andyr@sag4.ssl.berkeley.edu (Andy Rose) (03/20/91)
Data received from the Extra UltraViolet Explorer is in the form of time stamped photon events and a number of 'engineering parameters' or 'monitors'. There are some hundred monitors for each of the seven instruments. These monitors range in importance. Daylight/Nightlight, % coverage by South Atlantic Anomoly, total flux, and voltage are important. The nature of the instrument is that the distortion of the image (2D spectrum) is dependent on flux. So distortion is view dependent and can't be sampled out from initial (on the ground) calibration tests. Reduction of this data (lists of photon events) can be automated to some extent. Interaction with the data is required to understand how monitor values effect the instrument. As this knowledge grows, better reduction methods can be automated. The entire set cannot be reduced by hand (six months real-time events) and in fact can only be reduced automatically as compute power permits. (You see, if it takes 1/6 of real time to reduce the set, that's a month of crunching). Proposal is to create an interactive environment for 'acting' on the data given monitor conditions. Photon events are selected from the database given a window in the sky. Since these events are not collected at the same time, the satellite is in different states of 'health' for a given event. The idea is to build an accurate 'image' of that part of the sky. The detection of a photon (it's position) is dependent on the monitors, so an astronomer must be able to use the monitor data to 'build' an image. The interesting thing here is the time dependent nature of the data. An image will be made of events or groups of events which occured at different times. It could be night at the first pass and day the next, or the phase of the moon changes, adding more background photons. I would like to display an initial image based on default parameters and then allow the user to interact with some tools to fine tune the display. One tool would be a clock which could highlight those pixels representing photons collected at that time (or two clocks to set a range, or a color pallette which maps color to time collected for the field). This would allow a time based visual interpretation. I am concerned that the human mind will fail to see significant patterns if more than a few colors are used. Another tool which would be generic for many monitors is a 2D plot of value versus time. The user could map color to the events according to monitor value for that event. For example, a monitor such as reference voltage may indicate that the instrument behaved differently from one time to another. So the user selects a few values for voltage and maps colors like white for calibrated voltages, red for near unacceptable and black for way-out-there unreliable data. For a twist, new monitors could be created from old. Example is only display points when moon_monitor is "new" and reference_voltage is within acceptable range. More effects than coloring selected points must be supported. If a monitor's condition is known to shift data, those point's positions need to be adjusted by a factor, i.e. Add a*mon_5 to x position of points selected. It seems that multi-variate analysis of this sort could be useful for all sorts of data (census,...) and that frameworks such as AVS, apE, etc. can support this kind of interaction(?). I wonder if a 2D plotting widget with selectable thresholds for color assignment (my,my,these things are difficult to describe) exists. andy Scientific visualization in the global research community is virtually real.