The Intercomparison of X-Band SAR Images from COSMO-SkyMed and TerraSAR-X Satellites: Case Studies

The analysis of experimental data collected by X-band SAR of COSMO-SkyMed (CSK®) and TerraSAR-X (TSX) images on the same surface types has shown significant differences in the signal level of the two sensors. In order to investigate the possibility of combining data from the two instruments, a study was carried out by comparing images collected with similar orbital and sensor parameters (e.g., incidence angle, polarization, look angle) at approximately the same date on two Italian agricultural test sites. Several homogenous agricultural fields within the observed area common to the two sensors were selected. Some forest plots have also been considered and used as a reference target). Direct comparisons were then performed between CSK and TSX images in different acquisition modes. The analysis carried out on the agricultural fields showed that, in general, the backscattering coefficient is higher in TSX Stripmap images with respect to CSK-Himage (about 3 dB), while CSK-Ping Pong data showed values lower than TSX of about 4.8 dB. Finally, a difference in backscattering of about 2.5 dB was pointed out between CSK-Himage and Ping-Pong images on agricultural fields. These results, achieved on bare soils, have also been compared with simulations performed by using the Advanced Integral Equation Model (AIEM).


Introduction
With the launch of COSMO-SkyMed (CSK) and TerraSAR-X (TSX) missions, a significant quantity of X-band backscattering data, useful in several hydrological applications, was made available to the scientific community [1][2][3].
The presence of these SAR satellites represents an excellent opportunity to monitor the parameters involved in the hydrological cycle, thanks to a very short revising time in convenient and various configurations of incidence angles and polarizations.
Some preliminary considerations on the sensitivity of X-band SAR to surface parameters have been mostly performed in the framework of the SIR-C/X-band experiment [4][5][6].Thanks to the COSMO-SkyMed Announcement of Opportunity funded by the Italian Space Agency (ASI), significant experimental studies for exploiting the capabilities of the X-band SAR mission in monitoring soil and vegetation parameters were recently carried out.Among them, the project ASI/1720 HydroCOSMO demonstrated a considerable sensitivity of X-band backscatter to soil, snow and vegetation features [7,8].Although this frequency is not the optimal one for the investigation into soil and vegetation cover, a reasonable sensitivity to soil moisture and vegetation biomass of agricultural crops has been observed [9][10][11][12].Analogous results have been obtained from the exploitation of TSX data, as it has been demonstrated in [13] and [14].
Complementary information can be derived from the data fusion of these two sensors, for a better scene understanding, which is very useful for all the techniques of change detection [15,16].However, absolute radiometric calibration, geometric differences due to conditions of acquisition and temporal decorrelation make the joint use of multi-modality images a very challenging task, especially in the case of multi-sensor satellite SAR images, for which relatively few works have been proposed until now [17,18].
The synergism between CSK and TSX missions and the interchangeability of SAR data coming from different X-band sensors is, therefore, an added contribution to scientific research, in general, and to hazard management and monitoring [19,20], which are applications where the revisit time is of vital importance.The opportunity of obtaining information about the seasonal variations of soil moisture, vegetation biomass and snow cover at the X-band is, in general, very important for hydrology, water management, climatology and natural hazards.However, if these satellites have different calibration, their datasets cannot be directly compared, hampering their combined use in inversion algorithms and, consequently, the retrieval of geophysical parameters.For this reason, a comparative analysis of these SAR sensor images is essential.This study aims to compare data from the two missions, and in this paper, some comparisons have been made between CSK and TSX images in different acquisition modes.The data processing was carried out by using standard calibration procedures provided by the space agencies, implemented by commercial software (i.e., SARSCAPE ©).The work, which was focused on image comparison, took into consideration some natural surfaces (i.e., forests and agricultural bare soils) used as reference targets.The choice of the images was then carried out so that the natural targets did not change significantly during the acquisition period.The final goal of this investigation was to provide useful information and some advice for the simultaneous use of SAR images acquired by different sensors.The comparison between TSX-Stripmap and CSK-Ping Pong was performed on an agricultural test site located in the watershed of the Scrivia River, in Northwest Italy.The other comparisons were performed on a test site located close to Florence in Tuscany (Central Italy).A further analysis of the data acquired by both TSX and CSK sensors was carried out by using the Advanced Integral Equation Model (AIEM) [21,22], in order to check the level of backscatter.The backscatter of bare soil was simulated for different fields, taking into account the values of soil moisture and surface roughness measured on the ground.Although the AIEM is not always able to fit the SAR signal correctly, this model is the most currently used for this type of analysis, also thanks to its extended validity limits.It should be noted that the results shown in this paper are based on a limited quantity of data and, consequently, require further data analysis and confirmation.

The Experimental Data and the Test Sites
A picture of the selected test areas is shown in Figure 1.The watershed of Scrivia is a flat alluvial plain of about 300 km 2 located near the confluence of the Scrivia and the Po rivers in Northwest Italy (central coordinates: 44.98°N, 8.88°E).It is characterized by large, homogeneous, agricultural fields of wheat, corn, sugar beet and potatoes and has been the test site for other SAR investigations [23].The Tuscany area, called Sesto, is a flat plain close to Florence of about 50 m (a.s.l.) (central coordinates: 43.81°N, 11.20°E).Forests are also present, mainly in the northern side of the agricultural test site features, and are mainly constituted by closed (greater than 40%) broad-leaved deciduous forests (oaks and hashes), with an average height greater than 5 m.A smaller part of the forest is represented by closed (greater than 40%) needle-leaved evergreen forest (black pine, white spruce), with an average height greater than 5 m [7,8].Most images were acquired in HH polarization (except for a few images acquired on the Tuscany test site) and the look direction was 'right' for the entire dataset.In Table 1, a summary of SAR images collected on these test areas is represented.The SAR data were required in 'single look complex' format both for CSK (Single-look Complex Slant product Balanced, SCS-B) and TSX (Single-look Slant Range Complex representation, SSC), and the imaging mode is 'Stripmap' for the two sensors.The 'Stripmap' mode represents the best compromise between spatial resolution (few meters) and the extent of the observed surface in a single acquired SAR frame (tens of kilometers).The other configurations have, on one hand, too low spatial resolution with respect to the size of the selected agricultural fields (i.e., Scansar) and, on the other hand, a small frame in terms of observed surface (i.e., Spotlight).The task of this paper required that SAR images be acquired, as much as possible, at the same time and with similar instrument and orbital parameters (i.e., incidence angle, polarization and spatial resolution).However, since CSK is a dual (civil and military) mission, the planning of the exact acquisition time on a test area is problematic.For this reason, the matching with TSX was extremely difficult and the common acquired dataset very slim.

Experimental Results
In order to compare the acquired data, the SAR images were geocoded and calibrated with a standard procedure, using slant range single-look complex data for both sensors, by means of a commercial software (SARscape ©) that implements the radiometric calibration according to the official documents of the corresponding space agencies [24][25][26].The acquired images have been processed by using the following usual procedures.Multi-look detected images were generated from single-look complex data by averaging the intensity in azimuth (10 looks for CSK and 5 looks for TSX) and range (5 for CSK and TSX) direction.The number of looks was chosen in order to filter the speckle and retrieve a square pixel in the multilooked image.The geocoding was performed to convert the position of the backscatter elements from SAR geometry to three-dimensional object coordinates by using a DEM (derived from the SRTM mission) and the satellite orbital parameters.The geocoded images have a pixel size of 10 × 10m 2 .In addition, layover and shadow effects in every acquired image were identified.Finally, the stack of images was generated and merged with the classification map of the observed area.The noise equivalent sigma zero (NESZ) was neglected, since the value is −19 dB for TSX and −22 dB for CSK, in the worst cases, as it can be observed in [24] and [25].A preliminary investigation of the characteristics of the images acquired by the two different sensors was carried out on a rather stable and homogenous target, such as a forest area, previously described in Section 2. The histogram of the mean σ° values is presented in Figure 4 for different satellites: CSK1, 2 and 3 and TSX1.In general, it can be observed that CSK Ping Pong (PP) σ° data show the lowest values (average: −13.17 dB, standard deviation, SD = 3.41 dB) and a very spread histogram, whereas TSX σ° in HH polarization shows the highest values (−8.11, −8.61 dB, SD ≈ 2.4 dB) and generally higher than the corresponding CSK Himage (HI) in HH polarization too (see Table 2).Although at X-band, σ°, on forest is not necessary as stable as σ° at the L-band, due to the wind and the presence and absence of leaves, the differences observed between the various sensors and, in particular, between CSK2-PP and the others, seem indeed rather high for a forest area.This fact suggested a more in-depth investigation of the performances of CSK and TSX satellites.

Conclusions
A cross comparison of CSK and TSX data taken on extended targets has been carried out to exploit the possibility of using combined data from the two sensor systems.For this comparison, several pairs of images from TSX and CSK with similar orbital parameters (in terms of date, time, incidence angle and polarization) have been selected.The comparison was carried out on flat agricultural areas only, in order to reduce the effect of the orography and on bare soils.The performed analysis has shown that CSK and TSX sensors produce different σ° values for the same surface types.More in detail, it has been observed that TSX Stripmap generally shows higher backscattering values than CSK Himage, with a mean difference of 3.15 dB (±0.90 dB).In turn, CSK-HI images show higher σ° (2.4 dB ± 0.02 dB) than the corresponding Ping Pong.It should be noted that CSK Ping Pong values are in some cases close to the noise level, which in general is about −22 dB.Model simulations, carried out by using the AIEM, resulted in good agreement with the TSX acquisitions, whereas the model was less able to reproduce CSK data, which can be simulated by forcing some of the surface roughness parameters beyond the range of ground measurements.
The results obtained from the performed comparisons lead to the following preliminary observations: The comparison between CSK/TSX presented in Section 3 and 3.3 did not show discrepancies attributed to evident target variations.The comparison between CSK Himage and Ping Pong, presented in Section 3.1, is considered a very stable target, due to the absence of rainfall events during the acquisitions and the presence of similar orbital parameters.The comparison between CSK-PP and TSX, presented in Section 3.2 (after that the necessary corrections for the different incidence angles have been applied), showed differences between the two sensors on bare soils that cannot be due to the observed target.
The most anomalous data seem to correspond to CSK-PP images, which showed very low values of about 4 dB compared to TSX data.
From the analysis of ground-truth data, they did not present features able to justify the differences in the measured σ°.
As a general consideration, we can suppose that the observed discrepancies can be attributed to different calibration procedures and calibration coefficients applied to the raw data of the various sensors.
We know that the relatively small SAR dataset used for this comparison, which is due to the extreme difficulty in obtaining simultaneous acquisitions (with similar orbital parameters) of the two satellites, hampers the generalization of the results.A higher number of images on different land types would therefore be necessary for achieving more universal results on a wider dynamic range of backscattering.Nevertheless, we find the obtained results to be a significant and useful guide to those who plan to combine X-band data from the two satellite systems for land applications.Taking into account TSX data as a reference, on bare soils, CSK data can be therefore corrected according to the differences found in this analysis (i.e., by adding about 4 dB in PP mode and about 2.5 dB in HI mode).This is maybe little too simplistic of a procedure, but the possibility of combining TSX and CSK data in order to shorten the revisit time of X-band SAR images is very important in various applications, such as disaster management.
Figure dates i

Table 2 .
Mean and SD of backscattering values collected from CSK and TSX in different configurations over a forest plot in Sesto area.
3.1.Comparison of COSMO-SkyMed Stripmap Mode: Himage (HI) and Ping Pong (PP)The first comparison was carried out between the CSK Himage (HI) and Ping Pong (PP) data (Image 4 and 5 in

Table 1 )
, in VV and VV/VH polarization, respectively.In this case, the Sesto test