During the 2nd SMOS workshop a one-day splinter session was held on "image
reconstruction" for the SMOS aperture synthesis instrument. A list of the
participants is available. The discussion was focused to the following topics
of currently highest priority:
1. Status of the SMOS simulator software.
2. Polarimetric requirements for proper retrieval.
3. Potential instrumental error sources (PIES) to be identified and corrected.
4. Realistic estimation of the strength and shape of the PIES.
5. Potential image reconstruction methods (PIRM) and algorithms suited for SMOS.
6. Estimation of the achievable accuracy of the PIRM.
7. Definition of suited test scenes for the PIRM assessments.
8. Potential calibration and validation sites suited for SMOS.
9. Formulation of mandatory actions for ESA to be carried out next and soon.
The discussion content and outcome of the single topics is briefly reviewed in the following subsections. The action items identified are highlighted in bold letters, numbered, and furnished with a deadline. Concerning documents to be exchanged or provided, it was clarified during the drafting of the minutes, that all the documents have to be officially finalized and accepted by the originator, and that it is legal to use them according to related contracts.
1. The SMOS simulator software tool shall be used for the prediction of the overall SMOS performance and for the determination of the key design parameters. The actual status of the SMOS simulator software developed by UPC is more or less unknown to the participating splinter experts. Especially the comprehensive realization of realistic PIESs, the detailed modelling of the upwelling brightness temperature to be received by the single antennas of the instrument, and the extent and types of PIRMs currently implemented have to be identified to determine potential lacks to be addressed next. (1) For this, UPC and ESA agreed to distribute a software description and a software requirements document to the SAG (31 Jan 01). Furthermore it was criticized that a detailed knowledge about the existing simulator software is only available through more or less one person. This situation has to be changed as soon as possible although it was not discussed here how to arrange that practically.
2. The actually intended operation of SMOS is to receive a horizontal and vertical linear polarisation at the antenna terminals. Due to the basically inherent variation of the incidence angle within the FOV and the additonally tilt angle of the instruments main axis relative to the nadir direction, the polarisation bases are changing for each image pixel. This results in a spatially varying mixture of received polarisations even in the case of a single polarisation receiving antenna. It was noted by referencing to an earlier paper that the only measurement of two independent polarisations is not sufficient to retrieve the pure polarisations by mathematical means. It would be necessary to have a fully polarimetric instrument to achieve that goal. A document (Pol-Switching: Switching Scheme for Single Channel Polarimetric Aperture Synthesis Radiometers) was distributed to the group which explains a technique to make MIRAS working fully polarimetric without additional hardware, except for an increase in data rate and a degradation in effective sensitivity (TBC). (2) UPC agreed to check how this problem is addressed by the current simulator version. ESA noted that this problem is treated within a current ESA study and that results will be made available for SMOS (Both informations shall be provided by the end of the year and latest for the next SAG meeting). It was furthermore noted that this polarimetric correction has possibly to be incorporated within the image reconstruction procedure and not only be applied to the final brightness temperature maps.
3. Generally it was agreed to summarize the PIES in a common equation modelling the instrument as realistic as possible. For this purpose a first draft equation was discussed. The following key suggestions have been made: The brightness temperatures at the antenna terminals, the antenna patterns, and the fringe wash functions should be considered in a matrix representation to incorporate properly the polarimetric character of the upwelling radiation. The modelled output of the correlator shall be called a correlation sample instead of a visibility sample to discriminate between uncorrected and corrected values. The incorporation of the calibration switch mismatch is of big importance because this component is not considered by the noise injection procedure for error correction and calibration. (3) UPC agreed to provide their equations used in the current simulator version together with a detailed description of the quantities to the SAG (31 Jan 01). Based on that and the suggestions of this splinter group the final equation shall be treated as the general modelling equation for the SMOS system and used for the performance assessment, the system design, and a general image reconstruction starting point. The equation shall be well-described in detail and validated by accompanying measurements provided by the developed subsystems like antennas, LICEF receivers, or complete systems like the anticipated HUT airborne aperture synthesis radiometer. (4) ESA agreed to provide an actual block diagram of the overall system to the SAG to assist the equation formulation adequately. This document should be redistributed whenever major technical changes are made (31 Jan 01). The current duration for a simulation using the existing UPC simulator is in the order of one day, which is too long for a working tool. It was mentioned that the current UPC simulator was optimised for accuracy rather than speed, but can be improved for speed in future versions.
4. The proper and realistic simulation of a real system requires obviously realistic estimates of the PIES as well. Especially the accurate polarimetric and two-dimensional measurement of the single antenna patterns down to required levels of -30dB and less and considering some mutual coupling is a challenging work. Such comprehensive measurements on existing related antennas from the former MIRAS project or the LICEF developments are currently not existing. A draft plan promises measurements for a single antenna for December 2000, for a four-elements subarray for March 2001, and the fully integrated subarray together with the receivers for summer 2001. Furthermore it was discussed to add a dummy element on each arm of the Y to have more symmetry in the mutual coupling between the elements. The addition of a total power radiometer to each arm was also suggested to assist the antenna pointing estimation and to make the calibration redundant. The tilt error for the single arms of the Y is guaranteed by industry to be less than 5mm maximum and to have a resonance frequency of 3.5Hz. These values are uncritical for the SMOS performance so that there is no need for further assisting measurement equipment. Currently the fringe wash functions of the receivers can be estimated to vary within 0.2-0.6% in amplitude and a few degrees in phase, which has to be confirmed by measurements on the LICEF receivers. Additionally a sufficient out-of-band rejection of the filters is of big importance to protect the SMOS receivers from interference. This requirements together with steep transitions to the in-band area are technically difficult to achieve. (5) MIER agreed to provide ESA with the measurements of the filters to be used and ESA agreed to provide such measurements to the SAG with a list of hardware specifications delivered to industry (ASAP). Data on the correlator performance have not been discussed.
5. The following proposed naming of the basic image reconstruction (assisting) methods have been agreed by the splinter group: the "direct method", the "closure phases and closure amplitudes method", the "point source response method", the "CLEAN method", and the "maximum entropy method". Briefly, the first one uses measurements on well-known scenes to determine a matrix based transfer function to inversely apply to the actual measurements. It is also known as G-matrix approach and was used by the American ESTAR system. The intended noise injection procedure for error correction and calibration belongs as well to the direct methods. The second method allows to correct for amplitude and phase imbalances of the individual receivers using the redundancy of repeated baselines. It does not provide a brightness temperature map but corrects the measured correlations to the visibilities to be used for producing the temperature map. The third method can be used as well to correct the correlations by the imaging of a point source. Using multiple source positions a set of correction equations can be found from comparisons between the ideal and the system-model-based point source response. It was again mentioned that it is more or less impossible to have an artificial point source, e.g. a transmitter at the pole caps, within this protected frequency band. For that reason the use of the moon or the sun was considered as a potential natural point source of sufficient background contrast in combination with a repeated tilt of the instrument into space from time to time. Here an additional problem could be the cosmic variance of the background in the order of around 3-8 Kelvin. The known brightness temperatures of the moon or the sun can be used for additional calibration purposes. To overcome the transmit prohibition within the protected band the use of a sufficient power transmitter in the out-of-band area of the receivers and outside the protected band was mentioned. The fourth and fifth image reconstruction methods have not been discussed further because their major use is for the case of gaps in the visibility function which do primarily not occur. These methods are useful for the unintended loss of receivers or correlators due to damage or malfunction.
6. The splinter group agreed that there is a potential lack in the knowledge of the accuracy of the image reconstruction process. A basic starting equation to start an iterative error correction algorithm was discussed. This equation basically requires the accurate modelling of the instrumental errors intended to be provided by the PIES equation of section 3. For one numerical run this algorithm allows a worst case estimate of its accuracy but for a standard deviation estimate multiple time consuming runs of many hours for a proper statistic are necessary. It was noted that it is mandatory to verify this required basic equation by experimental means on the developed hardware to achieve highest accuracy and reliability. (6) In this context ESA agreed to provide the SAG as soon as possible with numbers of all relevant hardware measurements (ASAP). Generally it was highlighted that each algorithm or combinations of them should produce the same reconstruction result accuracy when provided with the same input. Furthermore it was mentioned to check the use of AIPS++, which is a potential image reconstruction tool for aperture synthesis in radio astronomy. (7) UPC agreed to provide the SAG with the gaps of the actual simulator software concerning the image reconstruction. Finally UPC was asked for their handling with Gibb's phenomenon in the simulator arising from Fourier decomposition as used in the basic Fourier inverse transform of the visibilities (31 JAN 01). As potential references to the aperture synthesis theory the following literature was recommended: The thesis of A. Camps of UPC, a calibration definition document of ESA, the LICEF specification document, and various other ESA documents on aperture synthesis. (8) ESA agreed to provide a detailed list (ASAP).
7. For testing the overall system performance and especially the image reconstruction theoretically, the splinter group agreed that it is necessary to have a well-defined set of test scenes, i.e. polarimetric brightness temperature maps as seen by the antenna terminals. It was noted that both quasi-realistic as well as artificial scenes should be defined and generated. For the quasi-realistic case the help of the land and ocean groups is required to produce typical soil moisture and ocean salinity influenced brightness temperature maps. Those scenes should be pure ocean areas, coastal areas, and pure land areas which itself should contain forests, lower vegetation, bare soils, and high contrast areas like lakes to incorporate as well the mixed pixel situation. (9) It was agreed that the SAG will define the number, size, resolution and anticipated content of the scenes to provide potential scene generation experts of the land and ocean group with those numbers (ASAP). During the discussion the problem of resampling a given data point grid arised which can have serious error sources as mentioned by some specialists of the splinter group. It was agreed to perform data comparison for validation and assessment purposes on a hexagonal grid, which is the primary output format due to the SMOS Y-shaped array configuration.
8. Only brief reflections on the calibration validation have been finally done. The moon and the sun have been taken into closer consideration as already mentioned in section 5. Furthermore it is intended to use the well-investigated and operationally validated ground truth areas of some watersheds in Europe and the US, which have a size of more than one SMOS ground resolution cell.
9. Following the previous discussions the splinter group finally identified and recommend the mandatory actions to be undertaken as soon as possible by ESA: