Characterizing Error in the Verification Procedure of the ICESat-2 ATLAS
Instrument's Level-1B Product
Abstract
The Advanced Topographic Laser Altimeter System (ATLAS) is the sole
science instrument on NASA’s ICESat-2 mission. It was designed to
measure elevation simultaneously along six tracks on the earth’s surface
with centimeter-level vertical precision, demanding a picosecond-level
precision in photon time of flight. To ensure this precision requirement
was met, we developed for the Level-1B ATLAS data product a careful
verification procedure. To quantify the amount of acceptable error we
needed to understand the effects of the various floating-point
precisions at all steps of the calculations and the limitations of the
programming languages used for the production software (Fortran) and for
the verification software (Python). For example, we found in the 64-bit
photon time-of-flight calculations that differences even at 13 or 14
decimal places often revealed that an incorrect calibration value had
been selected. Without first characterizing the acceptable error, such
small errors could be overlooked and could propagate into critical
inaccuracies in the science products. We describe our methods and
lessons learned in order to inform future remote sensing verification
efforts.