Ok, this paper is a very cheeky one to include in this blog, because I think 90% of my contribution to it was looking confused and forcing the clever computational geniuses who were working on it to slow down and explain everything in shorter words. But I am still very excited by the work coming from Nick Wright as part of Nik Callow (yes, we are drowning in Nic/k/ck/ch type people around here)'s Water Smart Dams project, which is pushing the limits of what AI can do for automating detection of water in the small dams that are so crucial to Western Australian farmers.
One of the big challenges in observing small areas from remote sensing is having enough data, and some of the biggest bogey men that limit remotely sensed data availability are clouds and their shadows. Lots of people have thrown lots of time and effort into creating cloud-free products (think of those cloud-free LandSat Google Earth Engine mosaics we're all so fond of), but Nick wanted to go one better, and mask and classify types of cloud. And to do so he pulled out every trick in his little book of computational processing tricks.
I can't do justice to the novelty of the methods (see previous "can you use smaller words, please Nick?" comment), so you should read the paper https://www.sciencedirect.com/science/article/pii/S0034425724001330, but I am impressed by the results - substantial improvements in processing time, and small but meaningful improvements in performance against all benchmarked alternatives - and I will certainly be recommending CloudS2Mask to any colleagues needing a cloud masking approach for their remote sensing workflows.
Comments