The aforementioned issues have been extensively analyzed by meteorologists
and climate scientists using increasingly sophisticated statistical methods, see e.g. the
recent papers by Haylock et al. (2008) for statistically motivated interpolation methods,
and Menne et al. (2010) for the methods used to correct for spurious changepoints and
trends. Nevertheless, the methods currently in use do not reflect the latest research in
statistics; see e.g. Smith and Cressie (2010) for a review of spatial interpolation methods
in the light of modern developments in spatial statistics, and the papers of Cassinus and
Mestre (2004) and Reeves et al. (2007) for modern developments in changepoint
detection. Continued interaction between statistical scientists and experts in both land-
based and ocean-based observing systems is needed to deliver high-quality data
products for research and policy decision-making.
2.1.3 Human-element Datasets
From an anthropocentric perspective, sustainability is typically defined as
improving human well-being while maintaining the life-support systems of the planet.
Thus, measuring and monitoring human variables is crucial for assessing whether we
are making progress toward sustainability. Human-focused variables include population,
health, disease, wealth and its distribution, security, economic indicators, education and
governance structures and institutions. Variables measuring human well-being and
activities are collected by a myriad of governmental agencies (e.g. census data),
international organizations such as the UN (e.g. poverty indicators) and commercial
companies (e.g. marketing data). Each entity uses different approaches for designing
their data collection and assimilation methods. Apart from availability, proprietary, and
privacy issues, these multiple datasets have different spatial scales and boundaries (e.g.
nation-states, geographic regions, political units or discrete survey locations), different
temporal sampling periods (with frequency ranging from 1 year for census data to nearly
zero for not-reproduced reports) and different levels of uncertainty (which is frequently
left unspecified). Often, data are altogether missing in some spatial regions or are
measured inconsistently over time. While the numerous sustainability indicators
proposed in the past have different aims and specifics (Parris and Kates 2003), many of
them have in common the need to perform calculations of these human variables
(together with relevant biotic and abiotic variables) in a unified spatio-temporal grid.
Getting the appropriate data from the diverse world of online data warehouses requires
both statistical methodologies and mathematical process-modeling to fill-in, interpolate
and extrapolate data, methods the mathematical sciences have been developing the
tools to do.
2.2. How to measure and monitor? Sampling design and scales
The examples above show the challenges of constructing and maintaining data
products for sustainability science. These examples illustrate the importance of bringing
general statistical principles of good sampling design to the forefront. The examples
also show some of the challenges of establishing baselines and modeling using data
collected over long periods and/or multiple instrumentations.
Good sampling design should follow the three principles of design espoused by
Fisher (1935), namely blocking (stratification), randomization, and replication. Even if a
spatio-temporal component was not incorporated in the design, an observation was