of virtually every instrument are necessary to empirically determine the magnitude of these climatically large and unpredictable intersatellite biases. The overlapping observations should last at least 1 year as some biases depend on season.
The present two-orbit configuration (nominal northbound equatorial crossing times of 1330 and 1930) is manageable for bias calculations for certain climate data sets. Overlapping observations from a 1330 spacecraft may be used to determine the bias of the 1930 instrument as long as their respective biases relative to the diurnal component are known. This is possible for quantities such as stratospheric temperature, which is fairly coherent in space and time. However, quantities with substantial diurnal variations, such as cloudiness or surface properties, will require overlapping observations between old and new spacecraft in the same orbit node. (The shift from 0730/1930 to 0530/1730 in the AM node, as has been proposed for the National Polar-orbiting Operational Environmental Satellite System (NPOESS), will probably cause some discontinuity. Additionally, some instruments will fly on only one node, again requiring overlapping observations of the old and new spacecraft in each node.) The following two examples demonstrate the critical value of overlapping observations.
A time series of total solar irradiance (TSI) solar variations was pieced together by Willson (1997) using data from two consecutively flown Active Cavity Radiometer Irradiance Monitor instruments (ACRIM I and II). A gap of about 2 years in the ACRIM data was bridged by using mutually overlapping observations from two other TSI instruments, the Nimbus-7 Earth Radiation Budget (ERB) and the Earth Radiation Budget Satellite (ERBS). The results from the ERB and ERBS were less precise than those from ACRIM I and II and the absolute irradiance values differed by 0.5 percent (Figure 3.1). However, the extent of the overlap in the data made possible a sufficient reduction of noise in the bias calculations, allowing for the production of a time series with a trend of 0.032 (±0.0009) percent per decade.
Christy et al. (1998, 2000) utilized the MSUs on National Oceanic and Atmospheric Administration (NOAA) polar orbiters in both 1330 and 1930 orbits, alternating between the two nodes, to merge data from nine satellites into a time series of daily global atmospheric temperatures. The duration of most overlapping periods was 1 to 3 years; however, two periods were less than 7 months long. Biases, which could be as large as 0.5 K, were determined to be globally accurate to a precision of 0.01 to 0.02 K (as estimated through a variety of subsampling experiments) only because data from overlapping observations were available.
Flexible strategies for providing overlapping data should be investigated because data from the older spacecraft are not required in real time for climate purposes. Thus, information from instruments for which overlapping observations are necessary may be stored on board for downloading at more convenient times. Additionally, though an overlap period of 1 year is a requirement, the sampling during this year may be less than continuous, being whatever is sufficient to characterize the annual cycle of the bias. Most instruments will need overlapping observational periods for each orbit node (e.g., from an old 0530 to a new 0530 spacecraft).
The past generation of polar-orbiting satellites was injected into orbits that included slow longitudinal (east-west) drifting relative to local equatorial crossing time (LECT), to prevent the spacecraft from approaching local solar noon. The rate of the drift was up to 15 degrees per year in some cases, representing up to an hour drift per year in terms of LECT. Two consequences of this drift conspire to corrupt the observations from the standpoint of their stability.
Because these spacecraft drifted to earlier or later LECTs, observations were influenced by the local diurnal cycle (e.g., afternoon temperatures are warmer than morning temperatures) in the quantity measured. Trend calculations are thus skewed by the local time at which observations are taken and may be misinterpreted as a trend in the absolute instrument bias. This is particularly important for most surface quantities, atmospheric temperature, and cloud observations, where diurnal signals and solar angle changes, which might affect the upwelling radiance, are substantial relative to decadal trends.
One method of dealing with this problem was developed by Waliser and Zhou (1997) for outgoing longwave radiation (OLR) and highly reflective cloud (HRC) data sets. Because the spatial variation of the changes in OLR and HRC is substantial as a satellite drifts through the diurnal cycle, Waliser and Zhou based their corrections on