Riverside team members Eric Maddy and Adam Neiss are supporting NOAA’s Center for Satellite Applications and Research (STAR) to develop a proof of concept algorithm utilizing machine learning (ML) and artificial intelligence (AI) techniques applied to remote sensing, data assimilation, numerical weather prediction, and radiative transfer. The team leveraged the Multi-Instrument Inversion and Data Assimilation Preprocessing System (MIIDAPS) - a software package for remote sensing, data assimilation, preprocessing, and quality control - to develop MIIDAPS-AI, the artificial intelligence extension of MIIDAPS.
Currently, MIIDAPS processes passive microwave (PMW) as well as geostationary and polar orbiting infrared (IR) sounder and imager observations to provide the simultaneous inversion / of profiles of temperature, humidity, cloud, rain and ice, trace gases (CO, CO2, CH4, N2O), as well as the surface emissivity and skin temperature.
MIIDAPS-AI is applicable to the same sensors and applications and matches the performance of the traditional remote sensing algorithms. The main advantage of MIIDAPS-AI is the greatly improved efficiency in computing performance - excluding I/O and data preprocessing. The full-physics MIIDAPS-AI product generation takes just a few seconds compared to several hours for the 1DVAR physical MIIDAPS algorithm.
Eric Maddy began development of MIIDPAS-AI 10 months ago by training the AI to process the data produced by the Suomi-NPP ATMS sensor. In the last six months, with the assistance from Adam Neiss, MIIDAPS-AI has been extended to six additional sensors: Suomi-NPP CrIS, MetOp-A AMSU-A/MHS, MetOp-B AMSU-A/MHS, NOAA-18 AMSU-A/MHS, NOAA-20 ATMS, and GOES-16 ABI. MIIDAPS-AI runs daily and produces research-level products for each sensor. Below is an animation of GOES-16 ABI, ECMWF water and ice cloud top pressures, and ABI window channel radiances. MIIDAPS-AI products and more information can be found at: star.nesdis.noaa.gov/MIIDAPS-AI.