The geolocation of satellite imagery plays an important role in their applicability, as the horizontal accuracy has a series of implications, eventually affecting the overall reliability of the derived products. Additionally, the coregistration of different satellite image types (e.g. for data fusion), in particular between optical and SAR, is problematic, due the different nature of these products, which render the identification of corresponding pixels very difficult.
A reliable assessment of geolocation accuracy of satellite imagery typically necessitates the existence of a number of Ground Control Points (GCPs). Nevertheless, the recognition of GCPs on satellite imagery with spatial resolution of several meters is not a trivial task. It may be somewhat of an easier procedure on optical imagery than on SAR data, but, in practice, unless very high resolution (sub-meter) data are used, it is virtually impossible to reliably recognize any individual points on either image type.
In this context, the purpose of this study is to validate the geolocation accuracy of Sentinel-1 products, as they have an increased nominal geolocation accuracy, which provides major improvements and overcomes some of the aforementioned issues of previous SAR missions (Envisat, ERS), which are also assessed in this paper.
In order to overcome the hurdle of GCPs recognition, we employ an approach that attempts to recognize geometries - rather than specific points – in order to a) validate the geolocation accuracy and b) extract GCPs that would be useful either in georeferencing and/or as an intermediate step for the coregistration of SAR with optical images.
In particular, we use KGNSS (Kinematic Global Navigation Satellite Systems) measurements with GPS and GLONASS and a local adjustment of characteristic geometries (such as crossroads) strategy, in order to identify GCPs in a Geographical Information Systems (GIS) environment. An advantage of this approach is that a large number of GCPs can be extracted, without having to actually identify each and every point on the satellite imagery.
In the results, we primarily discuss the geolocation accuracy of Sentinel-1 when compared against Envisat/ASAR and ERS data. We also address the overall advantages and limitations of collecting KGNSS data, in order to use them as GCPs, taking into account the possibility of using Galileo GNSS satellites to further increase the accuracy of KGNSS data.
Finally, the prospects of automating the procedure for recognizing geometries and extracting GCPs and the implications on data fusion techniques at data (pixel or even sub-pixel) level are discussed.
This study is part of an experiment on geolocation accuracy and GCP extraction of Copernicus data also performed on optical images (Sentinel-2 and Landsat)over Central Macedonia, Greece (Perivolioti et al., also presented at this conference).

Area B - Board 158

The FY-3 satellite series is China’s second-generation polar-orbiting satellites, which consists of four satellites (FY-3A, FY-3B, FY-3C, and FY-3D), with about a two-year separation between two subsequent launches. A Microwave Radiation Imager (MWRI) was aboard this series of satellites which observes the Earth’s surface at five different microwave frequencies ranging from 10 to 89 GHz. Observations from the MWRI were further used to retrieve the land surface parameters such as soil moisture, vegetation water content, and land surface temperature. An official soil moisture (SM) product derived from the MWRI observations are being distributed by the National Satellite Meteorological Centre (NSMC) of China, which is available for all registered users.

As we know, SM products derived from passive satellite missions are playing a more and more important role in agricultural applications, especially in crops monitoring and disasters warning. It is crucial to evaluate their dependability of those products before they can be widely used at a large scale. Numerous studies have assessed the accuracy of the soil moisture products from SMOS, SMAP, AMSR-E, and AMSR2 by comparing the estimations against the ground measurements from monitoring networks around the world. However, to our knowledge, there was limited research focusing on evaluating the accuracy of the soil moisture product from FY-3 series satellites, particularly over agricultural areas.

In this study, we assessed the L2 SM product from the Chinese Fengyun-3C (FY-3C) radiometer against in-situ measurements collected from the Chinese Automatic Soil Moisture Observation Stations (CASMOS) within a 1-year period (January 1st to December 31st, 2016) in an agricultural province of China, Henan (Fig.1). As shown in Fig.1-a, in Henan nearly half of the regions are planted with crops, and the prevailing double-cropping system is winter wheat and summer maize. Due to agricultural irrigation, Henan suffers from serious water shortage and environmental problems relating to groundwater overexploitation. Therefore, strengthening soil moisture monitoring in the agriculture areas is of great significance for improving the water use efficiency in this province. Four statistical parameters were employed to evaluate the products’ reliability, including the mean difference (MD), the root mean squared error (RMSE), the unbiased root mean square error (ubRMSE), and the correlation coefficient (R).

We first examined the temporal performance of FY-3C L2 product during the 1-year period for each footprint in our study region. Figure 2 gives the spatial distribution difference between the FY-3C retrievals and in-situ measurements. In general, for most grid cells, their biases are negative (see green and yellow colors in Fig.2-a), approximately ranging from -0.1 to -0.03 m³/m³. In terms of RMSE and ubRMSE, they share a similar distribution pattern that most grid cells high values are located in the eastern part of Henan (Fig.2-b, Fig.2-c), which suggested that the FY-3C SM retrievals have better consistency with in-situ measurements in the western part than the eastern agricultural regions in Henan (see in Fig.1). Similarly, the grid cells with higher correlation coefficients (see red color in Fig.2-d) are also located in the western Henan. However, there are also some exceptional regions, for example, the southeast part of Henan, where they have a positive bias, larger RMSE, and ubRMSE, as well as higher correlation.

Next, we continued to evaluate the spatial performance of FY-3C SM product at different times of the year. As shown in Fig.3 the varying patterns of the four indicators along time in the year were generally different. The MD shows a ‘double-peak’ variation trend with the peaks around May and August (Fig.3-a). Except for the peaks, basically at all the dates, FY-3C SM retrievals show a negative bias compared to in-situ observations. The monthly mean bias across the year ranges from −0.11 m³/m³ to 0.08 m³/m³ with an averaged value of -0.04 m³/m³. This kind of underestimation has also been revealed by many previous validation studies (Choi and Hur, 2012, Jackson, Cosh, 2010, Wu, Liu, 2016b), for the soil moisture products of AMSR-E and AMSR2, which were derived using a similar retrieving algorithm as the FY-3C L2 product. Regarding the RMSE, they were roughly ranging from 0.08 m³/m³ to 0.13 m³/m³, with an average of 0.11 m³/m³, indicating the inconsistency between the two datasets. Seasons are one factor influencing the RMSE (Fig.3-b), but its variation behaviors are quite different from the MD and not easy to conclude. For most of the year, the RMSE tend to be higher than 0.1 m³/m³ except April, May and some dates in the autumn. Whereas, the daily and monthly averaged ubRMSE (Fig.3-c) show that spring (January-March), and winter (October-December) have the smallest average values (0.06 m³/m³), and the dates from May to August have the largest average ubRMSE (0.08-0.09 m³/m³), which may imply that the ground vegetation density is a key factor determining the FY-3C SM performance. Low correlations (R < 0.2) between FY-3C retrievals and in-situ measurements are found in the most time of the year, indicating the poor temporal consistency between the two compared data sets in our study area. The level of positive correlation coefficients (0.21) computed in our study is a bit lower than the reported results in previous satellite-derived SM products evaluation studies. For example, R = 0.24 in (Wu, Liu, 2016b), R = 0.31 in (Cho et al. , 2015) and R = 0.38 in (Kim et al. , 2015).

The increase in uncertainty in FY-3C soil moisture products with increasing vegetation is expected from theory and has been widely confirmed in our analysis of the soil moisture products. For instance, the temporal performances of FY-3C SM product for the footprints in western Henan showed better agreements than the footprints in eastern Henan, where is dominated by the agricultural regions. While the spatial performances at different times of the year have close relevance to the seasons; e.g., the daily mean bias between the two datasets shows a ‘double-peak’ variation pattern, tending to overestimate the soil moisture during May, August, and September and to underestimate the soil moisture during the rest of the year. This varying trend is in line with the cropping system in Henan, where the winter white and summer maize reach their vegetation density maximum in May and August, respectively. Herein, we used the NDVI collected from the MODIS satellites to represent the ground vegetation density. Figure 4 gives the temporal variations of the FY-3C SM and MODIS NDVI for the FY-3C footprint covering the CASMOS station O2073, which demonstrated that the two datasets shared a similar temporal varying trend and that the influence of vegetation could be captured by using a time series-based approach to soil moisture assessment (Wagner et al. , 1999).

Further, we calculated the temporal correlation coefficients (R) between the FY-3C L2 SM and MODIS NDVI for all the footprints in Henan. FY-3C SM is closely related to MODIS NDVI with averaged R of 0.55, which is consistent with the findings of several previous studies (Brocca et al. , 2013, Parinussa, Wang, 2014, Zhang, 2017). As in Henan province, especially in the agricultural regions, the cropping system dominates the NDVI variations over the year. The high correlations among the two products revealed that the cropping system has considerable influence on the soil moisture retrievals, which simultaneously could quantitatively explain the influence of the crops for the soil moisture products. To improve the accuracy of the FY-3C soil moisture product, an improved algorithm that could filter out the influences of the crops be applied in future studies.

References
Cho, E.; Moon, H.; Choi, M. First assessment of the advanced microwave scanning radiometer 2 (amsr2) soil moisture contents in northeast asia. Journal of the Meteorological Society of Japan. Ser. II 2015, 93, 117-129.
Brocca, L.; Tarpanelli, A.; Moramarco, T.; Melone, F.; Ratto, S.; Cauduro, M.; Ferraris, S.; Berni, N.; Ponziani, F.; Wagner, W. Soil moisture estimation in alpine catchments through modeling and satellite observations. Vadose Zone Journal 2013, 12.
Kim, S.; Liu, Y.Y.; Johnson, F.M.; Parinussa, R.M.; Sharma, A. A global comparison of alternate amsr2 soil moisture products: Why do they differ? Remote Sensing of Environment 2015, 161, 43-62.
Wagner, W.; Lemoine, G.; Rott, H. A method for estimating soil moisture from ers scatterometer and soil data. Remote sensing of environment 1999, 70, 191-207.
Wu, Q.; Liu, H.; Wang, L.; Deng, C. Evaluation of amsr2 soil moisture products over the contiguous united states using in situ data from the international soil moisture network. International Journal of Applied Earth Observation and Geoinformation 2016, 45, 187-199.
Zhang, X.; Zhang, T.; Zhou, P.; Shao, Y.; Gao, S. Validation analysis of smap and amsr2 soil moisture products over the united states using ground-based measurements. Remote Sensing 2017, 9, 104.

Area B - Board 159