In search of the unusual: on the benefits of long-term ecosystem-atmosphere exchange observations

Observations in any field of scientific enquiry are used to examine hypotheses and questions, to develop parameterisations, or to evaluate and drive models. As such, observational data are commonly interpreted in the framework of preconceived notions; they form the foundation of established knowledge. However, progress beyond the bounds of knowledge becomes possible when observations deviate from the usual and indicate events, trends or patterns that cannot readily be attributed to known forcings. In most traditional disciplines of science, the envelope of what is known can be extended by manipulative experimentation, but in environmental science manipulation is not generally possible, with few exceptions. Here we argue that, in the study of ecosystem-atmosphere exchange, long-term programs of continuous observation take the place of manipulative experiments. Long-term observations explore trends and patterns and establish the ensemble of possible states in the joint distribution of variables. Continuous observations ensure that singular events and sharp transitions are detected at all. We further argue that the most important prerequisite for the detection of something unusual is the establishment of what is usual. Using examples of long-term measurements of ecosystem-atmosphere exchange over Morgan-Monroe State Forest (MMSF) in the US Midwest, we show that the definition of "usual behavior" (and consequently the identification of outliers in inter-annual variability) changes, as the observational record gets longer.

We argue that a comprehensive long-term observation program, with continuous data quality control, analysis and interpretation, combined with modeling, goes far beyond mere fishing for serendipity. Such programs are invaluable tools to detect the scales of environmental variability and long-term trends; they form the basis for the identification of anomalies and their underlying processes; they are the most important data source for the independent evaluation of Earth-system-climate models.

The most obvious significance of continuous long-term observations is their utility for continuous evaluation of a model over long time periods. A more subtle point arises from the recognition that every environmental observation site is to a certain extent unique, and thus the dimensionality of the manifold of drivers and forcings in which a measured parameter assumes a given value at a given time is very large. In consequence, every observation must be seen a-priori as a unique value in a non-stationary, non-homogeneous field, collected in a unique set of conditions. Without the availability of an associated value, to which it can directly be compared (e.g., a known reference) it is not possible to obtain a well constrained estimate of its uncertainty. Without a measure for uncertainty, the utility of such observations for comparison with others or with modeling results is jeopardized. However, if the data point is embedded in a comprehensive long-term series, including observations that characterize the environmental forcing conditions in which the data were collected, it is possible to stratify the data set and objectively select a sample with comparable environmental conditions. Because the stratified sub-set exhibits a degree of homogeneity, it is permissible to use it for statistical analysis and uncertainty estimation. Clearly, the probability that a sufficient number of data points that meet given comparability criteria within a narrowly stratified data set can be identified, increases with the length of the data series. Thus, long-term observations provide an essential tool for environmental science to escape a fundamental methodological dilemma.