Implementing a spatially variable LAI?

[Utkarsh Mital]

Hello,

If I understand correctly, watershed workflow takes LAI from MODIS and then spatially averages the value for each land cover type, using an NLCD-MODIS cross-walk. Is it possible instead to project the MODIS LAI dataset on to the ATS grid (much like what is done for weather)?

Thanks,

Utkarsh

[Coon, Ethan]

Yes, certainly.

 

I’ve done some work on largish basins (HUC4s) to investigate how much of the variability in LAI is explained by land cover type vs by e.g. latitude or elevation, and found that within-class variation in LAI is much smaller than across-class variation.  Specifically, I first took the basin-average (spatially) LAI as a function of time, and computed an MSE compared to the “real” MODIS data.  Then I took a model of LAI where we accumulate time-series of each land-cover type and assign N-types time series by the land cover (default WW model).   The error in this model reduces the MSE by 90-99% (different in time).  So allowing within-class variability as a function of space really can’t buy you much more.

 

But if you look and find it is important for your domain, by all means, it’s doable.  My concern is that you may find some inconsistencies between LAI and land cover type.  Let’s say that the domain-averaged “grass” LAI is 2, and the domain-averaged “tree” LAI is 10.  If you just take LAI as a spatially interpolated field across all types, you will end up with some NLCD “grass” pixel (30m)  that have 10 LAI because they are in an area with a lot of trees that, at the MODIS (500m) scale, looks like trees.  So the downscaling of using NLCD land cover with MODIS LAI, to me, is better represented with a class-averaged time series.

 

You could also do more clever things like a latitude- or elevation-based corrections, by land class type (e.g. make LAI piecewise bilinear in x & y, or piecewise a linear function of elevation, where the constant term is the class-average time series and the linear terms are fits to data).  This would probably be the “most accurate” choice and would have the best of both worlds.

 

Ethan

 

--
You received this message because you are subscribed to the Google Groups "Amanzi-ATS Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ats-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ats-users/a5d3a82e-fe73-4c34-a334-af6b826f109cn%40googlegroups.com.

[Utkarsh Mital]

Thank you for sharing that observation from your past work, Ethan. It's quite helpful! Any chance that work is documented in a publication?