This section clarifies definitions and reviews some of the extensive literature relating to forest resource assessments, manipulated experiments and observational studies in forest ecosystems. Some parts were previously presented in two workshop proceedings that had not been subject to peer review (Zhao et al. 20112012; Amateis and Burkhar 2012).
Forest resource assessments (forest inventories)
The objective of a resource assessment, such as a National Forest Inventory (NFI), is to provide unbiased estimates of particular target variables and to present statistics about forest resources. NFIs assess forest areas, growing stock volumes, and changes in biodiversity status, land-use, carbon stock and ecosystem services. International agreements, including the Montréal Process (1998), the Ministerial Conference on the Protection of Forests in Europe (2008) and the Convention on Biological Diversity (2008), require information which is provided by NFIs. In North America, the Forest Inventory and Analysis (FIA) Program of the U.S. Forest Service provides the information needed to assess America’s forests (LaBau et al. 2007). Naturally, sampling designs, sample plot configurations and other assessment methods may vary among countries and there is diversity in the definitions of forest area and growing stock volume among countries. Thus, harmonizing assessment and reporting is essential to make NFI results comparable (Tomppo et al. 2010).
Manipulated experiments
A manipulated experiment is an investigation that attempts to establish a particular set of conditions under a specified protocol with the aim of testing a hypothesis. The adjective manipulated implies the establishment of a set of predefined treatments which allow comparison of the effects/responses resulting from these treatments (Fisher 1935; Cox 1958; Gadow and Kleinn 2005). Thus, the experiment deliberately imposes a treatment on a group of objects in the interest of observing the response. Examples of manipulated experiments include medium and long-term growth studies in response to different fertilizer applications and stand densities (O’Hehir 2001; Burkhart and Tomé2012), growth studies of clones on different sites, including marginal ones (Bungart and Hüttl 2004), growth and competition effects in multi-species forests (Vanclay 1994; Pretzsch 2009), or evaluating effects of afforestation on water yield in mountain catchments (Bosch and Gadow 1990). Despite attempts to create homogenous conditions, manipulated field experiments always involve uncertainties in controlling ceteris paribus conditions, which are necessary for obtaining noise-free “dose/response” relations.
Manipulated field experiments do not seem to be as popular as they used to be during the 1960’s and 70’s. Possible reasons are the high cost of maintaining the field experiments and the restricted ability to generalize beyond the homogeneous, and therefore limited, experimental conditions. An exception is the impressive network of the Forest Modeling Research Cooperative at Virginia Tech which has employed a combination of permanent sample plots in operational stands of loblolly pine (Pinus taeda) plantations and designed experiments to provide data bases needed to construct robust growth and yield models for projecting inventories and estimating response to a wide range of silvicultural treatments for plantations established primarily for wood production (Burkhart 2008). These extensive networks have provided the empirical base for models of tree taper and volume, site index, and tree and stand increment and mortality. Data from spacing trials that were measured annually over a 25-year period (Amateis and Burkhart 2012) and from silvicultural trials designed for estimating response to control of competing vegetation and fertilizer applications extend and enhance the basic modeling framework. (Amateis and Burkhar 2012). Because the plots were stem-mapped and measured regularly, it was possible to obtain data on decay rates and amounts of carbon in the dead trees at the time plot measurements were scheduled to be terminated in the first region-wide study (Radtke et al. 2009). With increased interest in climate influences on forests, both region-wide data sets, which cover a broad range of soil and climate zones where loblolly pine is planted, are providing highly valuable data for a large multi-disciplinary, multi-institutional research effort aimed at assessing climate influences on southern U.S. conifers (http://www.pinemap.org).
Forest observational studies
Models of forest ecosystem dynamics, including tree growth, mortality and recruitment, are often developed on the basis of repeated observations in specially selected observational field plots. Selection of sites is not random or systematic, as in forest inventories, but based on particular data requirements. Ecologists tend to use the term “opportunistic sampling” for this approach (De Barba et al. 2010). The aim of the early field experiments established during the 19th century was to measure the growth of trees which were numbered for re-identification during successive measurements. Some of these experiments have been maintained for over a century, providing valuable information on long-term developments (Spellmann et al. 1996). The field plots are usually larger than inventory plots, and measurements are often much more detailed. Despite advances in resource assessment technology, there seems to be a real need for long-term observational field studies with mapped trees in large field plots. Examples of the design and implementation of such studies are presented by Zhao et al. (2011), Corral-Rivas et al. (2012), Tewari et al. (2014) and Kiviste et al. (2012).
In contrast to a manipulated experiment, which deliberately imposes treatments on experimental plots with the aim of observing a particular response, a comparative observational study involves collecting and analysing data from different site conditions without actively pre-defining these conditions (Kuehl 1994). Comparative observational studies are also known as quasi-experiments (Campbell and Stanley 1963; Cook and Campbell 1979). Typical quasi-experiments are longterm forest observational studies. Extensive networks of longterm forest observational studies have been established since the middle of the nineteenth century. Franz v. Baur (1830–1897), Bernhard Danckelmann (1831–1901) and Gustav Heyer (1826–1883) were among the first who devised a concept for long-term investigations in forest science, emphasising the importance of experimental field plots. In 1929 the International Union of Forest Research Organisations (IUFRO) was established with the objective of standardising the design and analysis of long-term field experiments (Pretzsch 2009). The majority of these experiments are observational studies (Szaro et al. 2006). Numerous observational studies were also established in North America. (Seymour et al. 2006a) reviewed long-term silvicultural experiments in four regions of the United States. As early as the 1920s, when there was much interest in multi-aged silviculture, scientists recognized that silvicultural systems involving within-stand variation of tree age and size could not be tested effectively on small (<1 ha) plots, and began installing compartment-scale (10–20 ha) trials on many experimental forests throughout the United States. Such large-scale trials have experienced a revival during the 1990’s in response to a renewed interest in management methods aiming to maintain within-stand structural complexity and biodiversity at larger scales.
Longitudinal, cross-sectional and interval studies
Forest ecosystem studies are often designed as longterm experiments, which are known in the statistical literature as longitudinal studies. The key feature of longitudinal data is the fact that the same individual is repeatedly measured at successive points in time. Thus, the set of observations on one individual subject will tend to be positively correlated and this correlation needs to be taken into account (Crawley 2005, p. 180). Longitudinal studies are commonly used to describe and explain trends. One disadvantage of a longterm or “permanent” experiment is the high maintenance cost of the research infrastructure and the long wait for results (Zhao et al. 2011).
On the other hand, results are available relatively quickly in a cross-sectional study (also known as a chronosequence) which involves one-time measurements of a set of field plots which usually cover a wide range of ages and environmental conditions. Thus, the sequence of remeasurements in time (longitudinal study) is substituted by simultaneous measurements in space. This method has been used extensively during the 19th century (Kramer 1988, p. 97). Cross-sectional studies may be combined with stem analyses to reconstruct the development of tree height, for example (Lee 1993). They may provide information relatively quickly, but do not capture the response of a target variable to a given initial state. The initial state may be defined for example, by the competition effects in the past. A cross-sectional study evaluates differences in growth in response to different site conditions. It does not provide evidence that can be used to test an effect such as past forest density or competition. A cross-sectional study is the only possible approach, however, in experiments requiring destructive sampling, i.e. in biomass studies where individual trees, or a cohort of trees within a given area, have to be cut up, dried and weighed. Obviously, the selection of the particular individuals or the particular cohorts will greatly influence the results. A cross-sectional study provides one-time measurements for many individuals, but no rates of change. A longitudinal study does assess rates of change, but usually only for few individuals. Thus, there are two extreme cases (Crawley 2005, p. 180):
-
1.
few measurements on many individuals
-
2.
many measurements on few individuals
Cross-sectional studies typically involve few observations on a large number of individuals while in longitudinal studies few individuals are typically measured many times. A practical compromise between a longitudinal study and a cross-sectional study is an interval study. Interval plots are measured at least twice with the objective of assessing the rate of change of a target variable between successive re-measurements. As many sites as possible should be covered and there is no immediate need to observe the target variable for long periods of time. Thus interval studies strike a compromise, exploiting the primary advantages of longitudinal experiments (gathering specific age and cohort-related responses) and cross-sectional studies (broad coverage of site conditions and minimum wait for data). However, continuous observations over long time periods are indispensable for assessing certain longterm effects, such as maximum density and tree survival for different ages and initial espacements, and the effects of changing environmental conditions.
An important question concerns the balance between the number of growth intervals that should be assessed in one particular location and the range of environmental and treatment conditions that needs to be covered. In the ideal case, one growth interval will provide sufficient response and the funds that would be required for assessing a second interval at the particular location can be spent on gathering another set of interval data in a different locality. An interval plot should be continued after one growth period if new results are expected, or it may be abandoned and the available funds used in another location to increase the variety of initial states for which the response needs to be evaluated. Generally, interval plots are continued for multiple growth periods, but ongoing evaluation should be carried out of the tradeoff between abandoning current interval plot locations and establishing new plots in additional locations to increase the variety of initial states for which response data are needed.
Some plots have been continuously re-measured over long periods of time. Pelz and Kohnle (2012) list numerous examples of long-term field studies in Germany. New networks of such studies, some rather extensive, were established more recently. The impressive examples from the US, China and Mexico are described below in more detail. However, there are also examples of observational studies that had been established during the early years of the 20th century, were subsequently abandoned (usually after political change caused by a revolution or independence) and are being re-initiated. One such series was described by Sims et al. (2009). Concerted efforts are also being made in India to locate and revive old plots which had been established during colonial times (Tewari et al. 2014).
The ultimate aim of all field studies is to provide external validity, the ability to generalize from a limited set of observations. Generalizability depends on whether the observed response measurement is representative of a population of interest. Scientists need to clarify whether the results of the observations may be legitimately extended to the general population of interest. In theory, it should be possible to extend observations beyond the particular restricted data set. However, this is not always possible in practice, for example in the case of rare biotopes or endangered species that occur in geographically restricted areas. A comprehensive description of the study sites and of the methodology is helpful for users to judge whether the results are applicable to a particular situation.
Resource assessments, manipulated experiments and observational studies
Thus, according to Zhao et al. (2011) we may distinguish three common approaches for collecting field data in forest ecosystems: Resource Assessments, Field Experiments and Observational Studies. Resource Assessments are usually carried out at regular intervals, providing geographical information about specific target variables (biomass; tree species). The data are used for formulating policy by governments, land managing agencies and NGOs, for forest planning, and for locating timber resources and new processing facilities. A manipulated experiment deliberately imposes treatments on experimental plots with the aim of observing a particular response. Field experiments are more often designed for planted forests with the aim of developing optimum silvicultural treatments. Observational studies are often laid out in natural forests with irregular structures where treatments are difficult to define. Data are collected from different site conditions without actively pre-defining these conditions. Field plots are usually large to capture effects of scale, and assessments are scheduled to coincide with a disturbance. Resource Assessments, Field Experiments and Observational Studies may complement each other. However, in Observational Studies the primary objective is not hypothesis testing or geographical coverage, but improved understanding of ecosystem dynamics, especially regarding the immediate and long-term effects of harvest events and other disturbances.
Numerous long-term observational studies with small, large and nested plot structures have been installed and maintained by various institutions in different countries covering different time periods and diverse forest ecosystems. The utility of each design depends on specific data requirements (forest trees, or other vegetative life forms) and available funds. The following two examples represent successful new implementations.