We are thinking of running SLAMM using 1983 NWI data and 5 year time steps until 2005.
Would it be valid to compare these SLAMM results to past aerial photography/imagery?
We are looking to see if SLAMM will show tree islands (Swamp) in marshes converting to something else,
changes are visible in the photography. Our known problem is that our LiDAR-derived elevation data
is from 2007. Vegetation correlates highly with elevation, would it be sensible to turn
the preprocessor on in the above scenario? Thinking about the model's T0 has generated the question,
if the model start date is 1990 and the time step is 25 years, how is the first time step ending in year 2025?
Such a hindcasting effort would seem to be both valid and very interesting. Usually with a hindcasting effort you would need to use the pre-processor as you will never have LiDAR data for a historical NWI dataset. So really, you will be verifying the model when initialized with NED-quality data which would ostensibly have different results than if you initialized the model with (nonexistent) historical LiDAR data.
The model is not set up for hindcasting right now, though. The years are fixed so that simulations will culminate in the year 2100
QuoteI guess I could just have the software just step forward from the initial condition (NWI photo date) by the time-step but most users want to see multiple years of results that start on the same date. Therefore if they have a site with a photo date of 1986 and one with 2005 they still want to see results from 2025, 2050, 2075, 2100 for both sites. For that reason I implemented a bit of control:
25 year timesteps run with the timesteps shown above. 15 or 20 year timesteps start at 2020. Timesteps below 15 start at 2010...
However, fear not, I need to make the model compatible with "hindcasting" in the very near future and will release a patch for the model within April.
-- Jonathan
Colleagues have suggested trying the following to accomplish a hindcast. Our NWI map was created in 1983 - 1984 and our NED is from approximately 1999. We were wondering if we could accomplish the equivalent of a hindcast using the following technique: Instead of entering 1984 as our map date, we would enter 2010. Then we would subtract 26 (actually rounding to 25) years from each of the resulting outputs to view results for our period of interest. Thus, the model output for 2035 would actually represent the change in condition from 1985 through the year 2010. What do you think of this approach?
One problem that I see with this approach is that the model assumes historical SLR until 1990 and then applies the IPCC scenario selected (adjusted to local conditions based on the discrepancy between historical local SLR and historical eustatic SLR, OR based on the subsidence/uplift raster). So running from 2010 forward is not the same as running from 1985 to 2010 because of the SLR experienced...
However, I have completed the hindcasting capability. Actually the way the model works is the user can enter a text string with the years they wish to run (e.g. "1956, 1968, 1978, 1990, 2000") and the model then runs these time-steps and produces output.
I'll send that to you this afternoon by email. I just need to track down one range-check error (that may be the model running out of memory again). Oh, and this version allows the users computer to use 4GB if it's a 64 bit OS...
-- J