Remote sensing instrumentation can be launched into space in a variety of orbital altitudes and inclinations; instruments can be flown on endo-atmospheric systems--aircraft, balloon, and remotely piloted aircraft; or they can be sited on the ground. The selection of a particular "system architecture" for a given mission typically involves many compromises and tradeoffs among both platforms and sensors. For imaging missions based on satellites, the most important factors in determining overall system architecture include the required geographical coverage, ground resolution, and sampling time-intervals. These affect platform altitude, numbers of platforms, and a host of sensor design parameters. Each remote sensing mission will have unique requirements for spatial, spectral, radiometric, and temporal resolution. A number of practical considerations also arise, including system development costs; the technical maturity of a particular design; and power, weight, volume, and data rate requirements.
Spectral resolution refers to the capability of a sensor to categorize electromagnetic signals by their wavelength. Radiometric resolution refers to the accuracy with which the intensities of these signals can be recorded. Finally, temporal resolution refers to the frequency with which remote sensed data are acquired. It is also possible to categorize the "coverage" of three of the instruments' four resolutions: spatial coverage is a function of sensor field of view; spectral coverage refers to the minimum and maximum wavelengths that can be sensed; and radiometric coverage refers to the range of intensities that can be categorized. The required measurement intervals vary widely with mission. For example, data on wind conditions might be required on time scales of minutes; data on crop growth might be needed on time scales of a week or more; and data on changes in land use are needed on time scales of a year or more.
Sensor design requires tradeoffs among the four "resolutions" because each can be improved only at the expense of another. Practical considerations also force tradeoffs; for example, on Landsat, multispectral and spatial data compete for on-board storage space and fixed bandwidth data communication channels to ground stations. For a given swath width, the required data rate is inversely proportional to the square of the spatial resolution and directly proportional to the number of spectral bands and the swath width. For example, improving the resolution of Landsat from 30 m to 5 m would raise the data rate by a factor of 36. Adding more bands to Landsat would also increase the required data rate. Changing the width of coverage can increase or decrease the required data rate proportional to the change in swath width. The baseline design for a proposed high-resolution imaging spectrometer (HIRIS) sensor would have 192 contiguous narrow spectral bands and a spatial resolution of 30m. To accommodate these requirements, designers chose to limit the ground coverage and thereby reduce the swath width of the sensor. HIRIS would have been used as a "targeting" instrument and would not acquire data continuously.
Spatial resolution drives the data rate because of its inverse square scaling. One way to reduce the data rate requirements without sacrificing spatial resolution is to reduce the field of view of the sensor. Designing multispectral sensors that allow ground controllers to select a limited subset of visible and infrared bands from a larger number of available bands is another option to lower data rates.
1 HIRIS was eliminated as an EOS instrument during the restructuring of EOS (see ch. 5: Global Change Research).
2 The different resolutions can be traded against ground coverage. For example, the French SPOT satellite offers 10 m resolution in black and white, but its ground swath width is 60 km versus Landsat 5's 185 km.
3 Data compression is another option to reduce data rates. A lossless compression would allow the full set of raw data to be recovered; reductions in data rates of approximately a factor of two might be gained implementing these algorithms. Most researchers prefer this to a data set that has been pre-processed in a way that destroys some data (but reduces data rate requirements) because "one person's noise can prove to be another person's signal."
SOURCE: 1983 Landsat Short Course, University of Calitornia Santa Barbara and Hughes SBRC; Office of Technology Assessment, 1993.