Premium
Using biological metrics to score and evaluate sites: a nearest‐neighbour reference condition approach
Author(s) -
BATES PRINS SAMANTHA C.,
SMITH ERIC P.
Publication year - 2007
Publication title -
freshwater biology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.297
H-Index - 156
eISSN - 1365-2427
pISSN - 0046-5070
DOI - 10.1111/j.1365-2427.2006.01675.x
Subject(s) - metric (unit) , nearest neighbour , measure (data warehouse) , reference data , data mining , computer science , statistics , k nearest neighbors algorithm , set (abstract data type) , selection (genetic algorithm) , range (aeronautics) , regression , artificial intelligence , pattern recognition (psychology) , mathematics , engineering , operations management , programming language , aerospace engineering
Summary 1. Reference (i.e. least or minimally impaired) sites can provide important information about the expected range of biological metrics and can be used to establish impairment or non‐impairment of a test site. A problem with using reference data is that biological metrics are affected by natural conditions. We present an approach that uses local information to adjust for natural conditions and a method for statistically evaluating condition at a test site using biological metrics. 2. Our method consists of four steps: selection of a distance measure to find neighbours of a test site, selecting natural variables to measure the distance, selection of the number of neighbours and calculating a scored metric. 3. We use a simulated example to illustrate when the nearest‐neighbour approach improves classification of sites as reference or not reference. 4. Using a set of data from the Mid‐Atlantic Highlands, we show that the nearest‐neighbour method improved on the ability of a regression approach to correctly classify test sites known to be from a non‐reference group without affecting the ability to correctly classify test sites known to be from the reference group.