USGS Home PageGigalopolis Home Page
Home | About | Implement | Download | Discussion | Publications | Site Map
Home
About
Implement
functions & modes
data base setup
download and verify
calibrate
selecting ranges
      coarse
      fine
      final
      forecasting
predict
build products
Download
Discussion
Publications
Site Map
UCSB Home Page
 

Selecting coefficient ranges from coarse calibration

This is an example of one decision algorithm for narrowing coefficient ranges from a coarse calibration of Demo_city using the demo50 input images.

Using file control_stats.log:

  1. Sort file in descending order using the Lee Sallee metric
    Download entire control_stats.log file used for this sort here
  2. For the top three rankings: take high and low values of each of the coefficients
    It is possible for more than one run to have the same score, creating a tie
  3. For each coefficient: in the scenario file to be used for fine calibration, low values are set to _START
  4. For each coefficient: in the scenario file to be used for fine calibration, high values are set to _STOP
  5. A _STEP value is selected that will increment between the _START and _STOP values 4-6 times
    If only one coefficient value sorts into the top 10 (e.g.; "1" for dispersion and spread) select _START, _STEP, and _STOP values that will explore a finer coefficient space around the value

Top 3 scores from demo50 coarse calibration, sorting only on the Lee Sallee metric:

 
sort value
initial coefficient values
Run Product Compare Pop Edges Clusters Cluster
Size
Leesalee Slope %Urban Xmean Ymean Rad Fmatch Diff Brd Sprd Slp RG
15 0.08234 0.64331 0.92749 0.92189 0.99241 0.86224 0.34437 0.99988 0.88769 0.91861 0.94748 0.94694 0.6944 1 1 1 75 1
16 0.08234 0.64331 0.92749 0.92189 0.99241 0.86224 0.34437 0.99988 0.88769 0.91861 0.94748 0.94694 0.6944 1 1 1 75 25
17 0.08234 0.64331 0.92749 0.92189 0.99241 0.86224 0.34437 0.99988 0.88769 0.91861 0.94748 0.94694 0.6944 1 1 1 75 50
20 0.00213 0.61402 0.98361 0.9887 0.99985 0.89286 0.33744 0.85379 0.91667 0.97736 0.0221 0.99266 0.7044 1 1 1 100 1
21 0.00213 0.61402 0.98361 0.9887 0.99985 0.89286 0.33744 0.85379 0.91667 0.97736 0.0221 0.99266 0.7044 1 1 1 100 25
22 0.00213 0.61402 0.98361 0.9887 0.99985 0.89286 0.33744 0.85379 0.91667 0.97736 0.0221 0.99266 0.7044 1 1 1 100 50
145 0.02022 0.64854 0.98469 0.98736 0.97959 0.7033 0.31584 0.29837 0.91965 1 0.80372 0.99308 0.6728 1 25 1 100 1
146 0.02022 0.64854 0.98469 0.98736 0.97959 0.7033 0.31584 0.29837 0.91965 1 0.80372 0.99308 0.6728 1 25 1 100 25
147 0.02022 0.64854 0.98469 0.98736 0.97959 0.7033 0.31584 0.29837 0.91965 1 0.80372 0.99308 0.6728 1 25 1 100 50

Coefficient range for fine calibration:

coefficient type {_START - _STOP, _STEP}
dispersion {0 - 20, 5}
breed {0 - 25, 5}
spread {0 - 20, 5}
slope {75 - 100, 5}
road gravity {0 - 50, 10}