermal remote sensing of near surface environmental variables: application over the Oklahoma Mesonet,” The Professional Geographer, vol. 52, no. 2, pp. 345–357, 2000.
View at: Publisher Site | Google Scholar
G. P. Petropoulos, G. Ireland, and B. Barrett, “Surface soil moisture retrievals from remote sensing: current status, products & future trends,” Physics and Chemistry of the Earth, Parts A/B/C, vol. 83-84, pp. 36–56, 2015.
View at: Publisher Site | Google Scholar
T. N. Carlson, R. R. Gillies, and E. M. Perry, “A method to make use of thermal infrared temperature and NDVI measurements to infer surface soil water content and fractional vegetation cover,” Remote Sensing Reviews, vol. 9, no. 1-2, pp. 161–173, 1994.
View at: Publisher Site | Google Scholar
S. Stisen, I. Sandholt, A. Nørgaard, R. Fensholt, and K. H. Jensen, “Combining the triangle method with thermal inertia to estimate regional evapotranspiration: applied to MSG-SEVIRI data in the Senegal River basin,” Remote Sensing of Environment, vol. 112, no. 3, pp. 1242–1255, 2008.
View at: Publisher Site | Google Scholar
N. R. Patel, R. Anapashsha, S. Kumar, S. K. Saha, and V. K. Dadhwal, “Assessing potential of MODIS derived temperature/vegetation condition index (TVDI) to infer soil moisture status,” International Journal of Remote Sensing, vol. 30, no. 1, pp. 23–39, 2009.
View at: Publisher Site | Google Scholar
W. Wang, D. Huang, X. G. Wang, Y. R. Liu, and F. Zhou, “Estimation of soil moisture using trapezoidal relationship between remotely sensed land surface temperature and vegetation index,” Hydrology and Earth System Sciences, vol. 15, no. 5, pp. 1699–1712, 2011.
View at: Publisher Site | Google Scholar
H. Sun, “Two-stage trapezoid: a new interpretation of the land surface temperature and fractional vegetation coverage space,” Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 1, pp. 336–346, 2016.
View at: Publisher Site | Google Scholar
T. N. Carlson, E. M. Perry, and T. J. Schmugge, “Remote estimation of soil moisture availability and fractional vegetation cover for agricultural fields,” Agricultural and Forest Meteorology, vol. 52, no. 1-2, pp. 45–69, 1990.
View at: Publisher Site | Google Scholar
I. Sandholt, K. Rasmussen, and J. Andersen, “A simple interpretation of the surface temperature/vegetation index space for assessment of surface moisture status,” Remote Sensing of Environment, vol. 79, no. 2–3, pp. 213–224, 2002.
View at: Publisher Site | Google Scholar
Z. Wan, P. Wang, and X. Li, “Using MODIS land surface temperature and normalized difference vegetation Index products for monitoring drought in the southern Great Plains, USA,” International Journal of Remote Sensing, vol. 25, no. 1, pp. 61–72, 2004.
View at: Publisher Site | Google Scholar
A. Ghulam, Z. L. Li, Q. Qin et al., “A method for canopy water content estimation for highly vegetated surfaces-shortwave infrared perpendicular water stress index,” Science in China, Series A D, vol. 50, no. 9, pp. 1359–1368, 2007.
View at: Publisher Site | Google Scholar
M. E. Holzman, R. Rivas, and M. C. Piccolo, “Estimating soil moisture and the relationship with crop yield using surface temperature and vegetation index,” International Journal of Applied Earth Observation and Geoinformation, vol. 28, pp. 181–192, 2014.
View at: Publisher Site | Google Scholar
Q. Yuan, H. Xu, T. Li, H. Shen, and L. Zhang, “Estimating surface soil moisture from satellite observations using a generalized regression neural network trained on sparse ground-based measurements in the continental U.S,” Journal of Hydrology, vol. 580, Article ID 124351, 2020.
View at: Publisher Site | Google Scholar
A. Elshorbagy and K. Parasuraman, “On the relevance of using artificial neural networks for estimating soil moisture content,” Journal of Hydrology, vol. 362, no. 1-2, pp. 1–18, 2008.
View at: Publisher Site | Google Scholar
S. S. Zanetti, R. A. Cecilio, E. G. Alves, V. H. Silva, and E. F. Sousa, “Estimation of the moisture content of tropical soils using colour images and artificial neural networks,” Catena, vol. 135, pp. 100–106, 2015.
View at: Publisher Site | Google Scholar
E. Santi, S. Paloscia, S. Pettinato, and G. Fontanelli, “Application of artificial neural networks for the soil moisture retrieval from active and passive microwave spaceborne sensors,” International Journal of Applied Earth Observation and Geoinformation, vol. 48, pp. 61–73, 2016.
View at: Publisher Site | Google Scholar
J. Kolassa, R. H. Reichle, Q. Liu et al., “Estimating surface soil moisture from SMAP observations using a Neural Network technique,” Remote Sensing of Environment, vol. 204, pp. 43–59, 2018.
View at: Publisher Site | Google Scholar
L. Gao, Q. Gao, H. Zhang et al., “A deep neural network based SMAP soil moisture product,” Remote Sensing of Environment, vol. 277, Article ID 113059, 2022.
View at: Publisher Site | Google Scholar
J. Liu, Y. Xu, H. Li, and J. Guo, “Soil moisture retrieval in farmland areas with Sentinel multi-source data based on regression convolutional neural networks,” Sensors, vol. 21, no. 3, p. 877, 2021.
View at: Publisher Site | Google Scholar
S. Nativel, E. Ayari, N. Rodriguez-Fernandez et al., “Hybrid methodology using sentinel-1/sentinel-2 for soil moisture estimation,” Remote Sensing, vol. 14, no. 10, p. 2434, 2022.
View at: Publisher Site | Google Scholar
X. Yang, C. Zhang, Z. Cui, Y. Fan, and J. Wang, “Yingjuan Han (2018) Filling method for soil moisture based on BP neural network,” Journal of Applied Remote Sensing, vol. 4, Article ID 42806, 2018.
View at: Publisher Site | Google Scholar
K. Yang, J. Qin, L. Zhao et al., “A multiscale soil moisture and freeze–thaw monitoring network on the Third Pole,” Bulletin of the American Meteorological Society, vol. 94, no. 12, pp. 1907–1916, 2013.
View at: Publisher Site | Google Scholar
R. Zhuang, Y. Zeng, S. Manfreda, and Z. Su, “Quantifying long-term land surface and root zone soil moisture over Tibetan plateau,” Remote Sensing, vol. 12, no. 3, p. 509, 2020.
View at: Publisher Site | Google Scholar
M. Sadeghi, E. Babaeian, M. Tuller, and S. B. Jones, “The optical trapezoid model: a novel approach to remote sensing of soil moisture applied to Sentinel-2 and Landsat-8 observations,” Remote Sensing of Environment, vol. 198, pp. 52–68, 2017.
View at: Publisher Site | Google Scholar
L. Zhang, Y. Zeng, R. Zhuang et al., “In situ observation-constrained global surface soil moisture using random forest model,” Remote Sensing, vol. 13, no. 23, p. 4893, 2021.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2023 Dayou Luo et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

498

Downloads

416

Citations