Effect of sampling frequency on trend detection for atmospheric fine mass

Abstract

A time series model was fitted by site to atmospheric fine particle mass data collected on a daily basis at several sites of the Eastern Fine Particle Visibility Network. The estimated parameters from the model fitting were used in evaluating the effect of different sampling frequencies on the sensitivity of a trend test. The presence of autocorrelation in the data was confirmed and indicates some degree of temporal redundancy in the fine particle mass data. This redundancy was quantified in terms of relative variances of annual means computed for different sampling frequencies and in terms of power curves for a trend test. A diminishing return in terms of power of the trend test was seen for increasing investments in sampling effort. As an example, an increase in sampling rate from once every two days to once every day resulted in a maximum increase in the probability of detecting a trend of 0.04 for the scenarios examined in this report. Thus, sampling frequency can be reduced from once per day to once every other day without a major penalty in terms of decreased power. In general, power of the trend test was insensitive to sampling frequency for short (less than 5 yr) or long (greater than 15 yr) time horizons. However, there were substantial trade-offs between sampling frequency and power of the test for intermediate time horizons (between 5 and 15 yr).