A method for jointly estimating the noise and bias of AIRS and TES over homogeneous ocean scenes

Abstract

Accurate estimation of measurement noise in remote sensing instruments is critically important for the retrieval of geophysical quantities and the analysis of bias and trends. It is difficult to estimate noise directly from observed scene data because it is a combination of many sources, including instrument quiescent noise, scene inhomogeneity and random background fluctuations. Multiple datasets can be used to separate the instrument and scene noise. A noise estimate based on staring at cold space or a calibration source constitutes a lower limit, while noise estimates derived from the difference between scene observations and a model (such as forecast) convolves the true noise with the model uncertainty. Ideally, noise should be estimated directly from the observation of the scene. We have developed a Bayesian hierarchical model to jointly estimate the scene noise, instrument noise and instrument biases from sets of overlapping footprints. Informative prior distributions are constructed from pre-launch test results and inference is done by using Gibbs sampling to sample from the posterior distribution of the instrument parameters. We demonstrate this model by estimating and comparing the relative noise and bias of the Atmospheric InfraRed Sounder (AIRS) instrument on board the Aqua platform to the Tropospheric Emission Spectrometer (TES) aboard the Aura platform over the tropical latitudes using the Real-time, global, sea surface temperature (RTG-SST) analysis as a ground truth.