Deriving sampling frequency charts in Brittany, France

Timeline: Started in 2009 - ended in 2011

Project Description

The main goal of this project is to be able predict, at the watershed scale in Brittany, the level of uncertainty in the estimation of nitrate loads and other water quality indicators such as annual average concentration, etc. Having this information at hand opens the possibility to adjust sampling frequency and essentially human and financial resources on a per watershed basis, and yet still know a priori the uncertainty of the indicators estimated.

Why are there uncertainties in load estimations?

First, let us restate that all measurements are associated with uncertainties. There are many sources of uncertainties (details in publications available in the links section below) when measuring nutrient loads and indicators in watersheds. An intriguing and obvious one is the fact that flow and concentrations vary rapidly through time (e.g., figure above) and that infrequent sampling only gives a very partial representation of these variations. Additionally, for a given sampling frequency, there is essentially an infinite number of combinations of samples (see video). So for a particular sampling frequency and for a particular wateshed, there is not one uncertainty value, but a distribution of uncertainty values. The algorithm used to compute loads also play a major role (details in links below).

Frequency charts

In this project we found a way to predict uncertainty, obviously as a function of sampling frequency, but also as a function of a dimensionless parameter that characterizes the hydrological reactivity, V2%. The advantage of this parameter is that is applies to all watershed sizes! V2% corresponds to the percentage of the annual flow that has occurred in 2% of the time, corresponding to the highest flows. Intuitively, for the more flashy or reactive watersheds, flow and concentration variations are more poorly characterized by a set sampling frequency. We were eventually able to make a correlation between a level of uncertainty, a sampling frequency and a reactivity indicator. From that we were able to draw sampling frequency charts for loads and other water quality indicators such as the annual average concentration (Cavg) illustrated below.

The way to read these charts is as such. The watershed you are interested has a reactivity of V2%=10%. You are looking for no more than ±5% error in your annual load estimates. You find the value for the ±5% error (green lines) corresponding to V2%=10%, on the Y axis and you see that you should sample either every 9 or 10 days. There are two values because the charts are derived from the overestimation or underestimation limits of the uncertainty distributions. To have all the details on how we got there, you can find links to the articles and one presentation below. Other publications on sampling strategies can also be found here.

links

Return to top

This page was last updated on February, 13 2013.