Implementing Snow surface roughness on SMRT

[peilinUW]

Hi all, I am working on some forward modelling on Ka-band active systems at VV channel at near-nadir angle on snow-covered ice. I attempted to add some macro-scale roughness information on the snow cover (snow-air interface) to represent some minor wind ripples, as shown in the photo here:

I interpreted that as a similar roughness level compared to Meloche et al. 2021 with RMSH 1.65cm and Correlation length 40cm, which let me use the GO model for this interface (errors emerged when I tried using GO backscatter model, I don't think that is compatible with the DORT solver; for IEM, ks is around 12 and kskl is in thousands), I also used the GO backscatter model, adding a water substrate under ice.

The returned backscatter at the 1.65 cm RMSH is overestimating at snow depth 10cm (13 dB simulated vs 4 dB observed). When I removed the snow surface roughness, the backscatter is more reasonable (3 dB simulated vs 4 dB observed). I tested with different roughness RMS with fixed correlation length (40cm), and here is what I simulated (and yes they are positive based on near-nadir return):
RMS = 1 mm: Backscatter = 40 dB
RMS = 1cm: Backscatter = 15 dB
RMS = 2cm: Backscatter = 13 dB

Documentation mentioned that the GO model will block most penetration (transmission) at the interface; is that the reason why GO is less reliable here?

Is there any other approach to represent the macro-scale roughness of the snow surface? Or does it actually matter to backscatter?

Thanks!

[JulienMeloche]

Hi peilinUW,

1. I think the rms value you are using is too high. The paper you are referencing was for soil, which is rougher. In my experience, rough snow surface is more in the mm range (Landy et al. 2019, DOI: 10.1109/TGRS.2018.2889763, Meloche et al. 2024, DOI:10.5194/egusphere-2024-1583). This will probably reduce the backscatter if you use iem... Lets say you use 5mm, it would put right on the edge ks < 3.
   Try surface = make_interface('iem_fung92_brogioni10', roughness_rms = 0.005, corr_length = 0.01)

2. The geometrical_optics_backscatter should not be use with DORT because its not a first-order scattering model. It was made to use with nadir_lrm_altimetry module... so better to stick with the geometrical_optics if you are using DORT. geometrical_optics has diffuse transmission but no coherent transmission. If you want to use geometrical_optics_backscatter, you could use the iterative_first_order.py solution instead of DORT. Be careful, it's a new implementation.

3. We are discussing here the micro-scale, or the roughness scale that interacts with the radar wavelength... For the macro-scale, there is nothing in SMRT yet.

Roughness is pretty tricky, and yes, you will be sensitive to it since you are working at nadir.

Hope this helps
Julien

[peilinUW]

Thanks, Julien!

[ghislainp]

Hi,

Is there a minus sign missing in all your dB values ? could you please edit the message for the future readers.

Both GO normal and GO backscatter can be used with DORT but have advantages and disavantages.

The normal GO implements the full GO, the reflectivity is bistatic. The transmittivity is also bistatic but it can be very peaked in the refraction direction. Because of this peaked nature, the RT solvers are struggling, certainly underestimating (probably a lot) the transmitted power. Even if well integrated over the peak, GO formulation does not conserve energy anyway.

This is why I implemented the GO_backscatter, which only consider the backscatter and does not compute the transmittance with GO. With moderate roughness, this approach should be fine. It is also significantly faster.

A more careful investigation is needed on this topic but the few tests I've done today (passive microwave) show that the normal geometrical_optics behave correctly. This is far from sufficient.

BTW, I've submitted a few changes, one is to check the validity of the geometrical optics. Nothing major.