Recommendations for appropriate parameterization of Level 2 ImproPhe submodule

[HeatherRCox]

Hi all,
I'm trying to use S2A/ S2B to enhance the resolution of Landsat (8/9) data to 10m using the FORCE Level 2 ImproPhe submodule. Does anybody have recommendations as to what makes the most sense for the parameters KERNEL_SIZE (which deals with the prediction kernel) and KERNEL_TEXT (which deals with the kernel for the heterogeneity proxies)? The defaults are 2500 and 330m respectively, but the parameter file notes that these parameters should reflect the scale differences between the sensors. I went back to David's original paper on the ImproPhe algorithm, and I think these defaults might be appropriate for enhancing MODIS to Landsat resolution (approximately 100 and 11 Landsat pixels respectively, as mentioned in the paper?) The full-width at half-maximum (FWHM) of the line spread function in the scan direction for the red Landsat band is approximately 3 S2 pixels, so I am wondering if setting KERNEL_TEXT to 30m makes more sense for this application. But I see that others posting about this same application have used the default values. Given that this module is computationally intensive, I would rather avoid trying a bunch of different options. David, do you have a recommendation?

[davidfrantz]

Hi Heather,

sorry, this slipped through.

For the KERNEL_SIZE parameter, there is no physically-based recommendation. Usually, larger is better for the quality, but runtime is quadratic... Thus, a balance needs to be found. I would start with small kernels (e.g., 30), and then iteratively increase until results look usable (test with a very limited number of data).

The KERNEL_TEXT is best based on the scale ratio. Thus, 30 would make sense here.

Cheers,
David

[HeatherRCox]

Thank you David! I will try that.