Final parameter combinations for runs

During the past 2 weeks I have run additional sets of N x M parameter combinations in order to narrow down the search for a starting point and set of parameter values to use in the large set of runs. I think I've come up with such a set, all of which produce 'reasonable' or 'possible' looking configurations of the MNRR, although this evaluation is based primarily on my own subjective interpretation of the resulting geometry and dynamics. Using runs performed with paired combinations of 5 x 5 values (25 combinations) of {ldist and sdist}, {width and depth}, and {flow and diameter} (i.e. 3 independent 2-dimensional subspaces of the 6-dimensional parameter space), I've decided on the following discrete values: 

                   ldist: 1.0 only
                   sdist: 0.75, 1.0, 1.25 local widths
Maximum attenuated width: 500, 600, 700, 800 m
                   Depth: 2.75, 3.0, 3.25, 3.5 m
                    Flow: 800, 1000, 1200, 1400 m^3/s
       Particle diameter: 0.0005, 0.00075, 0.001, 0.00125 m
    Initial erosion rate: 1.2, 2.4, 3.6 ha/yr/km

This will yield 1x3x4x4x4x4x3 = 2304 total runs, requiring an estimated 33.6 days to complete, if all goes well. Assuming some unanticipated problems, this should realistically allow me to finish the runs by early- to mid-November, allowing sufficient time for the analysis and write-up. While the runs are in progress, I will be doing other things (see below).

Here are some representative results. Except as noted, all use the base values of: 

ldist: 1.0
sdist: 1.0
 wmax: 600
depth: 3.0
 flow: 1000
 diam: 0.001
erate: 2.4

sdist = 0.75

sdist = 1.0

sdist = 1.25

wmax = 500, depth = 2.75

wmax = 600, depth = 3.0

wmax = 700, depth = 3.25

wmax = 800, depth = 3.5

diam = 0.0005, flow = 800

diam = 0.00075, flow = 1000

diam = 0.001, flow = 1200

diam = 0.00125, flow = 1400

The 'median' simulation consists of parameters: 

ldist: 1.0
sdist: 1.0
 wmax: 650
depth: 3.125
 flow: 1100
 diam: 0.000875

erate = 1.2

erate = 2.4

erate = 3.6

coverage (yrs per (0.5 km)^2) at erate = 1.2

coverage at erate = 2.4

coverage at erate = 3.6

erosion (ha per (0.5 km)^2) at erate = 1.2

erosion at erate = 2.4

erosion at erate = 3.6

river length (measured every 100, 2500, 5000, 7500, 10000, 12500 m), sinuosity (same lengths), coverage (km^2), erosion rate, total eroded area (ha), and erate vs. sinuosity for initial erate = 3.6:

Caveats

Here is a comparison of the MNRR at the end of the 20th century and 100 years before that:


(Click for larger image)

It seems clear that:

  1. The JP method does not really produce results that look like the MNRR. It produces 'river-like' loops and all, but they don't have the same 'flavor' as the MNRR bends. Other meandering methods (for example, the variable-width method I tried briefly several months ago) produce potentially more similar results:

    Nevertheless, by virtue of the different parameter combinations, the JP method produces a wide variety of geometries which are probaby sufficient for the 'proof of concept' of this project.

  2. The MNRR hasn't changed all that much in 100 years. For examples, see figures 29 and 31 of EJ06. Similarly, the simulations performed at the nominal erosion rate of 1.2 ha/yr/km are pretty boring, and have few if any new loops or cutoffs. Only those simulations with exaggerated migration rates are of visual interest. Therefore, making use of such exaggerated simulations in this project might be called a 'dramatization' of the evolution of the river over the next 100 years. Or, what might happen if the circumstances conspired to create greater than average migration which was not properly managed by human intervention. By running many different simulations having different combinations of width, depth, flow, erosion rate, etc... it also becomes possible to delineate at what point the evolution of the river becomes dangerous to established human habitation and other property.

The 2304 simulations are expected to amass about 9.9 Gb of data, which will require some non-trivial monitoring and management, and the creation of several new programs for analysis and visualization. While they are underway, I will be also working on the following:

  1. Digitizing the 1890 maps of the MNRR.
  2. Trying to create a quantitative 'metric' for comparing the geometries of the river in the 1890s and the 1990s, and for ranking and sorting the resulting simulations based on similarity to these two configurations.
  3. If possible, deriving a 'vector space' of river configurations, in which the 1890 and 1990 points (and the 2304 new variations) form a line or cluster in 6d which can be extrapolated to predict other potential configurations at different times in the past or future of the river.

© Sky Coyote 2008.