Which spatial interpolators I should use? A case study applying to marine species

• A new method was developed to select among spatial distribution models using 2-steps. The 1st step uses the variance explained by predictive model (10-fold cross validation). 2nd 3 criteria: error based measures, equivalence measures and data integrity. is illustrated 175 marine species distributions (35 x 5. years). approach accessible, clear, multi-criteria universal as it does not depend on used or characteristics. Species are spread in space, whereas sampling sparse. Thus, describe map along environmental gradients , necessary interpolate abundance. Considering plethora of valid methods, researcher gets easily puzzled choose most appropriate interpolation with reference ecological question being asked. We propose a procedure alternative we illustrate * 5 In first step, (VEcv) given 10-fold validation estimated for each method. When inter-quartile range VEcv different methods overlap, selection passes second 11 belonging three 1) 2) (center gravity, inertia, isotropy index aggregation) 3) integrity after interpolation, example percentage area over maximum sampled data. applied our either stratified random survey (trawl) systematic (acoustic). found that 87% all had overlapping thus passed selection. best varied year, although general additive (GAM), Thin Plate Spline (TPS), Universal Kriging (UKr) Random Forest (Rfor) performed better trawl TPS, Ordinary (OKri) UKr acoustic . Further, results differed within (e.g. kriging neighborhood type kriging) small modifications specifications can have large impact surfaces produced. proposed accessible intuitive, require any complex software sophisticated methodology; shows exactly what aspects prevalent others permits make decision accordingly objectives study; takes into account criteria evaluate each, properties an method; 4) detailed review subject also included.