kvos
commented
1 week ago hi @phillipjws , nice that you got the slopes running with Planet data.
- you should select n_days as the most common tilestep, in your case 1 day. your Nyquist will be 2 days so you should pick that 14.8 days spring /neapfreq clearly.
- 250 timepoints is not ideal for estimating slopes. The algorithm was designed to run on Landsat and we use TimeSeries from 1999-date, so we usually get around 600 timepoints.
- that slope estimation did not work in your case as the curve doesn't have a minimum, in keeps decreasing with increasing slope values.
Anyways that's really good work and worth exploring more, would be great to have an idea of how to estimate slopes from Planet.
phillipjws
Hello, I've been working a lot with CoastSat.PlanetScope, working to modify it for 8 band imagery. I have trained a new model and have it working with 8 band imagery, I'm just curious as to the incorporation of CoastSat.Slope. I've gotten the tool working with my datasets, and I'm not too sure about what my settings should be, I know to keep my number of days above 7, although I'm still seeing flaky results.
It's mentioned in the example notebooks that beach slope can be found using the toolkit, and I'm curious if this meant using it with planetscope data, or using LandSat data and copying the results to complete tidal correction.
I think due to my use of 8 band imagery exclusively, I'm limiting myself to too little data for the slope toolkit to work effectively, as I end up with ~250 images between 2020 and now, with a significant peak in my timestep distribution towards 1 day.
Perhaps @kvos can weigh in here too, I think if I also used 4 band imagery (spanning back to July 2014) this may solve my problem.
Heres an example of some outputs from my specific use case:
I'm also in an area with rocky beaches, and so not too sure how accurate the slope calculations will be, going to compare with some in situ data (as I'm local to my site) and compare with both results from planetscope and landsat.
Thanks!