A Data Reduction Exercise to Detect Threshold Samples for Regression Models to Predict Key Water Variables
✍ Scribed by Lars Håkanson
- Book ID
- 102283535
- Publisher
- John Wiley and Sons
- Year
- 2007
- Tongue
- English
- Weight
- 185 KB
- Volume
- 92
- Category
- Article
- ISSN
- 1434-2944
No coin nor oath required. For personal study only.
✦ Synopsis
Abstract
Inherent uncertainties in empirical data limit our understanding of interrelationships among variables and constrain our possibilities to identify critical thresholds as well as our possibilities to develop practically useful predictive models for water management. This work concerns key water variables for water management and the first aim is to utilize a very comprehensive set of data set for Ringkobing Fjord, Denmark. The paper first presents the methods and data used, then a reference regression for chlorophyll, coefficients of variation (CV = SD/MV; MV = mean value; SD = standard deviation) for a variety of water variables and how these CV‐values influence n, the number of data used to determine coastal area characteristic mean or median values (note that the interest here is not on the conditions in sampling bottle but on the conditions in entire coastal areas, the ecosystem perspective). The main part of the work presents a data reduction exercise including a definition of an error function where the focus is on “large N”, i.e., the number of data in a regression. The results are summarized in a diagram relating the error in the regression to different water variables with different inherent CVs in rivers, lakes and coastal areas. Given the inherently high CV‐values of many of these water variables, more samples than generally taken in most regular monitoring programs are needed if scientific unassailable conclusions are to be made concerning interrelationships among the variables and to produce scientifically meaningful information to detect critical ecosystem changes and threshold values. (© 2007 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)