Potential Land Applications of Geostationary Spectrometers as Demonstrated with TEMPO

TEMPO is the first sensor in geostationary orbit with hourly scans over much of North America and spectral coverage from the UV (~300 nm) through the NIR (~740nm) with a small gap (500-540nm) at a spectral resolution of ~0.6nm. As such, it provides a unique opportunity to investigate potential land applications of such sensors. We have used machine learning with TEMPO radiances trained on MODIS data to show that the spectral properties of the land surface can be separated from those of the atmosphere (gases, aerosol, and clouds) even up to fairly high values of cloudiness. This allows for quantitative estimates of gross primary production (GPP) in near-real-time as well as qualitative vegetation indices. With multi-spectral data, such as from MODIS, typically processing is done only in clear skies and with a 16-day temporal window such that similar data is not available for more than a week after overpass. The coverage available from GEO is also vastly superior to that from low Earth orbit instruments. Another benefit of such machine learning algorithms is noise reduction. We will show that our approaches can also be applied to atmospheric products to reduce noise using only the spectral content of the data as opposed to spatial approaches. Although the spatial resolution of TEMPO is not as high as dedicated imagers, it may be possible to combine with data from GEO and/or LEO imagers to perform spatial downscaling. Such information may be useful for precision agriculture, forestry, and hazard monitoring. Similar spectral coverage will be achieved by planned the planned GeoXO ACX sensor as well as GEO ocean sensors such as NASA’s GLIMR and the NOAA GeoXO OCX but at spatial resolutions of about two orders of magnitude higher and with regular coverage of coastal and near-lake areas and capabilities to point to other locations.