Methods

Calibration

At present, calibration coefficients provided by Andrew Heidinger (NOAA) under SCOPE-CM project are applied for all satellites.

Pygac comes with one default calibration file, which can be found in pygac/data/calibration.json. The current version is PATMOS-x, v2017r1, including provisional coefficients for MetOp-C. A record of all versions is kept in pygac.calibration.Calibrator.version_hashs. Alternatively, it is possible to pass custom coefficients to the reader, see pygac.reader.Reader. This could also be a previous default version.

The solar channel calibration (Channels 1 and 2, and Channel 3a if available) takes into account inter-satellite differences and is derived using amalgamation of different calibration references including the most recent MODIS Collection 6 data, in-situ targets, and simultaneous nadir observations. The detailed methodology is presented in Heidinger et al. (2010). The resulting (inter)calibration coefficients are of highest quality and follow the Global Climate Observing System (GCOS) standards for deriving fundamental climate data records.

The reflectances are normalized by a factor (as a function of day of a year) to account for changing Earth-Sun distance. However, it is left to the user to apply further normalization using cosine of solar zenith angle for example (depending on application in question).

The thermal channel intercalibration is done from scratch, starting from obtaining Platinum Resistance Thermometer (PRT), space and Internal Calibration Target (ICT, blackbody) counts. ICT temperatures are obtained from PRT counts based on coefficients provided in the POD and KLM Data User Guides (Kidwell, 2000). For each thermal channel, a smoothing window of 51 successive PRT, ICT and space counts is used to obtain robust gain values and to dampen undue high frequency fluctuations in the count data (Trishchenko, 2002). Section 7.1.2.5 of KLM User Guide presents the summary of equations implemented in Pygac to calibrate thermal channels, including non-linearity correction (Walton et al. 1998).

The PRT readings are supposed to be present in a specific order, for example, the reset value followed by four readings from the PRTs. However, this may not always be the case for orbits that contain data gaps or due to any other unexplained reason. If not taken into account properly, this irregularity could result in the underestimation of brightness temperatures, when calibration information is smoothed over many scanlines. In Pygac, this inconsistency is handled properly while calibrating thermal channels.

In some cases it was found that, apart from the reset values, even the readings from any one of the four PRTs could also have very low suspicious values. This could also seriously affect the computation of brightness temperatures. Pygac detects such anomalies and corrects them using interpolation of nearby valid PRT readings.

Geolocation

Each GAC row has total 409 pixels. But lat-lon values are provided for every eigth pixel starting from pixel 5 and ending at pixel 405. Using Numpy, Scipy and Pyresample packages, inter- and extrapolation is carried out to obtain geolocation for each pixel along the scanline.

If the GAC data belongs to POD family, then clock drift errors are used to adjust existing Lat-Lon information. Here, Pygac makes use of PyOrbital package. Pygac interpolates the clock offset and adjusts the nominal scan times to the actual scan times. Since the geolocation was computed using the nominal scan times, Pygac interpolates the latitudes and longitudes to the actual scan times using spherical linear interpolation, aka slerp. However, in the case of a clock drift error greater than the scan rate of the dataset, the latitude and longitude for each pixel of the scan lines that cannot have an interpolated geolocation (typically at the start or end of the dataset) are recomputed. This is done using pyorbital, which in turn uses TLEs to compute the position of the satellite at each scan time and the instrument geometry compute the longitude and latitude of each pixel of the dataset. Since this operation can be quite costly, the interpolation is preferred whenever possible.

Computation of Angles

The azimuth angles are calculated using get_alt_az and get_observer_look from pyorbital. The azimuth described in the link is measured as clockwise from North instead of counter-clockwise from South. Counter clockwise from south would be the standard for a right-handed orthogonal coordinate system. Pygac was updated to use the same definition for angles as pyorbital (2019, September, version > 1.1.0). Previous versions used azimuth +/-180 degrees, which correspond to degrees clockwise from south. All angles are converted to degrees. All azimuth angles are converted to range ]-180, 180] (2019 October version > 1.1.0 ). Note that ]-180, 180] is an open interval.

Correction of Satellite Location

Whenever possible, Pygac uses RPY corrections along with other orbital parameters to compute accurate satellite location (e.g. instead of assuming constant altitude). However, RPY corrections are not available for all NOAA satellites. In case of the majority of the POD family satellites, these corrections are set to zero.

Correction of Scanline Timestamps

The geolocation in Pygac depends on accurate scanline timestamps. However, these may be corrupt, especially for older sensors. Assuming a constant scanning rate, Pygac attempts to fix them using extrapolation based on the scan line number and a reference time.

Finding the right reference time is difficult due to the multitude of possible timestamp corruptions. But the combination of the following three options proved to be a robust reference in many situations: Timestamp of the first scanline, median time offset of all scanlines and header timestamp. See pygac.reader.Reader.correct_times_median() and pygac.reader.Reader.correct_times_thresh() for details.

Finally, not only timestamps but also scanline numbers may be corrupt. Therefor lines with erroneous scanline numbers are removed before extrapolation, see pygac.reader.Reader.correct_scan_line_numbers().

Scan-Motor-Issue

Between 2001 and 2004 GAC data from NOAA-14, NOAA-15, and NOAA-16 frequently contain a significant amount of noise towards an edge of the swath. As reported by Schlundt et al (2017), section 5.2, this is probably caused by a temporary scan-motor issue. Pygac tries to identify and mask affected pixels.