HEMI 2 • Hyper-Envelope Modeling Interface 2 • Uses Bezier curves to minimize over-fitting • Allows automatic fitting with manual adjustment • Provides interactive mode for developing models and batch model (“Monte Carlo”) for validation and uncertainty analysis.
Transcript
Slide 1
HEMI 2 Hyper-Envelope Modeling Interface 2 Uses Bezier curves
to minimize over- fitting Allows automatic fitting with manual
adjustment Provides interactive mode for developing models and
batch model (Monte Carlo) for validation and uncertainty
analysis.
Slide 2
Germination Percentage Shrestha, A., E. S. Roman, A. G. Thomas,
and C. J. Swanton. 1999. Modeling germination and shoot-radicle
elongation of Ambrosia artemisiifolia. Weed Science
47:557-562.
Slide 3
Maraghni, M., M. Gorai, and M. Neffati. 2010. Seed germination
at different temperatures and water stress levels, and seedling
emergence from different depths of Ziziphus lotus. South African
Journal of Botany 76:453-459. Over-fitting The Data? What should
the model look like? Maxent model for Tamarix in the US: response
to temperature when modeled with temperature and precipitation
Slide 4
Niche Temperature Salinity Population Growth +0-+0- +0-+0-
Temperature Salinity From the Theory of Biogeography Brown, J.H.,
Lomolino, M.V. 1998, Biogeography: Second Edition. Sinauer
Associates, Sinauer Massachusetts Environmental Space
Slide 5
Problems with Polynomials Lack flexibility Not well behaved at
boundaries f(x) = a 0 + a 1 x + a 2 x 2 +... + a n x n, where a n 0
and n 2 Wikipedia.com
Slide 6
US Environmental Space
Slide 7
Tamarisk Occurrences United States Environmental Space
Slide 8
Alternatives to Polynomials BioClim max and max boxes BioMapper
25x25 grid of boxes DOMAIN Distance to existing point NPMR
Hyper-Envelope Modeling Bezier Curves Multivariate adaptive
regression splines
Slide 9
Bezier and Spline Curves Wikiepedia.com P1P1 P0P0 P2P2
P3P3
Tamarisk Occurrences United States Environmental Space
Slide 12
HEMI Tamarix Model 08/10/2009International Biological
Information Science 338546 Annual Precipitation (mm) 107 -231 Min
Temp of coldest Month (C*10) Potential Habitat Unsuitable Habitat 5
Points -> 10 Parameters AUC: 0.76
Slide 13
Models for Tamarisk in W.U.S.
Slide 14
Results MethodTrain AUC Number of Parameters Test AUC HEMI
Automatic0.75100.76 HEMI Edited0.73100.73 BioClim0.6780.68
BioMapper Median0.746250.72 Maxent0.771820.78
Slide 15
Potential Habitat 08/10/2009International Biological
Information Science Potential Habitat Unsuitable Habitat
Slide 16
Global Tamarisk Map? 08/10/2009International Biological
Information Science 1.0 Habitat Suitability 2.0
Slide 17
Limits of HEMI 1 Computation limits as the number of predictors
increased Problems visualizing combinations of predictors beyond
2
Slide 18
HEMI 2 Limited to one response curve per predictor Similar to
MaxEnt and traditional regression Predictors can be transformed to
adapt for this to some extent Up to 10 predictors New interactive
interface with immediate feedback
Slide 19
HEMI Model 1 st point is tied to the left edge and the bottom
Last point is tied to the left edge and the bottom Remaining Points
can be moved vertically and horizontally