Get Our e-AlertsSubmit Manuscript
Journal of Remote Sensing / 2022 / Article

Research Article | Open Access

Volume 2022 |Article ID 9871246 | https://doi.org/10.34133/2022/9871246

Haoxuan Yang, Qunming Wang, Wei Zhao, Xiaohua Tong, Peter M. Atkinson, "Reconstruction of a Global 9 km, 8-Day SMAP Surface Soil Moisture Dataset during 2015–2020 by Spatiotemporal Fusion", Journal of Remote Sensing, vol. 2022, Article ID 9871246, 23 pages, 2022. https://doi.org/10.34133/2022/9871246

Reconstruction of a Global 9 km, 8-Day SMAP Surface Soil Moisture Dataset during 2015–2020 by Spatiotemporal Fusion

Received16 Dec 2021
Accepted05 Jul 2022
Published25 Jul 2022

Abstract

Soil moisture, a crucial property for Earth surface research, has been focused widely in various studies. The Soil Moisture Active Passive (SMAP) global products at 36 km and 9 km (called P36 and AP9 in this research) have been published from April 2015. However, the 9 km AP9 product was retrieved from the active radar and L-band passive radiometer and the active radar failed in July 2015. In this research, the virtual image pair-based spatiotemporal fusion model was coupled with a spatial weighting scheme (VIPSTF-SW) to simulate the 9 km AP9 data after failure of the active radar. The method makes full use of all the historical AP9 and P36 data available between April and July 2015. As a result, 8-day composited 9 km SMAP data at the global scale were produced from 2015 to 2020, by downscaling the corresponding 8-day composited P36 data. The available AP9 data and in situ reference data were used to validate the predicted 9 km data. Generally, the predicted 9 km SMAP data can provide more spatial details than P36 and are more accurate than the existing EP9 product. The VIPSTF-SW-predicted 9 km SMAP data are an accurate substitute for AP9 and will be made freely available to support research and applications in hydrology, climatology, ecology, and many other fields at the global scale.

1. Introduction

Soil moisture is considered to be a significant property in a range of models including those in applied ecosystem cycling [1, 2], water resources [3], drought monitoring [4, 5], and soil evaporation studies [6, 7]. Sensors installed on the ground are one of the earliest solutions to measure soil moisture and can acquire real-time in situ data. These in situ data provide accurate measurements of soil moisture, and they are commonly regarded as a ground reference for quantitative evaluation of soil moisture predicted by other platforms (e.g., satellite sensors) [810]. However, in situ data are commonly spatially sparse, that is, observations are provided only at the limited sensor locations available. Furthermore, the high economic and labor costs further hamper development of this measurement approach, especially for large area monitoring [11]. In contrast, satellite sensors can provide spatially continuous soil moisture data with large spatial coverage (i.e., including at the global scale). Therefore, the soil moisture data acquired by satellite sensors have been used widely in practical applications [12].

The Soil Moisture Active Passive (SMAP) satellite supported by the National Aeronautics and Space Administration (NASA) was launched in January 2015 and started to provide data in April 2015. The SMAP product is always considered as a type of dataset with greater accuracy than the other soil moisture products [1315], as exhibited in Table 1. For the SMAP satellite, two sensors, including an L-band active radar (1.26 GHz) and a passive radiometer (1.41 GHz), are carried by to measure global surface soil moisture at a fine temporal resolution [16]. As a result, a spatially complete global soil moisture map can be acquired every 2 to 3 days. The SMAP data are published at spatial resolutions of 36 km (called P36 in this paper), 9 km (called AP9 in this paper), and 3 km, which are retrieved using the passive radiometer observation, the combined active radar and passive radiometer, and the active radar observation, respectively [17, 18]. However, the onboard active radar operated only for approximately 3 months (13 April 2015 to 07 July 2015) and failed on 08 July 2015. Subsequently, directly retrieved SMAP soil moisture data at 3 km and 9 km were not provided.


Satellite/sensor/productGridded spatial resolutionTemporal resolutionLaunched time (mm-yyyy)Reference

Soil Moisture Active Passive (SMAP)3, 9, and 36 km01-2015[16, 17]
Soil Moisture and Ocean Salinity (SMOS)25 km11-2009[19]
Advanced Microwave Scanning Radiometer 2 (AMSR2)25 km05-2002[20]
Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) European Space Agency’s Soil25 km days05-2012[21]
Moisture Climate Change Initiative (CCI)25 km11-1978 (the earliest release)[22]
Advanced Scatterometer (ASCAT)25 and 50 km10-2006[23]
Fengyun-3 (FY)25 km11-2010[24]

To cope with the failure of the active radar, the enhanced SMAP passive 9 km soil moisture product (called EP9 in this paper) was created to extend the 9 km soil moisture data. The EP9 data were retrieved using interpolated brightness temperature based on the single-channel algorithm [25, 26]. Although both EP9 and AP9 are at the same 9 km spatial resolution, AP9 has been observed to contain more spatial details than EP9. To increase the spatial details, another product at 3 km/1 km spatial resolution was published by NASA, using a C-band synthetic aperture radar from Sentinel-1 A/B to substitute the defunct L-band active radar on-board SMAP, that is, the SMAP/Sentinel-1 L2 soil moisture product (called SM_SP in this paper). Nevertheless, the temporal resolution of Sentinel-1 A/B is coarser than SMAP due to the narrow swath [2729].

In recent years, a variety of methods have been developed to downscale the 36 km soil moisture data (i.e., P36) from the passive radiometer of SMAP. Zhao et al. [9] developed an ensemble learning-based random forest model to downscale the 36 km SMAP product to 1 km, in which 1 km optical and thermal infrared Moderate Resolution Imaging Spectroradiometer (MODIS) observations were used as auxiliary data. Likewise, Hu et al. [30] incorporated finer spatial resolution visible and shortwave-infrared data (such as land surface temperature (LST), normalized difference vegetation index (NDVI), and brightness temperature (TB)) in a random forest model to downscale the 36 km SMAP data. Wei et al. [31] proposed a gradient boosting decision tree method to downscale the P36 data to 1 km, coupled with a large number of soil moisture indices derived from MODIS and a digital elevation model (DEM). On the other hand, spatial interpolation-based methods have also been developed for downscaling the P36 data. Based on the well-established area-to-point regression kriging (ATPRK) method [32, 33], a geographically weighted version proposed by Jin et al. [34] replaces the original linear regression in ATPRK to account for being spatially nonstationary in downscaling. Wen et al. [35] further extended the geographically weighted version with a self-adaptive window in regression modeling. In addition, similar methods have also been developed for coarse soil moisture data acquired by other sensors, such as the European Space Agency’s Soil Moisture Climate Change Initiative (CCI) [36], Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) [37], and Advanced Microwave Scanning Radiometer 2 (AMSR-2) [38].

All of the aforementioned methods require auxiliary data (e.g., NDVI, TB, LST, and DEM) at finer spatial resolution. The auxiliary data are acquired by other sensors and influence the accuracy of downscaling directly. In particular, a key issue is that the auxiliary data acquired from other sources represent different physical meanings. Thus, the spatial texture of the fine spatial resolution auxiliary data may not be highly correlated with that of the fine spatial resolution SMAP data. Moreover, auxiliary data acquired by optical sensors can be contaminated easily by cloud and haze, leading to spatially incomplete data and increased uncertainty in the downscaling predictions [39].

Jiang et al. [18] first proposed to extend the 9 km SMAP product (i.e., AP9) by fusing the P36 product with the available AP9 data. As the baseline data, the 8-day (the last eight days of working active radar in the three months) global P36 and AP9 were considered as the known image pair of the fusion model, which was used to downscale the P36 data on other days (i.e., between 8 July 2015 and 12 April 2017) to 9 km. Jiang et al. [18] provided a pioneering study blending AP9 data for downscaling P36 data. However, two issues remain to be addressed. Firstly, only a single image pair was used in spatiotemporal fusion in [18]. The image pair plays a crucial role in spatiotemporal fusion, as the included fine spatial resolution image provides indispensable information for predicting the target image at the same fine spatial resolution on other days. Generally, the accuracy of spatiotemporal fusion tends to be greater when more image pairs are considered [40]. Thus, it is worthwhile to use more image pairs for downscaling the P36 product. Secondly, only two years (from July 2015 to April 2017) of AP9 data were produced in [18]. It is worthwhile to create longer time-series AP9 data at the global scale, which can be applied straightforwardly by researchers, for example, in hydrology, meteorology, climatology, and ecology.

In this paper, we created a long time-series, 8-day composited soil moisture product at 9 km and at the global scale, by spatiotemporal fusion of P36 and available AP9 data. We tackled directly the two abovementioned issues. Firstly, we considered multiple image pairs in the fusion of P36 and AP9. The use of multiple AP9-P36 pairs is feasible, as the AP9 data are available from 13 April 2015 to 07 July 2015 and there are about 10 AP9 images (8-day composited product) in total, with each providing full cover at the global scale. The composited period is the same as the well-known 8-day composited MODIS products to facilitate the postapplications where MODIS data are required. Secondly, 9 km global soil moisture data were produced over five years (from July 2015 to August 2020). The new product will be made publicly available to support global research in hydrology and other related fields.

2. Data and Methods

2.1. Data
2.1.1. Satellite Sensor Data

Three level 3 SMAP products with global coverage were applied in this research, including P36, AP9, and EP9. Different versions of the SMAP product were released by NASA due to continuous optimization of the SMAP products through time. In this research, the version 6 P36 and version 3 AP9 and EP9 were considered. It should be noted that there are no substantial changes between different versions and only one version was used for each SMAP product. Thus, the difference in data version was neglected in this research. Two groups of experiments with disparate periods were designed. Specifically, the period of experiment 1 is the same as that of AP9 (i.e., 14 April 2015 to 07 July 2015), while the period of experiment 2 is after the failure of AP9 (i.e., 12 July 2015 to 19 August 2020). In addition, regional SM_SP data at 3 km were also used in this research. Details of the used SMAP data are listed in Table 2. The data can be acquired freely from the National Snow and Ice Data Center (NSIDC) (https://nsidc.org/).


DatasetSpatial resolutionData levelData versionData period (mm-dd-yyyy)

P3636 kmLevel 3Version 6Experiment 1: 04-14-2015 to 07-07-2015, experiment 2: 07-12-2015 to 08-19-2020
AP99 kmLevel 3Version 3Experiment 1: 04-14-2015 to 07-07-2015
EP99 kmLevel 3Version 3Experiment 1: 04-14-2015 to 07-07-2015, experiment 2: 01-01-2019 to 01-02-2020
SM_SP3 kmLevel 2Version 3Discussion: 01-01-2019 to 11-30-2019

2.1.2. In Situ Soil Moisture Data

The in situ data are soil moisture data measured on the ground and have been used widely as reference for evaluating the accuracy of soil moisture products retrieved from satellite sensor data [4143]. The in situ data are provided freely by the International Soil Moisture Network (ISMN) (https://ismn.earth/en/). The soil moisture observation networks are set up across the entire globe by several institutions with different sensors for various purposes. Accordingly, the networks have different measurement periods and one or more observation depths in some cases [4447]. Thus, based on the availability of the in situ data at the given periods, different networks were considered in experiments 1 and 2 in this research. Additionally, two networks (i.e., Fort Cobb and Little Washita) in Oklahoma, USA, were used for the regional comparison, which are provided by the US Department of Agriculture, Oklahoma State University, and the Oklahoma Climatological Survey. The location of the in situ data is exhibited in Figure 1. It should be stressed that only topsoil moisture (<0.05 m) can be retrieved effectively from L-band radiometers (with ~1.4 GHz frequency and ~21 cm wavelength), since they cannot penetrate deep soil [4850]. Therefore, the measured depths of the in situ data selected for validation are less than 0.05 m. In addition, for some points, parts of the in situ data are missing along the time series, which were excluded for consideration in the experiments. The locations of the used in situ data are listed in Table 3.


SectionNetworkNumber of in situ pointsLocationDepth (m)Sensor

Experiment 1AMMA-CATCH3Benin, Niger, Mali0.05–0.05CS616
CTP_SMTMN14China0.00–0.055TM
DAHRA1Senegal0.05–0.05ThetaProbe-ML2X
FLUXNET-AMERIFLUX2USA0.00–0.00ThetaProbe-ML2X
FR_Aqui3France0.05–0.05ThetaProbe-ML2X
HOBE17Denmark0.00–0.05Decagon-5TE
MySMNet5Malaysia0.00–0.05WaterScout-SM100
NGARI7China0.05–0.055TM
OZNET16Australia0.00–0.05Stevens-Hydra-Probe
PBO_H2O82USA0.00–0.05GPS
REMEDHUS14Spain0.00–0.05Stevens-Hydra-Probe
RISMA17Canada0.00–0.05Hydraprobe-II-Sdi-12
RSMN16Romania0.00–0.055TM
SCAN105USA0.05–0.05Hydraprobe-Analog-(2.5-Volt)
SMOSMANIA16France0.05–0.05ThetaProbe-ML2X
SNOTEL35USA0.05–0.05Hydraprobe-Analog-(2.5-Volt)
SOILSCAPE47USA0.05–0.05EC5
TERENO4Germany0.05–0.05Hydraprobe-II-Sdi-12
USCRN70USA0.05–0.05Stevens Hydraprobe II Sdi-12

Experiment 2FLUXNETAMERIFLUX4USA0.00–0.00ThetaProbe-ML2X
FR_Aqui3France0.05–0.05ThetaProbe-ML2X
HOBE18Denmark0.00–0.05Decagon-5TE
NGARI8China0.05–0.055TM
REMEDHUS19Spain0.00–0.05Stevens-Hydra-Probe
RSMN14Romania0.00–0.055TM
SCAN82USA0.05–0.05Hydraprobe-Analog-(2.5-Volt)
SMN-SDR22China0.05–0.055TM
SMOSMANIA7France0.05–0.05ThetaProbe-ML2X
SNOTEL54USA0.05–0.05Hydraprobe-Analog-(2.5-Volt)
TERENO4Germany0.05–0.05Hydraprobe-II-Sdi-12
USCRN83USA0.05–0.05Stevens Hydraprobe II Sdi-12

DiscussionFort Cobb15USA0.05–0.05Stevens-Hydra-Probe
Little Washita20USA0.05–0.05Stevens-Hydra-Probe

2.2. Methods

The virtual image pair-based spatiotemporal fusion (VIPSTF) method provides a flexible framework for spatiotemporal fusion, by creating a virtual image pair (VIP) that blends the information from multiple image pairs [40]. Accordingly, two versions were derived, including spatial unmixing-based VIPSTF (VIPSTF-SU) and spatial weighting-based VIPSTF (VIPSTF-SW). VIPSTF-SW has been demonstrated to be the more accurate in [40]. Thus, this method was used to extend the AP9 data by fusion of P36 and existing time-series data. Specifically, the predicted AP9 is expressed as follows: where is the predicted 9 km SMAP data at time , is the AP9 image in the VIP, and is the 9 km increment image predicted using a spatial weighting scheme.

The image is produced with a linear combination of the multiple known AP9 images: where is a transformation coefficient for the th image , is a constant, and is the number of known AP9 images. In experiment 1, each of the 10 scenes was predicted in turn, using the other 9 scenes. Therefore, was set to 9 in experiment 1. In experiment 2, all the 10 scenes were used as the inputs. Thus, was set to 10 in experiment 2. Based on the assumption of scale invariance [40, 51], the optimal coefficient set (i.e., and ) is estimated based on the linear regression model fitted between the corresponding P36 images:

In equation (3), is the th known P36 image and refers to the residual image. The optimal coefficient set is predicted using the least squares fitting method. The 9 km increment image is predicted based on the following spatial weighting scheme: where is the spatial location of the similar pixels surrounding the center pixel (), refers to a weight calculated according to the distance between the center pixel and the th surrounding similar pixel, and is the number of the surrounding similar pixels [52]. was set to 5 in this research. Note that the image needs to be interpolated to 9 km in advance to match the spatial resolution of AP9 and the bicubic interpolation was employed in this paper for this purpose. Figure 2 shows the whole process of VIPSTF-SW for fusion of P36 and AP9.

2.3. Validation Strategy

The predicted SMAP soil moisture data at 9 km spatial resolution were validated with either real AP9 data or in situ data, or both. Specifically, for experiment 1, both the real AP9 and in situ data were applied for validation of the predicted AP9 data. For experiment 2, however, only the in situ data were used, as the real AP9 data were not available in the period (i.e., 12 July 2015 to 19 August 2020). For the evaluation using the real AP9 data, the pixels of AP9 data were used directly for evaluation, considering that the spatial resolution (i.e., 9 km) is consistent. For the evaluation using in situ data, the average in situ data at the level of networks were used for validation to reduce the measurement errors caused by the numerous stations. Specifically, the corresponding pixel of each measuring station was extracted from the satellite-derived soil moisture data based on their locations. Then, the soil moisture values of the extracted pixels were averaged as the measured value and compared with the data of the corresponding network. In addition to the spatial evaluation (i.e., evaluation was performed on the map at each time in turn), the temporal evaluation (i.e., evaluation was performed on each pixel with temporal profile in turn) was also implemented.

Six quantitative evaluation indices were employed for accuracy assessment in this research, including the correlation coefficient (CC), root mean squared error (RMSE), bias (Bias), unbiased root mean squared error (ubRMSE), relative global-dimensional synthesis error (ERGAS) [53], and universal image quality index (UIQI) [54]. For validation using real AP9 data in experiment 1, all six indices were utilized for accuracy assessment. With respect to the validation based on the use of the in situ data in both experiments 1 and 2, and were not considered, as these two indices are calculated based on spatially complete reference data, rather than spatially sparse in situ data. where represents the real soil moisture data (i.e., either AP9 or in situ data) and represents the predicted data. In addition, is the covariance function, means the expectation, represents the variance, and are the fine and coarse spatial resolution (i.e., ), respectively, and means the number of bands.

3. Experiments and Results

The experimental design is shown in Figure 3. In experiment 1, each of the 10 8-day composited AP9 data was predicted, in turn, with the other nine AP9 data as known images for fusion. The acquisition times for the AP9 data are shown in Table 4. Accordingly, all data in experiment 1 (i.e., P36, AP9, EP9, and in situ data) were composited to produce 8-day composited data for testing. Meanwhile, the composited data of AP9 and P36, that is, the 10 known AP9-P36 image pairs, were also applied in experiment 2. For clarity, we denoted the AP9 data predicted using VIPSTF-SW as VIPSTF-SW9 (i.e., in equation (1)). The published EP9 and P36 were used as benchmark data to show the advantage of VIPSTF-SW9. In experiment 2, the VIPSTFSW9 data were produced, in turn, from 12 July 2015. Again, all data in experiment 2 (i.e., P36, EP9 and in situ data) were composited with the same rule.


Data3

Time04-14-201504-22-201504-30-201505-08-201505-17-201505-25-201506-02-201506-10-201506-20-201506-29-2015
04-15-201504-23-201505-01-201505-09-201505-18-201505-26-201506-03-201506-11-201506-21-201506-30-2015
04-16-201504-24-201505-02-201505-10-201505-19-201505-27-201506-04-201506-12-201506-22-201507-01-2015
04-17-201504-25-201505-03-201505-11-201505-20-201505-28-201506-05-201506-13-201506-23-201507-02-2015
04-18-201504-26-201505-04-201505-12-201505-21-201505-29-201506-06-201506-14-201506-24-201507-03-2015
04-19-201504-27-201505-05-201505-14-201505-22-201505-30-201506-07-201506-15-201506-25-201507-04-2015
04-20-201504-28-201505-06-201505-15-201505-23-201505-31-201506-08-201506-17-201506-26-201507-05-2015
04-21-201504-29-201505-07-201505-16-201505-24-201506-01-201506-09-201506-18-201506-27-201507-06-2015
06-19-201506-28-201507-07-2015

The expression for the date is mm/dd/yyyy. We did not collect the P36 and AP9 data on 05-13-2015 and 06-16-2015 since they are unavailable.
3.1. Experiment 1
3.1.1. Validation Based on AP9 Data

The results at and are shown as examples in Figure 4, where four subfigures are also zoomed to aid comparison. It is seen clearly that the available EP9 data are visually smooth, even though they contain more spatial details than the original P36 data. Compared with the EP9 data, the VIPSTF-SW9 predictions reproduce more spatial heterogeneity and are closer to the reference AP9 data. Table 5 lists the quantitative evaluation VIPSTF-SW9 and EP9 results for the predictions made 10 times, where the real AP9 data are used as reference. By referring to the real AP9 data, there are obvious differences between the accuracies of VIPSTF-SW9 and EP9 data for all 10 cases. The VIPSTF-SW9 data have an average CC of 0.968, which is 0.095 larger than that of EP9. Furthermore, the average UIQI and ERGAS of VIPSTF-SW9 are 0.118 larger and 5.465 smaller than those of EP9, respectively. Both the average RMSE and ubRMSE of VIPSTF-SW9 are 0.41 smaller than those of EP9. In addition, The VIPSTF-SW9 data have an average Bias of 0.002, which is closer to the reference than that of EP9. The advantage of VIPSTF-SW9 over EP9 is also illustrated by the spatial bias in Figure 5.


CCUIQIERGASRMSEBiasubRMSE
VIPSTFSW9EP9VIPSTFSW9EP9VIPSTFSW9EP9VIPSTFSW9EP9VIPSTFSW9EP9VIPSTFSW9EP9

0.9490.8860.9490.8637.08310.7140.0520.0780.001−0.0120.0520.077
0.9600.8780.9600.8556.40111.1880.0470.0830.002−0.0090.0470.082
0.9600.8800.9590.8536.45411.1850.0480.0840.005−0.0060.0480.083
0.9690.8700.9680.8435.50011.1020.0420.0850.003−0.0030.0420.085
0.9760.8740.9750.8494.83810.9140.0370.0840.003−0.0040.0370.083
0.9780.8720.9780.8474.61711.0090.0350.0830.000−0.0050.0350.083
0.9760.8690.9760.8464.72510.8420.0360.0810.002−0.0050.0350.081
0.9740.8620.9730.8394.93611.0810.0370.0830.000−0.0060.0370.083
0.9720.8710.9710.8505.13210.7670.0380.0800.002−0.0060.0380.080
0.9710.8710.9700.8505.22810.7600.0380.079−0.001−0.0080.0380.079
Ave0.9680.8730.9680.8505.49110.9560.0410.0820.0020.0070.0410.082

The critical reason for the difference between VIPSTF-SW9 and EP9 is that the latter is produced using only interpolated brightness temperature without auxiliary fine spatial resolution information. In contrast, the VIPSTF-SW9 data are produced by taking the known fine spatial resolution AP9 data into account comprehensively.

3.1.2. Validation Based on the In Situ Data in Experiment 1

The in situ soil moisture data were measured continuously from fixed sensors and considered as the reference for validation in this experiment. As shown in Figure 6, the accuracy at each time of the 10 cases is provided. Among the official products (i.e., P36, AP9, and EP9 data), at each time, the real AP9 data have the greatest consistency with the in situ data in experiment 1. This is because the AP9 data are collaboratively retrieved from the active radar and passive radiometer. However, the other data are not directly derived from the active radar with reliable support to the prediction of 9 km data. Therefore, the accuracy of the other official products is smaller than that of the AP9 data. Besides, the accuracy of the predicted VIPSTF-SW9 dataset is similar to that of the AP9 data. Although the average accuracy and stability of VIPSTF-SW9 data are slightly smaller than those of the real AP9 data (see, for example, the box charts of Figure 6), the predicted VIPSTF-SW9 data still have a more accurate performance than the EP9 and P36 data.

3.2. Experiment 2
3.2.1. 9 km SMAP Data Predicted during 2015–2020

The 10 8-day composited image pairs (AP9-P36) in experiment 1 were applied to produce the VIPSTF-SW9 data after the failure of AP9 (i.e., from July 2015 to August 2020) in experiment 2. For experiment 2, a total of 230 global VIPSTF-SW9 maps were produced from 12 July 2015 to 19 August 2020 (marked as 2015193 to 2020225). Due to the continuous absence of P36 from 20 June 2019 to 22 July 2019, VIPSTF-SW9 predictions from 2019169 to 2019201 are unavailable. The repository holding the predicted 5-year product of global VIPSTF-SW9 data is available at doi:10.6084/m9.figshare.14634276. As shown in Figure 7, we exhibit all VIPSTFSW9 data in 2016 (46 scenes). For convenience of visual inspection, a subregion of both VIPSTFSW9 and the original P36 data is shown in Figures 8 and 9. It can be seen clearly that the predicted VIPSTF-SW9 data present more spatial details than P36 data, revealing the advantages of the VIPSTFSW9 data. In addition, the composited products for each quarter (i.e., three months) from 2015 to 2020 are displayed in Figure 10.

3.2.2. Validation Based on the In Situ Data in Experiment 2

As shown in Figure 11, the results in experiment 2 were evaluated using the in situ data, as the real AP9 data were unavailable for the five years. The in situ data in the whole year of 2019 were selected for validation. Note that the P36 and EP9 data were not available from 2019169 to 2019201. Noticeably, there are differences in the accuracies of the results between experiments 1 and 2. The main reason is that the 9 km data in the two experiments were predicted in completely different periods. It is observed that the VIPSTF-SW9 data are more accurate than that of the P36 and EP9 data. Moreover, the accuracy of the EP9 time series is close to that of P36. This is because the spatial details acquired from the available AP9 data were injected into VIPSTF-SW9 data by spatiotemporal fusion. Inversely, P36 and EP9 data were retrieved from coarse and interpolated brightness temperature of SMAP, respectively, without prior spatial details.

In addition to the abovementioned spatial evaluation, the accuracy was also evaluated in the temporal domain. That is, the temporal file of each network was compared with that of the in situ time series at the same location. As shown in Table 6, the results of all 11 networks are provided and the average of all results at each network is also provided for comparison. It can be observed that VIPSTF-SW9 data present the greater performance than the P36 and EP9 data for almost all networks. Checking the average results, the VIPSTF-SW9 data have the largest mean CC of 0.787, which are 0.014 and 0.017 larger than those of the P36 and EP9 data. The mean RMSE of VIPSTFSW9 data is 0.008 smaller than those of the P36 and EP9 data. Meanwhile, the mean Bias of the VIPSTF-SW9 data is −0.006, which is much closer to the reference (i.e., in situ data) than that of the P36 and EP9 data (with a Bias of −0.025).


NetworkDatasetCCRMSEBiasubRMSE

FR_AquiP360.9450.076−0.0720.027
EP90.9480.065−0.0610.023
VIPSTF-SW90.9540.0390.0240.031

HOBEP360.5180.0570.0410.040
EP90.5110.069−0.0590.037
VIPSTF-SW90.5190.064−0.0520.037

NGARIP360.9390.012−0.0060.010
EP90.9560.0090.0030.009
VIPSTF-SW90.9220.0320.0270.017

REMEDHUSP360.8490.035−0.0120.033
EP90.8110.0360.0040.036
VIPSTF-SW90.8860.0280.0070.027

RSMNP360.6440.063−0.0540.032
EP90.6010.060−0.0510.033
VIPSTF-SW90.7280.0460.0370.027

SCANP360.9810.043−0.0420.008
EP90.9750.035−0.0340.009
VIPSTF-SW90.9800.0110.0080.008

SMN-SDRP360.3320.0470.0250.040
EP90.3330.0470.0250.040
VIPSTF-SW90.3290.0530.0390.036

SMOSMANIAP360.8350.0470.0370.028
EP90.8900.053−0.0470.023
VIPSTF-SW90.9120.044−0.0380.022

SNOTELP360.7250.0640.0320.056
EP90.7460.0640.0330.055
VIPSTF-SW90.6680.0620.0270.056

TERENOP360.7750.067−0.0510.043
EP90.7420.071−0.0520.047
VIPSTF-SW90.8120.0470.0220.041

USCRNP360.9570.025−0.0190.017
EP90.9520.025−0.0170.019
VIPSTF-SW90.9410.0210.0100.018

AverageP360.7730.049−0.0250.030
EP90.7700.049−0.0250.030
VIPSTF-SW90.7870.0410.0060.029

4. Discussion

4.1. The Advantage of Using Multiple Image Pairs in the Spatiotemporal Model

Jiang et al. [18] used a single image pair in their spatiotemporal model for downscaling P36. For spatiotemporal fusion of reflectance images, the use of multiple image pairs has been demonstrated to be more beneficial than the use of a single image pair [40]. It is necessary to further examine whether this scheme is also more advantageous for spatiotemporal fusion of the SMAP data investigated in this paper. To this end, the performances of both single and multiple image pairs were compared here. For the single image pair case, each of the AP9-P36 image pairs was considered as the known data in VIPSTF-SW, in turn, for downscaling the other nine coarse P36 data. The multiple image pair case is exactly the same as that in experiment 1. The accuracies of both cases are shown in Figure 12. As shown in Figure 12, for any prediction time, the average accuracies of using multiple image pairs are obviously greater than those using a single image pair. This validates rigorously the advantage of using multiple image pairs in this paper over the scheme of using a single image pair in [18].

4.2. Which Data Are Closer to the True Soil Moisture?

For a given spatial location, the availability of the in situ data varies across time. Since in this research, we sought to maximize the number of in situ data for evaluation of the predictions, the spatial distributions of the in situ data in experiments 1 and 2 are not consistent. To further discuss the reliability of the products in a continuous period, the common in situ data in both experiments 1 and 2 were used. The in situ data from 174 stations and nine networks were collected. The averages of the in situ data at the station level were calculated for evaluation. As shown in Figure 13, three main points can be observed. Firstly, the time series of AP9 and VIPSTF-SW9 are the most similar to that of the in situ data. Secondly, the time series of P36 and EP9 are highly similar and both are more different from that of the in situ data and AP9 and VIPSTF-SW9 (i.e., the soil moisture values of P36 and EP9 are much larger than of the other datasets). Thirdly, AP9 and VIPSTF-SW9 are close to each other, as seen in the results for experiment 1. This further supports the possibility of using the VIPSTF-SW9 data instead of the AP9 data after the failure of AP9.

4.3. Regional Comparison of Derived VIPSTF-SW9 and SM_SP Data

The derived VIPSTF-SW9 data were compared with the SM_SP data at a regional scale, that is, in Oklahoma, USA, in 2019. The daily SM_SP data were composited into 8-day temporal resolution to match the VIPSTF-SW9 data. The in situ data at two independent networks (i.e., Fort Cobb and Little Washita) were used as the reference. As shown in Figure 14, the time series of the derived VIPSTF-SW9 data is more similar to the reference than that of SM_SP. Moreover, there exist many gaps in the time series of SM_SP data. The main reason is that the available SM_SP data are cooperatively derived from SMAP and Sentinel-1, which have disparate revisit orbits and swath width (the width of Sentinel-1 is much smaller than that of SMAP). Therefore, the spatiotemporal distribution of effective SM_SP data is irregular, leading to great uncertainty in the data composition process of the SM_SP data. As shown in Figure 15, there exist many gaps in the composited SM_SP data. Generally, compared with the SM_SP product, the VIPSTF-SW9 data are more suitable for applications at a large spatiotemporal scale.

4.4. Uncertainty in Residual Correction for VIPSTF-SW9

Residual correction is aimed at making the predicted fine spatial resolution image consistent with the original coarse data (i.e., perfect coherence). The well-established ATPRK method can achieve reliable residual correction to preserve perfectly the observed coarse images [55]. Thus, this method was also applied to SMAP downscaling, producing 9 km SMAP data with perfect coherence with the original P36 data. The data in experiment 1 were used for testing, and the predicted 9 km fine SMAP data were denoted as ATPRK9. The accuracies of the 9 km predictions are shown in Table 7. It is seen that the accuracy of ATPRK9 is similar to EP9, but obviously less than for VIPSTF-SW9. Since ATPRK9 has perfect coherence with P36, the results suggest that it is not necessary to preserve the original 36 km information in predicting AP9. To further investigate the relation between AP9, VIPSTF-SW9, and P36, both AP9 and VIPSTF-SW9 data were upscaled to 36 km. As the reference at 36 km, the upscaled AP9 data were used to evaluate the upscaled VIPSTF-SW9 and P36 data, as shown in Table 8 and Figure 16. We find that there exist obvious differences between P36 and the upscaled VIPSTF-SW9 and the latter is closer to the upscaled AP9. Consequently, it is not beneficial to perform residual correction for VIPSTF-SW9 due to the inherent gap between AP9 and P36.


CCUIQIERGASRMSEBiasubRMSE

0.8790.87310.8120.079−0.0100.079
0.8770.87011.0760.082−0.0080.081
0.8740.86611.1720.084−0.0050.083
0.8390.83312.0720.0930.0020.093
0.8680.86110.9530.084−0.0030.084
0.8670.86011.0330.083−0.0040.083
0.8650.85910.8780.082−0.0030.082
0.8580.85211.1570.083−0.0040.083
0.8670.86210.8440.081−0.0040.081
0.8650.86110.9060.080−0.0060.080
ATPRK9-Ave0.8660.86011.0900.0830.0040.083

EP9-Ave0.8730.85010.9560.0820.0070.082
VIPSTF-SW9-Ave0.9680.9685.4910.0410.0020.041


CCUIQIERGASRMSEBiasubRMSE
P36Upscaled VIPSTF-SW9P36Upscaled VIPSTF-SW9P36Upscaled VIPSTF-SW9P36Upscaled VIPSTF-SW9P36Upscaled VIPSTF-SW9P36Upscaled VIPSTF-SW9

0.8830.9840.8420.98338.71712.5120.0500.016−0.010−0.0010.0480.016
0.8740.9810.8340.98039.70513.5900.0530.018−0.010−0.0010.0510.018
0.8750.9880.8430.98737.88410.6470.0530.015−0.0100.0000.0510.015
0.8650.9860.8330.98638.35410.8810.0560.016−0.010−0.0010.0540.016
0.8660.9940.8370.99436.8386.9910.0570.011−0.0100.0000.0550.011
0.8620.9950.8290.99537.2146.4300.0580.010−0.011−0.0010.0560.010
0.8590.9950.8270.99536.8085.8980.0590.009−0.0110.0000.0570.009
0.8500.9930.8110.99238.2867.4840.0610.012−0.012−0.0010.0590.012
0.8540.9950.8180.99537.3786.0930.0610.010−0.0120.0000.0580.010
0.8500.9940.8090.99438.3216.5090.0610.010−0.012−0.0010.0590.010
Ave0.8640.9910.8280.99037.9518.7030.0570.0130.0110.0000.0550.013

4.5. Uncertainty in the Soil Moisture Data

There exists inevitable uncertainty in soil moisture data. The uncertainties are mainly twofold. Firstly, there is uncertainty in the spatial support of the in situ data, which are measured by fixed stations and almost at the point scale. However, the SMAP data are composed of pixels with regular spatial coverage and the signal of each SMAP pixel represents the convolution of soil moisture over a much wider area (e.g., 9 km by 9 km in AP9, but potentially much larger than this due to the point spread function effect). Therefore, the mismatch in the spatial scale between the in situ data and the SMAP data leads to uncertainty in accuracy evaluation. Secondly, vegetation commonly covers the topsoil in many regions and the interference of vegetation can increase the uncertainty of soil moisture prediction, especially for satellite soil moisture products that are captured from above the vegetation, to reduce the uncertainty caused by the cover of vegetation. Yang et al. [56] acquired satellite sensor images in periods when the soil was bare to reduce the effect of vegetation. However, this method is unsuitable for studies at the global scale. Furthermore, surface soil moisture retrieved from satellite sensors can be inevitably affected by the vegetation water content. The vegetation optical depth (VOD) data derived from several sensors can reflect the water content in vegetation to some extent. Therefore, with the continuous development of the VOD data, it is expected that the use of such type of data can reduce the uncertainties in soil moisture retrieval [57, 58].

5. Conclusion

In this research, we produced 9 km, 8-day composited soil moisture data at the global scale from 2015 to 2020. The advanced VIPSTF-SW method was applied to produce the data, and ten 8-day composited image pairs (AP9-P36) from 14 April 2015 to 07 July 2015 were used as model inputs. The experimental results show that the predicted VIPSTF-SW9 data are more accurate than EP9 (the alternative published by NASA), based on the evaluation of the available AP9 data. Meanwhile, VIPSTF-SW9 can provide more detailed spatial information than EP9 as well as the original P36. Moreover, the predicted VIPSTF-SW9 data are also closer to the available in situ data than EP9 and P36 in terms of evaluation based on both spatial and temporal domains. The use of using multiple image pairs in the VIPSTF-SW method is a preferable choice than the single image pair scheme. Consequently, we conclude that the predicted VIPSTF-SW9 data are an accurate substitute for AP9 data for monitoring surface soil moisture after the failure of the SMAP active radar.

Data Availability

The SMAP data are available through NSIDC (https://nsidc.org/). The in situ data are available through ISMN (https://ismn.earth/en/). The produced VIPSTF-SW9 data are publicly available at doi:10.6084/m9.figshare.14634276.

Conflicts of Interest

The authors declare no conflicts of interest relevant to this study. There are no financial conflicts and interests to related to the manuscript.

Authors’ Contributions

H.Y. and Q.W. designed the research; H.Y. analyzed the data, wrote the original manuscript, and produced the dataset; Q.W., W.Z., X.T., and P.A. provided direction and comments on earlier drafts. All authors edited and approved the final manuscript; Q.W. provided the funding to support the research.

Acknowledgments

This research was supported by the National Natural Science Foundation of China under Grants 42171345 and 41971297, and Tongji University under Grant 02502350047. The authors would like to thank the NSIDC, Global Energy and Water Cycle Experiment (GEWEX), and European Space Agency (ESA) for making the SMAP and ISMN data freely available. The authors would also thank the US Department of Agriculture, Oklahoma State University, and the Oklahoma Climatological Survey for providing the data at Fort Cobb and Little Washita networks. Moreover, the authors thank Ms. Wenqi Liu from Oklahoma State University for collecting the related data.

References

  1. L. Liu, L. Gudmundsson, M. Hauser, D. Qin, S. Li, and S. I. Seneviratne, “Soil moisture dominates dryness stress on ecosystem production globally,” Nature Communications, vol. 11, no. 1, p. 4892, 2020. View at: Publisher Site | Google Scholar
  2. J. K. Green, S. I. Seneviratne, A. M. Berg et al., “Large influence of soil moisture on long-term terrestrial carbon uptake,” Nature, vol. 565, no. 7740, pp. 476–479, 2019. View at: Publisher Site | Google Scholar
  3. L. J. Renzullo, A. I. J. M. van Dijk, J. M. Perraud et al., “Continental satellite soil moisture data assimilation improves root-zone moisture analysis for water resources assessment,” Journal of Hydrology, vol. 519, article 27472762, pp. 2747–2762, 2014. View at: Publisher Site | Google Scholar
  4. A. Mishra, T. Vu, A. V. Veettil, and D. Entekhabi, “Drought monitoring with soil moisture active passive (SMAP) measurements,” Journal of Hydrology, vol. 552, pp. 620–632, 2017. View at: Publisher Site | Google Scholar
  5. A. G. S. S. Souza, A. R. Neto, and L. L. de Souza, “Soil moisture-based index for agricultural drought assessment: SMADI application in Pernambuco State-Brazil,” Remote Sensing of Environment, vol. 252, article 112124, 2021. View at: Publisher Site | Google Scholar
  6. V. Mishra, W. L. Ellenburg, R. E. Griffin et al., “An initial assessment of a SMAP soil moisture disaggregation scheme using TIR surface evaporation data over the continental United States,” International Journal of Applied Earth Observation and Geoinformation, vol. 68, pp. 92–104, 2018. View at: Publisher Site | Google Scholar
  7. R. Abolafia-Rosenzweig, A. M. Badger, E. E. Small, and B. Livneh, “A continental-scale soil evaporation dataset derived from Soil Moisture Active Passive satellite drying rates,” Scientific Data, vol. 7, no. 1, p. 406, 2020. View at: Publisher Site | Google Scholar
  8. Z. Tian, D. Kool, T. Ren, R. Horton, and J. L. Heitman, “Determining in-situ unsaturated soil hydraulic conductivity at a fine depth scale with heat pulse and water potential sensors,” Journal of Hydrology, vol. 564, pp. 802–810, 2018. View at: Publisher Site | Google Scholar
  9. W. Zhao, N. Sánchez, H. Lu, and A. Li, “A spatial downscaling approach for the SMAP passive surface soil moisture product using random forest regression,” Journal of Hydrology, vol. 563, pp. 1009–1024, 2018. View at: Publisher Site | Google Scholar
  10. H. E. Beck, M. Pan, D. G. Miralles et al., “Evaluation of 18 satellite- and model-based soil moisture products using in situ measurements from 826 sensors,” Hydrology and Earth System Sciences, vol. 25, no. 1, pp. 17–40, 2021. View at: Publisher Site | Google Scholar
  11. 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
  12. J. Peng, A. Loew, O. Merlin, and N. E. C. Verhoest, “A review of spatial downscaling of satellite remotely sensed soil moisture,” Reviews of Geophysics, vol. 55, no. 2, pp. 341–366, 2017. View at: Publisher Site | Google Scholar
  13. H. Ma, J. Zeng, N. Chen, X. Zhang, M. H. Cosh, and W. Wang, “Satellite surface soil moisture from SMAP, SMOS, AMSR2 and ESA CCI: a comprehensive assessment using global ground-based observations,” Remote Sensing of Environment, vol. 231, p. 111215, 2019. View at: Publisher Site | Google Scholar
  14. S. V. Kumar, P. A. Dirmeyer, C. D. Peters-Lidard, R. Bindlish, and J. Bolten, “Information theoretic evaluation of satellite soil moisture retrievals,” Remote Sensing of Environment, vol. 204, pp. 392–400, 2018. View at: Publisher Site | Google Scholar
  15. H. Kim, R. Parinussa, A. G. Konings et al., “Global-scale assessment and combination of SMAP with ASCAT (active) and AMSR2 (passive) soil moisture products,” Remote Sensing of Environment, vol. 204, article 260275, pp. 260–275, 2018. View at: Publisher Site | Google Scholar
  16. D. Entekhabi, E. G. Njoku, P. E. O'Neill et al., “The Soil Moisture Active Passive (SMAP) mission,” Proceedings of the IEEE, vol. 98, no. 5, pp. 704–716, 2010. View at: Publisher Site | Google Scholar
  17. S. K. Chan, R. Bindlish, P. E. O'Neill et al., “Assessment of the SMAP passive soil moisture product,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 8, pp. 4994–5007, 2016. View at: Publisher Site | Google Scholar
  18. H. Jiang, H. Shen, X. Li, C. Zeng, H. Liu, and F. Lei, “Extending the SMAP 9-km soil moisture product using a spatio-temporal fusion model,” Remote Sensing of Environment, vol. 231, article 111224, 2019. View at: Publisher Site | Google Scholar
  19. Y. H. Kerr, P. Waldteufel, J. P. Wigneron, J. M. Martinuzzi, J. Font, and M. Berger, “Soil moisture retrieval from space: the Soil Moisture and Ocean Salinity (SMOS) mission,” IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 8, pp. 1729–1735, 2001. View at: Publisher Site | Google Scholar
  20. E. Cho, C.-H. Su, D. Ryu, H. Kim, and M. Choi, “Does AMSR2 produce better soil moisture retrievals than AMSR-E over Australia?” Remote Sensing of Environment, vol. 188, pp. 95–105, 2017. View at: Publisher Site | Google Scholar
  21. E. G. Njoku, T. J. Jackson, V. Lakshmi, T. K. Chan, and S. V. Nghiem, “Soil moisture retrieval from AMSR-E,” IEEE Transactions on Geoscience and Remote Sensing, vol. 41, no. 2, pp. 215–229, 2003. View at: Publisher Site | Google Scholar
  22. W. A. Dorigo, A. Gruber, R. A. M. de Jeu et al., “Evaluation of the ESA CCI soil moisture product using ground-based observations,” Remote Sensing of Environment, vol. 162, pp. 380–395, 2015. View at: Publisher Site | Google Scholar
  23. Z. Bartalis, W. Wagner, V. Naeimi et al., “Initial soil moisture retrievals from the METOP-A Advanced Scatterometer (ASCAT),” Geophysical Research Letters, vol. 34, no. 20, 2007. View at: Publisher Site | Google Scholar
  24. Y. X. Y. Liu, Y. Zhou, N. Lu et al., “Comprehensive assessment of Fengyun-3 satellites derived soil moisture with in-situ measurements across the globe,” Journal of Hydrology, vol. 594, article 125949, 2021. View at: Publisher Site | Google Scholar
  25. T. J. Jackson, “III. Measuring surface soil moisture using passive microwave remote sensing,” Hydrological Processes, vol. 7, no. 2, pp. 139–152, 1993. View at: Publisher Site | Google Scholar
  26. S. K. Chan, R. Bindlish, P. O'Neill et al., “Development and assessment of the SMAP enhanced passive soil moisture Product,” Remote Sensing of Environment, vol. 204, pp. 931–941, 2018. View at: Publisher Site | Google Scholar
  27. N. N. Das, D. Entekhabi, R. S. Dunbar et al., “The SMAP and Copernicus Sentinel 1A/B microwave active-passive high resolution surface soil moisture product,” Remote Sensing of Environment, vol. 233, p. 111380, 2019. View at: Publisher Site | Google Scholar
  28. S. L. Hyunglok Kim, M. H. Cosh, V. Lakshmi, Y. Kwon, and G. W. McCarty, “Assessment and combination of SMAP and Sentinel-1A/B-derived soil moisture estimates with land surface model outputs in the Mid-Atlantic Coastal Plain, USA,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 2, pp. 991–1011, 2021. View at: Publisher Site | Google Scholar
  29. N. N. Das, D. Entekhabi, S. Kim, S. Yueh, and P. O'Neill, “Combining SMAP and Sentinel data for high-resolution Soil Moisture product,” in 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, China, 2016. View at: Publisher Site | Google Scholar
  30. F. Hu, Z. Wei, W. Zhang, D. Dorjee, and L. Meng, “A spatial downscaling method for SMAP soil moisture through visible and shortwave-infrared remote sensing data,” Journal of Hydrology, vol. 590, p. 125360, 2020. View at: Publisher Site | Google Scholar
  31. Z. Wei, Y. Meng, W. Zhang, J. Peng, and L. Meng, “Downscaling SMAP soil moisture estimation with gradient boosting decision tree regression over the Tibetan Plateau,” Remote Sensing of Environment, vol. 225, pp. 30–44, 2019. View at: Publisher Site | Google Scholar
  32. P. M. Atkinson, E. Pardo-Iguzquiza, and M. Chica-Olmo, “Downscaling cokriging for super-resolution mapping of continua in remotely sensed images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 2, pp. 573–580, 2008. View at: Publisher Site | Google Scholar
  33. Q. Wang, W. Shi, and P. M. Atkinson, “Area-to-point regression kriging for pan-sharpening,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 114, pp. 151–165, 2016. View at: Publisher Site | Google Scholar
  34. Y. Jin, Y. Ge, J. Wang, G. Heuvelink, and L. Wang, “Geographically weighted Area-to-Point regression kriging for spatial downscaling in remote sensing,” Remote Sensing, vol. 10, no. 4, p. 579, 2018. View at: Publisher Site | Google Scholar
  35. F. Wen, W. Zhao, Q. Wang, and N. Sanchez, “A value-consistent method for downscaling SMAP passive soil moisture with MODIS products using self-adaptive window,” IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 2, pp. 913–924, 2020. View at: Publisher Site | Google Scholar
  36. D. Long, L. Bai, L. Yan et al., “Generation of spatially complete and daily continuous surface soil moisture of high spatial resolution,” Remote Sensing of Environment, vol. 233, p. 111364, 2019. View at: Publisher Site | Google Scholar
  37. Y. Jin, Y. Ge, J. Wang, Y. Chen, G. B. M. Heuvelink, and P. M. Atkinson, “Downscaling AMSR-2 soil moisture data with geographically weighted area-to-area regression kriging,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 4, pp. 2362–2376, 2018. View at: Publisher Site | Google Scholar
  38. Y. Jin, Y. Ge, J. Wang, and G. B. M. Heuvelink, “Deriving temporally continuous soil moisture estimations at fine resolution by downscaling remotely sensed product,” International Journal of Applied Earth Observation and Geoinformation, vol. 68, pp. 8–19, 2018. View at: Publisher Site | Google Scholar
  39. R. Cao, Y. Chen, J. Chen, X. Zhu, and M. Shen, “Thick cloud removal in Landsat images based on autoregression of Landsat time- series data,” Remote Sensing of Environment, vol. 249, p. 112001, 2020. View at: Publisher Site | Google Scholar
  40. Q. Wang, Y. Tang, X. Tong, and P. M. Atkinson, “Virtual image pair-based spatio-temporal fusion,” Remote Sensing of Environment, vol. 249, p. 112009, 2020. View at: Publisher Site | Google Scholar
  41. T. W. Ford and S. M. Quiring, “Comparison of contemporary in situ, model, and satellite remote sensing soil moisture with a focus on drought monitoring,” Water Resources Research, vol. 55, no. 2, pp. 1565–1582, 2019. View at: Publisher Site | Google Scholar
  42. A. Colliander, T. J. Jackson, R. Bindlish et al., “Validation of SMAP surface soil moisture products with core validation sites,” Remote Sensing of Environment, vol. 191, pp. 215–231, 2017. View at: Publisher Site | Google Scholar
  43. T. Zhao, J. Shi, L. Lv et al., “Soil moisture experiment in the Luan River supporting new satellite mission opportunities,” Remote Sensing of Environment, vol. 240, p. 111680, 2020. View at: Publisher Site | Google Scholar
  44. E. C. Osenga, J. C. Arnott, K. A. Endsley, and J. W. Katzenberger, “Bioclimatic and soil moisture monitoring across elevation in a mountain watershed: opportunities for research and resource management,” Water Resources Research, vol. 55, no. 3, pp. 2493–2503, 2019. View at: Publisher Site | Google Scholar
  45. Z. Su, J. Wen, L. Dente et al., “The Tibetan Plateau observatory of plateau scale soil moisture and soil temperature (Tibet-Obs) for quantifying uncertainties in coarse resolution satellite and model products,” Hydrology and Earth System Sciences, vol. 15, no. 7, pp. 2303–2316, 2011. View at: Publisher Site | Google Scholar
  46. K. M. Larson, E. E. Small, E. D. Gutmann, A. L. Bilich, J. J. Braun, and V. U. Zavorotny, “Use of GPS receivers as a soil moisture network for water cycle studies,” Geophysical Research Letters, vol. 35, no. 24, 2008. View at: Publisher Site | Google Scholar
  47. F. Schlenz, J. T. dall'Amico, A. Loew, and W. Mauser, “Uncertainty assessment of the SMOS validation in the Upper Danube Catchment,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 5, pp. 1517–1529, 2012. View at: Publisher Site | Google Scholar
  48. S. Raju, A. Chanzy, J. P. Wigneron, J. C. Calvet, Y. Kerr, and L. Laguerre, “Soil moisture and temperature profile effects on microwave emission at low frequencies,” Remote Sensing of Environment, vol. 54, no. 2, pp. 85–97, 1995. View at: Publisher Site | Google Scholar
  49. M. J. Escorihuela, A. Chanzy, J. P. Wigneron, and Y. H. Kerr, “Effective soil moisture sampling depth of L-band radiometry: a case study,” Remote Sensing of Environment, vol. 114, no. 5, pp. 995–1001, 2010. View at: Publisher Site | Google Scholar
  50. J. R. Adams, H. McNairn, A. A. Berg, and C. Champagne, “Evaluation of near-surface soil moisture data from an AAFC monitoring network in Manitoba, Canada: implications for L-band satellite validation,” Journal of Hydrology, vol. 521, pp. 582–592, 2015. View at: Publisher Site | Google Scholar
  51. Q. Wang and P. M. Atkinson, “Spatio-temporal fusion for daily Sentinel-2 images,” Remote Sensing of Environment, vol. 204, pp. 31–42, 2018. View at: Publisher Site | Google Scholar
  52. G. Feng, J. Masek, M. Schwaller, and F. Hall, “On the blending of the Landsat and MODIS surface reflectance: predicting daily Landsat surface reflectance,” IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no. 8, pp. 2207–2218, 2006. View at: Publisher Site | Google Scholar
  53. T. Ranchin and L. Wald, “Fusion of high spatial and spectral resolution images: the ARSIS concept and its implementation,” Photogrammetric Engineering and Remote Sensing, vol. 66, no. 1, pp. 49–61, 2000. View at: Google Scholar
  54. Z. Wang and A. C. Bovik, “A universal image quality index,” IEEE Signal Processing Letters, vol. 9, no. 3, pp. 81–84, 2002. View at: Publisher Site | Google Scholar
  55. Q. Wang, W. Shi, P. M. Atkinson, and Y. Zhao, “Downscaling MODIS images with area-to-point regression kriging,” Remote Sensing of Environment, vol. 166, pp. 191–204, 2015. View at: Publisher Site | Google Scholar
  56. H. Yang, X. Zhang, M. Xu et al., “Hyper-temporal remote sensing data in bare soil period and terrain attributes for digital soil mapping in the Black soil regions of China,” Catena, vol. 184, p. 104259, 2020. View at: Publisher Site | Google Scholar
  57. E. Bousquet, A. Mialon, N. Rodriguez-Fernandez, C. Prigent, F. H. Wagner, and Y. H. Kerr, “Influence of surface water variations on VOD and biomass estimates from passive microwave sensors,” Remote Sensing of Environment, vol. 257, p. 112345, 2021. View at: Publisher Site | Google Scholar
  58. J.-P. Wigneron, X. Li, F. Frappart et al., “SMOS-IC data record of soil moisture and L-VOD: historical development, applications and perspectives,” Remote Sensing of Environment, vol. 254, p. 112238, 2021. View at: Publisher Site | Google Scholar

Copyright © 2022 Haoxuan Yang et al. Exclusive Licensee Aerospace Information Research Institute, Chinese Academy of Sciences. Distributed under a Creative Commons Attribution License (CC BY 4.0).

 PDF Download Citation Citation
Views484
Downloads235
Altmetric Score
Citations