A new look at habitat suitability curves through functional data analysis

Habitat suitability curves (HSC) synthesize the preference of a species for important habitat variables and are, therefore, key components of various fish habitat models. However, HSC are developed at large scales (e.g. river or regional scales) that do not consider the differences that exist in available habitat conditions at smaller scales. To address this problem, a new look at HSC is taken through functional data analysis (FDA). It is an appropriate framework adapted for HSC construction because in FDA, each observation is a curve or a function. To illustrate the potential of FDA for HSC, a dataset of Atlantic salmon (Salmo salar) parr density and habitat variables constructed on two rivers was used. Functional regression models (FRM) were built to predict site-specific HSC based on the available habitat conditions for three salmon parr habitat variables: water depth, mean flow velocity and median substrate size. FRM explained a greater proportion of the variation in site-specific HSC (respectively 38.0%, 53.3% and 45.5% for depth, substrate size and velocity) compared to traditional HSC developed at the scale of each river or regionally that poorly fitted site-specific HSC. When HSC were aggregated into habitat suitability indices (HSI), weak relationships were found between HSI and parr density (R2 < 5%) for all models (traditional HSC and FRM). This study demonstrates that FDA is an innovative framework that can be used to predict more representative site-specific HSC adapted to differences in local available habitat. The results suggested that its potential should be further exploited in habitat modelling.