2.1

Optical Module

The Optical Module consists of nine identical pre-slit optical units –the polarization encoders– and a spectrometer. The interface between the two is an array of slits. The optical system, including several light paths in the plane of spectral dispersion, are shown in Figure 2. A physical representation of the module plus camera, mounted on a baseplate, is shown in Figure 4. Nine identical units with pre-slit polarization modulation optics (PMO) are responsible for encoding the linear degree of polarization of the incoming light. Light passes the polarization encoder optics first, ensuring that no instrument polarization corrupts the signal before polarimetric encoding occurs. Two spectrally modulated beams exiting each of the nine polarization encoders are focused in pairs, by a 4.4 mm diameter lenslet with a focal length of 10 mm, via a folding mirror onto a pair of slits of an array of eighteen identical slits. Each slit measures 0.2×1.4 mm². In the spectrometer, light is collimated via a spherical mirror, then folded via a flat mirror, and spectrally dispersed by a holographic grating. Light is subsequently re-focused by a set of a-spherical lenses onto a single CCD detector. The effective wavelength range of 400-760 is determined by the transmissive optics becoming opaque below 400nm, and by high order overlap from the grating at wavelengths above 760nm.

Figure 2.

Representation of the OM optical components and light paths entering via the polarization modulation optics. The slit array is oriented normal to the paper.

Encoding of the state of linear polarization by the PMO occurs by a combination of three optical elements: a quarter wave retarder, a (composite) multiple order retarder, and a polarizing beam-splitter [14], as shown schematically in Figure 3. Light passing through the MOR undergoes a wavelength-dependent phase shift, resulting in a change of incident linearly polarized light along the vertical into circular, elliptical or linear, as a function of wavelength. The PBS acts as a linear polarizer; the projection of the state of polarization onto the polarizer axis determines the intensity of the light passing through. The net effect is a sinusoidal modulation of the intensity spectrum:

Figure 3.

Schematic representation spectral modulation via the polarization modulation optics.

where I(λ) is the incident spectral radiance, λ is wavelength, δ is the retardance of the MOR, I₊(λ) and I_-(λ) are the two beams exiting the PMO. The retardance δ is ~21µm, resulting in a gradually increasing modulation period from 7nm at 400nm up to 28nm at 760nm. The modulation amplitude scales with the degree of linear polarization (DoLP), and the phase is equals angle of linear polarization (AoLP). Two beams I₊(λ) and I_-(λ) have orthogonal polarization, whereby the modulations are precisely in anti-phase. The function of the QWR is to change linearly polarized light with polarization axis along the optical axes of the MOR into circularly polarized light.

From Eq.(1) it is readily seen that the sum of the two modulation spectra I₊(λ) and I_-(λ) is modulation free and provides a direct determination of the spectral radiation field I(λ). DoLP and AoLP are derived by fitting the scaled modulation spectrum

with a fit function

over wavelength intervals one modulation period long. Equation (3) includes the polarimetric efficiency W(λ) to account for spectral smearing of the modulations by the slit and defocus. Note that in the ratio Eq.(2), the radiation field I(λ) cancels and that the polarimetric efficiency W and the retardance δ are the only instrumental parameters in Eqs.(1-2). Their (wavelength dependent) values are determined from polarization calibration measurements. One of the advantages of the spectral modulation method is that it provides a “snapshot” measurement of the spectral radiance and polarization with a single measurement. This feature is essential in obtaining high-accuracy polarization performance.

The instantaneous field of view in the along-track direction is determined by the slit width and the focal length of the small lenslet, and is 1°. In the cross-track direction, the slit provides a 7° field of view of which the central 6° is collected to avoid vignetting at the edges. In the focal plane the spatial resolution is 0.1° per pixel. We typically combine two spatial pixels so that the effective cross-track field of view is 0.2°. The effective ground pixel size is determined by the 1.75 sec integration time per image (see next section), the cruise altitude of the aircraft at 20km and the ground speed of 210m/s. The effective ground pixel measures 70×350 m² for the nadir viewport, and increases with viewing angle, up to 125×1120 m² for the aft- and foremost viewing angles at ±56°. The spectrometer optics are contained in a stiff monolithic aluminum structure, with the PMO contained in separate housings. An external baffle, part of the Optical Module housing (see Figure 4), blocks light from outside the nominal field of view. The OMA is mounted onto a baseplate that also holds the camera, the RMU, and the optical reference cube. The camera is mounted via an interface bracket that allows for alignment of the CCD sensor relative with the focal plane (Section 2.3).

Figure 4.

Optical Module mounted on the adjustable baseplate, together with Bobcat CCD camera, the reference measurement unit (RMU), and reference cube.

The camera is a 12 bit BOBCAT B2041M(onochrome), with a KAI-04070 interline transfer CCD from ON Semiconductor. The CCD has 2072×2072 pixels of 7.4µm, with a full well of 40.000e⁻, readout noise of 12e⁻ rms and dark current of 150 e⁻/s. A region of 1600×1800 pixels is collected each integration that contains the eighteen spectra for nine viewports, where 1800 column pixels sample the spectral domain over 380-820nm. The Reference Measurement Unit reads temperature sensors placed at seven locations in and around the Optical Module, and reads the output of a three-axis tip-tilt sensor (ADIS16448 from Analog Devices) which also provides a compass heading, air pressure, and humidity. The optical reference cube defines the instrument coordinate system.

2.2

Electronics

IPU and ICU combined take care of all electronic functions. All electrical connections from aircraft to SPEX airborne are made via the IPU. The IPU is connected to the Electrical Interface Panel of the aircraft, where it receives 28 Volt to power the instrument. It has a data connection to a control panel in the cockpit for the pilot to start or reboot the instrument and a connection to two LEDs that signal the instrument health status. The IPU serves as a feed-through ethernet connection for the ICU to the aircraft LAN. It also connects to a GPS antenna, mounted onto the exterior skin of the tail cone. An important function of the IPU is to deliver driver voltages for four controlled heaters and four uncontrolled survival heaters which are part of the thermal control system. The ICU is a ruggedized, conduction cooled Air Transport Rack with a VPX backplane. It holds a Single Board Computer with an Intel 7 core, running under Linux, a 128 Tb SSD storage, and an inhouse developed frame grabber. A software process written in Python runs on the OBC and handles all ICU tasks. The ICU as a whole is responsible for collecting camera data, flight navigation data and sensor data, for running the thermal control algorithm, monitoring the health status of the entire instrument, and running a Failure Isolation Detection and Recovery process. The frame grabber configures the camera at startup and reads camera images collected from the region of interest at a rate of 25.2 frames per second. Each camera frame has an exposure of 12.6msec, such that for the bright (cloudy) scenes, pixels are charged to 75% of the full well (to avoid nonlinearity). To achieve adequate signal to noise, the frame grabber co-adds 44 frames real time, which are storing to disk with 16bits. Effectively, images are sampled repeatedly over a 44/25.2=1.75 period.

2.3

Alignment Unit

The Optical Module must be aligned such that its nine viewing directions can have an unobstructed view through a slit opening in the tail cone. This is enabled by the Alignment Unit shown in Figure 5. It extends downwards from the SSU such that the OM is brought close to the slit opening. The AU enables angular alignment around three axes over a maximum range of ±3°. The three axes nominally coincide with roll, pitch and yaw axes of the aircraft. The AU also provides translational adjustment along the same three axes by ±5mm. The OM is mounted such that positive viewing angles are directed towards the flight direction, and the center viewport labeled “0°” nominally points to Nadir.

Figure 5.

Alignment Unit holding the Optical Module, camera and Reference Measurement Unit.

Angular adjustment around roll and yaw axes is achieved by a single ball-joint connection between the AU and the baseplate holding the Optical Module plus camera, and two support points which can be altered in length by set-screws. The baseplate is connected to an adjustable segment of the AU acting as a pivot table, rotating around the pitch axis, with the pitch angle determined by a set-screw. The alignment system is designed such, that it allows removal of the Optical Module on its baseplate (including camera, RMU and reference cube) and re-mounting without loss of alignment. Being able to remove the OM is beneficial for the purpose of instrument calibration, which requires only the Optical Module. Using the AU, possible misalignment with the aircraft reference can in principle be adjusted. In particular, compensation of a yaw-angle mismatch would be desirable to optimize projected swath overlap of fore and aft viewports. During the ACEPOL campaign, the only angular adjustments made was a fixed roll-offset of 0.8° to match the average orientation of the wing relative to the fuselage during the flight. With the plane fueled up, the wings flex “up” a bit, and relax during the flight as fuel is consumed. With the invoked offset, the center viewport is pointing close to nadir half-way during the flight, and also effects optimized viewing angle overlap of SPEX airborne and RSP (inside the opposite wing-pod).

6.1

Georeferenced SPEX airborne imagery

Clear sky conditions allow for validation of the geo-located data recorded with SPEX airborne by comparison with reference images. When the geolocation is properly performed, georeferenced imagery from SPEX airborne, represented on a longitude-latitude grid, should coincide with the reference images. Figure 9 shows an example from a segment of a flight on October 23, when the ER-2 passed over the San Francisco Bay. Two panels in the figure display radiance and polarization respectively, both rendered as RGB images, obtained by convolving spectral data with typical RGB response curves. Data are from the nadir viewport which has a projected swath of 2.5km. The reference image is taken from World Imagery service from ArcGIS. Clearly there is good overlap between SPEX airborne and the reference where in particular the coastline in the SPEX airborne imagery lines up well with the reference. The DoLP image provides a complementary view, with high contrast between ocean and land. During this particular segment of the flight, the ER-2 aircraft performed corrective manoeuvring, which shows up as wiggles in the projected data. While such angular motions are not favored for co-registration of multi-angle data because they result in reduced overlap between different viewing angles, they do provide a consistency check of the projection algorithms for large attitude variations of the aircraft. From comparisons between SPEX airborne imagery and World Imagery, performed for multiple tracks from multiple flights, we were able to determine the orientation of the SPEX airborne instrument relative to the ER-2, or, to be more specific, relative the ER-2 navigation data system reference. Along three axes defining ER-2 roll, pitch and yaw angular motions, we find offsets of 0±°0.2, +0.8°±°0.5 and 0°±°0.2 degrees respectively. The stated accuracies correspond with the IFOV pixel size. Due to the elongated pixel shape, SPEX airborne provides more detail along the swath direction than along track. A pitch offset causes shifts of the images in the along-track direction, while roll and yaw result in shifts along the swath. Therefore roll and yaw offsets are determined somewhat better than the pitch. Only flat terrain was used to determine the instrument alignment, using coastlines and large agriculture fields as matching features. On flat terrain, the effect of illumination in different viewing angles (e.g. shadows, hidden features for outermost viewing angles), which can be misinterpreted as spatial misalignments, is reduced. Color rendering has proven valuable for understanding SPEX airborne imagery and for matching features with those in reference images.

Another example of SPEX airborne imagery is shown in Figure 9, collected on October 27 2018 when a small controlled forest fire occurred in the Grand Canyon. The left panel shows a MODIS image of that day, overlaid with a color image from the AirHARP instrument, and with SPEX airborne placed on top. Both instruments capture the smoke plume and allow to pinpoint the fire which is directly below the brightest spot. Images of the two airborne instruments line up with MODIS and with respect to each other. Because the terrain is erratic, with large variations in height, comparisons like this provide a strong test for the quality with which the geolocation is performed. Note also that the smoke patterns, a transient feature in time, overlap from both polarimeter instruments. Even though the swath and spatial resolution of SPEX airborne are moderate compared to AirHARP, there is ample detail, giving the impression that SPEX airborne is rendered semi-transparently. In the DoLP image (right panel in Figure 4), the smoke appears dark-brownish; multiple scattering in the smoke depolarizes the light. Again, color rendering of DoLP provides complementary qualitative information. The brownish color of the smoke in the DoLP color image is distictly different from clouds, which appear as dark shades of gray-blue.

So far, the comparative analysis of SPEX airborne has been made by visual inspection. In the future it is planned to employ more advanced photogrammetry tools for automated image registration and image alignment.

Figure 8.

Segment of a flight track over San Francisco Bay. The area shown is 60km × 30km. SPEX airborne data from the nadir viewport is the narrow strip (2.5km wide swath), plotted over a reference image from World Imagery from ArcGIS, ESRI. Left image displays RGB radiance, right panel shows RGB polarization (DoLP).

Figure 9.

Flight track over Grand Canyon, capturing a smoke plume from a forest fire. The area shown is 80km × 80km. Left image displays RGB radiance, right panel shows RGB polarization (DoLP). The left image includes a pushbroom-color image from AirHARP, with the SPEX airborne strip overlaid. The background image from MODIS of the same day.

6.2

Data comparison methodology

For an inter-comparison of measurements, we must collect data which are coincident in the temporal, spatial and spectral domains, and take into account operational and conceptual differences between the two instruments. SPEX airborne can be considered as a set of nine separate pushbroom imaging spectrometers, each with a swath of 6° and an instantaneous field of view of 1°×0.2°. RSP is a scanner, collecting light from within a single 0.8° field of view which repeatedly sweeps along the flight direction (rotating around the pitch axis). On the ER-2, RSP has a useful angular range from -63.2° to +42.7°. With each scan, RSP intercepts eight out of nine of SPEX airborne’s viewports somewhere along the swath. RSP has overlapping wavelengths with SPEX airborne at 410, 460, 555, and 670 nm. Of these, the 410nm band partly lies outside of SPEX airborne’s effective wavelength range. When comparing instrument, RSP is adopted as the reference for SPEX airborne. With RSP being equipped with in-flight calibration, and having a track record of close to two decades of collecting aerosol and cloud data, its measurements are considered to be among the best to-date.

For an inter-comparison of measurements, we must collect data which are coincident in the temporal, spatial and spectral domains. A straightforward approach would be to collect measurements with coinciding geolocation, with matching longitude and latitude coordinates and matching viewing zenith- and azimuth angles, to within pre-set matching accuracy. For this to work however, pointing knowledge (directly impacting geolocation) must be right on the mark and that cannot be assumed to be the case. For SPEX airborne, knowledge of instrument to aircraft instrument orientation is obtained using rather rudimentary methods which are still in development. A complicating factor is that during the campaign, SPEX airborne was in the right wing-pod, while RSP was in the left. As a consequence the two instruments experienced opposite wing-flex. Level 1 data of the two instruments are obtained with slightly different models of wing-flex. Moreover, the wingflex models describe the average flex of the wings with time and deviations, order of 0.2° inferred from the tiltsensor, occur from track to track. The magnitude of the amount of wingflex may not even be the same for the two wings. Such considerations led to an approach for which as little assumptions are made as possible about the pointing of SPEX airborne: SPEX airborne data are correlated with RSP data in the time domain and best matches are selected. This was done taking several steps. First, RSP data were resampled at the temporal grid of SPEX. Next, SPEX airborne polarization and radiance spectra were folded with the spectral response curves of the four RSP bands to obtain spectrally averaged data at RSP wavelengths (for the 410nm wavelength band, the response curve was cut off at 400nm). Then, the root mean square (rms) difference between SPEX airborne and RSP data was determined per track, for each of SPEX airborne’s viewport-swath combination with each of RSP’s scan angles. At a given wavelength and viewing angle of SPEX airborne, this results in a two-by-two array of rms values, from which the optimum provides the best-matching RSP scan angle and best-matching angle along the swath of SPEX-airborne. The average of the best-matching RSP scan-angles over all tracks (of all flights) is then determined for each viewport of SPEX airborne. In the final step, the data sampling is repeated for all tracks, but using RSP scan-angles fixed at the average best-matching values, and optimizing only for the swath angle. In this last step a filter is included: the best matching SPEX airborne data should have a swath angle within half a degree from the value predicted by the wingflex model, otherwise it is discarded.

6.3

SPEX airborne – RSP comparison results

Here we discuss comparisons of polarimetric and radiometric measurements that were made for two types of scenes denoted as “ocean” and “cloud”. The “ocean” scenes include flight tracks over ocean and lakes under clear sky conditions, while “cloud” scenes denote low clouds over land and ocean (not cirrus). These scenes were selected to maximize the dynamic range of polarization and radiance. Ocean data include flights on two days with the ER-2 heading aligned with the principle plane, and the solar zenith angle near the Brewster angle, resulting in DoLP near unity for solar glint observations. An advantage of ocean scenes is that they are very homogeneous, having little variation of intensity and DoLP along the swath. Therefore errors due to pointing mismatches which are amplified by scene inhomogeneities, are reduced. Observations of (thick) clouds provide peak radiance levels at all wavelengths, and minimal DoLP for backscatter viewing geometries. Selected ocean and cloud scenes both comprise eight tracks from four flights. For these selected scenes, the inter-comparison of SPEX airborne with RSP is shown in Figure 10, where the DoLP of SPEX airborne (DoLP_SPEX) is plotted versus RSP (DoLP_RSP) in the upper two panels for ocean and cloud scenes, and the respective radiances, I_SPEX and I_RSP are in the lower two panels. Data from different wavelengths are offset successively by fixed amounts along the vertical axis. In each panel the level of agreement between the data is denoted by the root-mean-square values listed per wavelength band. For polarization data, this is the root-mean-square (rms) of all DoLP_SPEX – DoLP_RSP values, while for radiance it the rms is determined of the relative difference (DoLP_SPEX/DoLP_RSP −1) in percent. Data from different SPEX viewing angles are color coded according to the legend in the top left panel. Almost the complete dynamic range of DoLP, zero to unity, is captured in the dataset. Ocean data almost cover the entire dynamic range of DoLP, with the highest DoLP values obtained over the ocean in sun-glint near the Brewster angle, recorded with SPEX with the aft-most viewing angle at -56°. For 670nm, DoLP is nearly unity. At shorter wavelengths, the maximum DoLP at the Brewster angle is reduced by atmospheric scattering. Radiance is large in the glint at all wavelengths, but well below the radiances encountered in cloud scenes. In the glint, radiance can potentially saturate the instrument, but due to omnipresent winds during the campaign, the ocean surface was roughened, resulting in reduced reflectance. Lowest DoLP values were recorded over clouds, with values near 0.001, the detection threshold of SPEX airborne.

Figure 10.

SPEX airborne data versus RSP data at four RSP wavelength bands centered on 410, 470, 555 and 670nm. DoLP is in dimensionless units, radiance is in units Watt/m²/nm/sterad. Top two panels are data collected over ocean and lakes, bottom two panels are from cloud scenes. Data from all viewports are combined, with different viewing angles colored separately. Data from different wavelengths are shifted vertically, with zero offset for 410nm. Solid lines represent the identity relation. For each panel, the rms difference is listed in a boxed insets. For radiance data, the rms is computed of the relative difference (DoLP_SPEX-DoLP_RSP)/DoLP_RSP.

The level of agreement that we find between SPEX airborne and RSP DoLP measurements varies with wavelength. The best match is found for the 555nm band, which for both scenes is better than the DoLP uncertainty target adopted for SPEX airborne (0.005). For the ocean scenes, the agreement at 470nm is near the target, while 410nm and 670nm are somewhat out of range. For cloud scenes, agreement is excellent for at all wavelengths except 470nm. At all wavelengths, SPEX airborne DoLP measurements are systematically below RSP by what seems to be a wavelength dependent factor. This is evident most clearly in the 410nm and 470nm cloud measurements, where the slope in the data is below the solid line denoting the DoLP_SPEX = DoLP_RSP line. If this slope, which is ~0.96 for 410nm, for would be extended to DoLP=1, then SPEX airborne would end up 0.04 below RSP. In this case the deviation of the slope from unity is the prime cause for the relatively high rms difference. If points would be equally distributed along the slope over the recorded DoLP range 0-0.6, then the rms would be 0.05, accounting for most of the observed 0.085 rms difference. The other part is due to scatter, constituting point-to-point scatter and track-to track variations. Track to track variation can be inferred from the ocean data (top left panel) where data points are aligned on straight line segments above or below the solid black line. The intrinsic point-to-point scatter is mostly much less than the track-to-track variations. Data in Figure 10 are from eight combined viewports of SPEX airborne. Individual viewports provide similar agreement with RSP; rms values of DoLP_SPEX – DoLP_RSP per viewport are all very similar, and there is no dependence of the agreement on viewport angle. One exception is the viewing angle -56°, which has rms differences with RSP a factor of two to three larger than the mean RMS of all the viewports for all tracks. It is unclear what is the cause for a larger mismatch at the outermost viewing angle.

The agreement in radiance data in the lower two panels in Figure 9 varies from 1% to 5% rms, depending on wavelength and scene, which is considered to be excellent considering the fact that the radiometric calibration accuracy is estimated to be 4%. The adopted uncertainty for SPEX radiance measurements appears to be too conservative, and the measurements provide a better indication of the quality of radiance measurements.

At this stage, the cause(s) for the observed discrepancies in polarization measurements are under investigation and will be reported on in a future publication.