Publications with HyperNiche

• Build habitat models for species presence-absence (estimate likelihood of occurrence in relation to multiple habitat parameters or other predictors).
  
  
• Build habitat models for species abundance (estimate abundance in relation to predictors).
  
  
• Estimate physiological response surfaces in relation to 1, 2, or more environmental parameters.  Open your physiological variables as the response matrix and your environmental variables as the predictor matrix.
  
  
• Build empirical models of species diversity in relation to multiple predictors.  Place your diversity measures (calculated with PC-ORD or a spreadsheet) in the response matrix.
  
  
• Relate community ordination scores to multiple environmental variables.  Save your ordination scores in a spreadsheet, then open this as the response matrix and your environmental variables as the predictors.
  
  
• Optimize sample stratification – choose combinations of variable to maximize differences in a response variable among strata.  Place potential stratifying variables in the predictor matrix, then conduct a free search to find the best combination of variables for predicting the response.
  
  
• Build multiple regression models for any nonlinear response to multiple interacting factors.

Simple, ecologically reasonable response surfaces pose difficult challenges to traditional habitat modeling tools.  HyperNiche provides tools that easily find complex response surfaces, such as this hump that combines sigmoid and Gaussian curves.

• GAMs and NPMR share the use of smoothing functions. Individual terms are combined additively in GAMs and multiplicatively in NPMR.
  
  
• GAMs model interactions by adding specific terms, while they are modeled automatically with a multiplicative kernel function in NPMR.
  
  
• Commercial software for GAMs is more expensive and harder to use than NPMR in HyperNiche.
  
  
• GAMs have no theoretical basis for additive terms in controlling species response functions, while NPMR has an explicit theoretical and biological foundation as a statistical representation of Shelford’s Law of Tolerance.
  
  

• In logistic regression (a form of generalized linear model, GLM) the analyst must specify in advance the form of the response function for each predictor and their interactions (e.g. sigmoid, hump-shaped, linear), even when the analyst has no theoretical basis for choosing one response type over another, while NPMR is open to any form of response function.
  
  
• Interactions must be modeled explicitly with logistic regression, by adding specific terms, while they are modeled automatically in NPMR.
  
  
• Commercial software for logistic regression is more expensive and harder to use than NPMR in HyperNiche.
  
  

• MARS fits separate splines to different intervals of the predictors, while NPMR applies a multidimensional kernel smoother.
  
  
• Both MARS and NPMR can fit response surfaces that differ fundamentally in shape in different parts of the surface.
  
  
• Both MARS and NPMR seek the best model by a numerically intensive and exhaustive search through the domain of possible models.
  
  
• Both MARS and NPMR are well suited to modeling interactions. NPMR does so automatically, while MARS users request consideration of interactions up to a specified level.
  
  
• Commercial software for GAMs is more expensive and harder to use than NPMR in HyperNiche.
  
  
• MARS has no theoretical basis in the biology of species response to habitat factors, while NPMR has an explicit theoretical and biological foundation as a statistical representation of Shelford’s Law of Tolerance.