point estimates of future quantities (e.g., amounts of available resources) will be
insufficient for evaluating progress towards sustainability or for making decisions
regarding certain actions to achieve particular sustainability markers.
3.2. Sampling designs for monitoring and measurement
Design criteria - Here we offer a few examples of sampling or monitoring
designs that were or could be motivated by specific design criteria. These examples are
not necessarily meant to serve as specific role models as they could be improved upon
for sustainability research. They also are not meant to represent the only existing
designs; there are likely other relevant monitoring and sampling networks that could be
expanded upon for the purpose of obtaining information to advance sustainability
research. Because many different variables are expected to be important to developing
and testing models related to sustainability, we anticipate that a wide array of sampling
and monitoring designs will be necessary to obtain relevant information, and the spatial
extent of each sampling or monitoring network will likely vary depending on the variable
of interest and the scope of the research problem. Below we highlight examples of
existing designs and/or design criteria that span local to national to global scales.
New methods for sampling designs will be required for measurement and
monitoring of baselines and to estimate trends in sustainability. Since the data will
typically be spatio-temporal in nature, both the frequency of observations as well as the
number and location of sites for monitoring stations will need to be considered. In most
cases, multiple criteria will need to be considered when designing a network. For
example, a sampling design may be constructed to achieve optimal predictions, to
produce accurate estimates of key quantities of interest, to allow for model assessment,
to estimate spatial and temporal trends, to detect catastrophic change, to estimate
dynamical invariants, and/or to estimate parameters in models for extreme events. A
common goal in all sampling design is to construct networks that use the available funds
for monitoring in the most efficient manner possible.
When constructing a sampling design to address sustainability issues, we will
also need to consider other sources of data. Sustainability science will require new
statistical methods for combining massive quantities of observational and experimental
data, combining and modeling data collected at different scales, and combining and
analyzing historical and new data. For example, should we design a new network so
that you can maximize the use of historical data in future analyses to enable analyses of
longer time series?
One example of the challenges of constructing sampling designs in the context of
sustainability science error are the multiple efforts that are now emerging on developing
methods for assessing carbon, carbon loss, and carbon sequestration in tropical forests.
Many such forests are remote and inaccessible and do not lend themselves well to
ground-based sampling. Thus, the data requirements and acquisition methods for
constructing remote sensing-based maps that are of sufficient quality (e.g., quality of fit,
precision) to serve as the basis for assessing carbon change is a huge emerging issue.
Two aspects are crucial: (i) acquisition of reference data to train or calibrate the models
that are then used to predict carbon or carbon change for non-sampled areas using
remotely sensed data, and (ii) acquisition of validation or accuracy assessment data to
evaluate the quality of the resulting map. The relevant issue here is construction of
sampling designs and plot configurations that are simultaneously efficient for the