You are here:
QUANTIFYING UNCERTAINTY IN NET PRIMARY PRODUCTION MEASUREMENTS
Citation:
HARMON, M. E., D. L. PHILLIPS, J. BATTLES, A. RASSWEILER, R. O. HALL, AND W. K. LAUENROTH. QUANTIFYING UNCERTAINTY IN NET PRIMARY PRODUCTION MEASUREMENTS. Chapter 12, Timothy J. Fahey and Alan K. Knapp (ed.), Principles and Standards For Measuring Primary Production. Oxford University Press, Cary, NC, , 238-260, (2007).
Impact/Purpose:
we review concepts related to uncertainty, factors that contribute to uncertainty, how it can be estimated, and provide examples of this term for selected biomes
Description:
Net primary production (NPP, e.g., g m-2 yr-1), a key ecosystem attribute, is estimated from a combination of other variables, e.g. standing crop biomass at several points in time, each of which is subject to errors in their measurement. These errors propagate as the variables are mathematically combined, and the distribution of these propagated errors reflects the uncertainty in the NPP estimate. While often not reported, quantification of the component error terms and the resultant NPP estimation error is important for several reasons. First, such information allows the user of the data to assess its reliability. A single point estimate of NPP does not convey any notion of how good the estimate is, but an estimate with an associated confidence interval does. Second, it allows for more meaningful comparisons because two estimates will never be exactly the same. The interpretation of a 10% difference in NPP between two forest stands would be viewed differently if the NPP estimate had an uncertainty of 5% rather than 20%. Third, dissection of estimation error into its constituent components allows one to understand what factors are the major contributors to uncertainty and where efforts might best be focused to reduce this uncertainty.
In this chapter we review concepts related to uncertainty, factors that contribute to uncertainty, how it can be estimated, and provide examples of this term for selected biomes. While uncertainty is often viewed negatively, we encourage the view that it is just another dimension of understanding an ecological system. Although a major goal of science is to reduce the uncertainty of prediction, this is difficult to achieve when uncertainty is not quantified or explicitly expressed. Finally, great progress has been made in methods to estimate uncertainty using statistical error propagation and other methods such as Monte Carlo analysis. Given the availability of software and computers to perform these estimates, this task has become relatively easy.