Principal Investigator

Yuliya Marchetti
Email: yuliya.marchetti@jpl.nasa.gov

Description

A machine learning approach is developed to improve the bad pixel map that masks damaged or unusable pixels in the imaging spectrometers of the Orbiting Carbon Observatory-2 and -3.

OCO-2 and -3 spectrometers use nearly 500,000 pixels to measure three infrared spectral channels with high resolution. Spectra of reflected sunlight are used to retrieve CO2 concentration in Earth’s atmosphere with a precision of 0.25% as specified by the mission objectives. Anomalous performance of even one detector pixel can add significant error to the retrieved CO2 measurement; thus, bad pixels are removed from the data stream by the flight software. Identification of bad pixels is challenging due to the large data volume (both number of pixels and mission duration) and the diversity of the undesirable behaviors (e.g. instability, responsivity to temperatures, degradation due to radiation or thermal cycles).

Machine learning approach characterizes each pixel’s behavior through a collection of interpretable and statistically well-defined metrics and use these metrics to build a binary “bad-good” classifier. Likelihoods produced by the classifier compress the available information about pixel behavior and simplify interpretation and bad pixel detection.

Machine learning helps to greatly reduces time and effort, combines many sources of input data and incorporates diverse features of bad pixels, improves accuracy of the resulting mask and enables new insights into pixel behavior. The current machine learning algorithm for bad pixel detection is used operationally in calibration to assess the health of the detector and potential changes in the instrument.

Select Publications

Classification of anomalous pixels in the focal plane arrays of Orbiting Carbon Observatory-2 and-3 via machine learning. Marchetti, Y., Rosenberg, R. and Crisp, D.
Remote Sensing, 11(24), p.2901, 2019. [link]