EMISAR data description (as of 1997)

Storage medium

Data are stored on 8 mm Exabyte tapes.

Storage format

Data are stored in tar format with record length 10240 bytes. 

Contents 

Each tape holds:

• a read_me file
• source code of a utility program written in C-language (util.c)
• an executable of the utility program compiled for DOS PCs (util.exe)
• the polarimetric image in covariance matrix format
• the polarimetric image in scattering matrix format

  .

  
  Read_me file
  The read_me file is an ASCII file with information on the mapping geometry, the image size etc. An example is enclosed in Appendix A.

  .

  
  util.c program
  util.c is an ASCII file with the program source code. The program can be compiled with C and C++ compilers. It is menu driven, and once a SAR image is transferred to disk using the tar command, the program can:

• pick out a subimage
• perform type conversion including byte swap (DOS to UNIX or UNIX to DOS)
• detect the image (power, amplitude, phase)
• lowpass filter and resample
• perform a radiometric scaling and transformation

Appendix B illustrates the main menu and the sub-menus.

Scattering matrix format

The complex imagery includes four files (HH, HV, VH, and VV) of one-look, slant range, scattering matrix data. HV means vertical transmit polarisation and horizontal receive polarisation. Data are sampled at a 1.5 ¥ 1.5 m pixel spacing (slant range ¥ azimuth), but focused to a nominal resolution of 2 ¥ 2 m in order to suppress sidelobes. Data are motion compensated and calibrated.

The calibration includes a channel imbalance correction (phase and amplitude) and an absolute radiometric calibration. This implies that the radar brightness b ° of a homogeneous area equals 4p times the spatial average of the pixel intensity (I2 +Q2 ). The backscatter coefficient s ° is obtained as b ° multiplied by sine to the local incidence angle (taking the topography into account). The radar cross section of a point target is 4p times the total energy of the target as found with the integral method. Cross-talk elimination is normally not performed as the EMISAR sensor provides a cross-talk ratio better than -25 dB.

Data are stored as range lines with the imaginary part of a pixel, Q, following the real part, I. The two byte "short float" format used (four byte IEEE float format with truncated mantissa) is supported by the util.c type conversion.

Covariance matrix format
The covariance matrix is defined as:

where <x> is the expectation value of the stochastic quantity x, and x* is the complex conjugate of x. S is the scattering matrix, and S T is the transposed matrix. The scattering matrix is written as a column vector with only three elements as the two cross-polarised channels, S hv and S vh , are identical for all natural scatterers (reciprocity). The covariance matrix is easily seen to have symmetry properties: the three elements below the principal diagonal are the complex conjugate of the three elements above the diagonal.

When calculating the covariance matrix in practice, the two cross-polarised channels are averaged coherently, i.e. on a pixel by pixel basis the complex numbers are added and divided by two. Thereby the signal-to-noise ratio is improved. Further, the expectation value is replaced by some spatial average over a small area around each pixel, i.e. upon a pixel by pixel multiplication of one image by the complex conjugate of the another image the product image is low-pass filtered. Finally, due to the symmetry properties only six of the nine images are computed.

The covariance matrix imagery is first processed like the scattering matrix imagery, but subsequently it is subject to a post-processing consisting of the following sequence of operations:

• the two cross-polarised scattering matrix channels are coherently averaged
• the six product images are generated
• using a lowpass filter the six product images are spatially averaged and resampled at a 5 ¥ 5 m pixel spacing in ground grid.
• the radiometric calibration of the six lowpass filtered ground range images is changed from a b ° calibration to a s ° calibration. Each pixel is multiplied by 4p · sin q , where q is the local incidence angle.

The filter impulse response has a fixed length in the slant range domain implying that the number of equivalent looks (the speckle noise reduction) is range independent, while on the other hand the ground range resolution is coarser in near range than in far range. In order to ensure that a subsequent high quality resampling can by applied to the covariance matrix data, aliasing has been suppressed by using an effective filter bandwidth somewhat smaller than the output sampling frequency. The penalty is a slightly blurred appearance of the imagery.

The calibration implies that the backscatter coefficients s °hh , s °hv , and s °vv are obtained as the spatial average of the pixels in the hhhh, hvhv and vvvv images, respectively. The radar cross section of a point target is the sum of the pixel values over the target extent multiplied by 25 m2 .

To be exact, the terms ground range and s ° should be replaced by pseudo ground range and pseudo s °, respectively, because a flat earth with no topography is assumed in the post processing.

The covariance matrix data are stored as range lines (pixels in one line have the same azimuth position) in a 4 byte floating point format (byte swapped for direct PC usage) with each of the six channels in a separate file.

Appendix A