I'm checking to see if Jonathan or anyone else has found a high degree of sensitivity to the time step chosen for modeling. I have been using 25 years as the model default but I ran a 5 year time step and came away with significantly different results especially regarding transitional marsh and salt marsh categories. I'm modeling a fairly small region (135K hectares) with a small tidal range (2 ft) and I'm curious as to the best time step.
QuoteI believe you had mentioned in our June phone call that you hadn't found much difference between a 25 year and a 5 or 10 year time step, so I had been modeling in 25 years for convenience. However I just tested 5 year and found significant differences. Areas that were forecasted as SLAMM cat 7 or 8 in a 25 year time step were most often tidal flat or open water in a 5 year.
Would this relate to the fact that the model only converts one class per time step? I have a fairly small tide range (2ft) so cells would not stay in a section of the inundation model for long.
Robert, you are right on. When I mentioned that I hadn't found much difference in my last time-step analysis this was in a system in which there were essentially no changes at time-zero.
Making sure the model behaves appropriately at time-zero is a very important step. In some cases where there is a horizontal DEM to NWI off-set you will have open water, or tidal-flat elevations in a location that is coded as dry land. In this case, the land type will move from dry land to transitional marsh to salt marsh to tidal flat to open water in each successive time-step. Going through the SLAMM decision tree each transition only occurs once per time-step.
In a case where the time-zero results look pretty good, and SLR is not so fast as to cause multiple transitions per time-step, I did not see much sensitivity to the time-step choice last time I checked.
I hope this helps to some degree.
Thanks for the reply. It seems that the 5 year timestep is best for modeling in my research areas because of the small tidal range.
I know you have mentioned before that it is important to have time zero and the initial condition essentially the same. What amount of difference have you found is acceptable. Most of the classes in my model are within a 5% difference in coverage between the T-O and IC though it does change depending on the protection scenario.
Thanks for your ongoing support.
I think that the discrepancy in "time-zero" results gives you a partial sense of the uncertainty of the model, even in the current condition. There will almost always be some discrepancy due to horizontal off-sets between the land-cover and elevation data-sets, general data uncertainty, or other local conditions that make a portion of your site not conform perfectly to the "conceptual model."
Five percent seems pretty good, in my opinion.