When I moved from the Space Science and Astronomy (uplooking) side of NASA to the Earth Sciences (downlooking) I was dismayed to see how much time scientists spent processing and correcting the data from satellites before they could do their scientific investigations. For Landsat this processing primarily involves geolocation and radiometric correction.
Using currently available technology, it is possible to automate both of these processes. The relevant technologies are pattern recognition and artificial intelligence (AI) or machine learning.
Geolocation is the easiest procedure to automate. From the satellite orbital parameters it is possible to obtain good geolocation. Using pattern recognition it is rather simple to then compare the new data to previously acquired and geolocated data to geolocate the new data.
Radiometric correction is a bit more difficult. This process involves both atmospheric correction and correction for instrument peculiarities. Fortunately, as with geolocation, one does not have to start from scratch. One can use archived information in the form of ground truth or previously corrected satellite data. In the case of hyperspectral data, absolute radiometric correction is not needed as the shape of the spectral curve contains the information desired. AI can be used to build up the knowledge base that it required to autonomously correct the raw data. As more data is collected, both from Landsat and from complementary sources, the performance of the AI system will improve.
I realize that automating these processes will take a lot of work away from researchers, but I would hope that this is viewed as a positive rather than a negative development. There is always the need for skilled researchers to review the autonomously corrected data to assure that glitches do not occur in the system.
This page was last modified on 2 February 2015