Skip to main content
ARS Home » Research » Publications at this Location » Publication #93798


item Meek, David
item Prueger, John
item Sauer, Thomas - Tom
item Kustas, William - Bill
item Hatfield, Jerry

Submitted to: Agricultural and Forest Meteorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/12/1999
Publication Date: N/A
Citation: N/A

Interpretive Summary: When comparing different environments, researchers try to understand what properties and conditions are similar and what are different. Some particular properties of interest to many researchers are incoming solar energy and the processes that utilize it at the air/land surface. Sensors called radiometers are employed for this job. Several studies have had problems with the data collected because sensors from different manufacturers or designs were used to measure the same quantity. Results from the comparative measurements of simultaneous recordings at the same location were sufficiently dissimilar to cause concern. In the analysis on this kind of data record, a statistical property known as serial correlation is almost always present but is not routinely considered. Two real world case studies are presented to demonstrate the benefit of considering the serial correlation property. Realization of the problem alone as well as adoption of the appropriate statistical methods would provide insight into the sensor difference problem, greatly help future research efforts, and further studies with existing data sets.

Technical Abstract: During particular ambient atmospheric and surface conditions, systematic bias between two different models of radiometers limits researchers' ability to compare instruments and have confidence in the calibrations. Previous work revealed that two different models of net radiometers were most dissimilar during periods of dry ambient surface conditions. Hence, during dry ambient periods there is a need to post-process net radiometer data (Rn) to track measurements from a high precision net radiometer chosen as the reference for the measurement. Furthermore, adjustment of the nonreference Rn data via a first-order autoregressive model was found to be a more statistically sound means of reducing the systematic error. Interpolation error was half of ordinary least squares regression. The underlying problem of serial correlation that required the use of autoregression seems not to be considered by researchers in agricultural and forest meteorology. Hence, two examples are provided that illustrate the phenomenon and show the merit of autoregressive models. The first example, precised in a recent publication, details the comparison of two Rn measurements with one measurement chosen as the reference Rn data. The other example details the comparison on global solar radiation (Rs) data from two different pyranometers with one measurement chosen as the reference Rs data. In the future, the possibility of as well as consequences of serial correlation needs to be considered. When and if need be, there are well established methodologies to deal with the problem.