472
decision environment through the use of interactive simulations, virtual reality, and
data-augmented discussions. This strategy goes by names like: “Serious Gaming”
(Sušnik et al. 2018), “Decision Theater” (White et al. 2010), and “Immersive
Analytics” (Chandler et al. 2015), among others; see also Sect. 20.3.3.
However, this strategy is more easily described than implemented. It takes a
great deal of investment in software, physical infrastructure, human resources, data,
and modeling, along with extensive research into the stakeholder’s identity and
decision processes, to create an effective, immersive decision environment. It is
essential that the decision-maker trusts the environment’s representation of reality.
This may explain why few of these environments have actually been constructed in
practice, and why those environments that have been constructed tend to be singlepurpose, highly focused environments intended for specific decisions. Generalpurpose decision environments are beyond the current state of the art.
17.4 Decision-Making Under Uncertainty
Imperfect spatiotemporal data from diverse sources related to food, energy, and
water systems creates uncertainty in data science approaches to FEW data analytics
and decision making. Modeling and efficient management of uncertain spatiotemporal data from the perspective of its broad impacts and applicability in the FEW
Nexus is a challenging task. The main reason behind this is that uncertainty can
come from various factors, including source data, generalization error of crossvalidation in supervised learning, the sensitivity of modeling choices for unsupervised learning, and a tendency to “false positives” due to multiple selections. This
chapter concludes with a brief overview of current issues and solutions in the management of uncertain spatiotemporal data and their transformative potential in
Nexus research.
One of the main motivations for investigating the efficient management of uncertainty in the context of agriculture and its impact on the consumption of other
resources—most notably, water and energy—is that it is simply a fact of life in
many different facets.
Most data-gathering devices, whether sensors for measuring different physical
phenomena (possibly organized in a network), imaging devices (e.g., satellitebased) for remote sensing, surveys, citizen science, or other, are inherently imprecise. There is always a limit on the accuracy of the measurements, and the accuracy
can vary based on the existing conditions. For example, at higher temperatures,
there may be a higher error in measuring humidity; clouds may occlude the view of
the satellite, etc. The impact is especially amplified when certain types of actual
measurements must be used for indirectly estimating other values.
In practice, regardless of the underlying technology, data samples are obtained in
discrete time-instants and at discrete geo-locations. However, the data are used to
approximate a continuous behavior of the underlying phenomena: implying that one
must resort to interpolation using models and take this into account in storage and
E. Eftelioglu et al.
decision environment through the use of interactive simulations, virtual reality, and
data-augmented discussions. This strategy goes by names like: “Serious Gaming”
(Sušnik et al. 2018), “Decision Theater” (White et al. 2010), and “Immersive
Analytics” (Chandler et al. 2015), among others; see also Sect. 20.3.3.
However, this strategy is more easily described than implemented. It takes a
great deal of investment in software, physical infrastructure, human resources, data,
and modeling, along with extensive research into the stakeholder’s identity and
decision processes, to create an effective, immersive decision environment. It is
essential that the decision-maker trusts the environment’s representation of reality.
This may explain why few of these environments have actually been constructed in
practice, and why those environments that have been constructed tend to be singlepurpose, highly focused environments intended for specific decisions. Generalpurpose decision environments are beyond the current state of the art.
17.4 Decision-Making Under Uncertainty
Imperfect spatiotemporal data from diverse sources related to food, energy, and
water systems creates uncertainty in data science approaches to FEW data analytics
and decision making. Modeling and efficient management of uncertain spatiotemporal data from the perspective of its broad impacts and applicability in the FEW
Nexus is a challenging task. The main reason behind this is that uncertainty can
come from various factors, including source data, generalization error of crossvalidation in supervised learning, the sensitivity of modeling choices for unsupervised learning, and a tendency to “false positives” due to multiple selections. This
chapter concludes with a brief overview of current issues and solutions in the management of uncertain spatiotemporal data and their transformative potential in
Nexus research.
One of the main motivations for investigating the efficient management of uncertainty in the context of agriculture and its impact on the consumption of other
resources—most notably, water and energy—is that it is simply a fact of life in
many different facets.
Most data-gathering devices, whether sensors for measuring different physical
phenomena (possibly organized in a network), imaging devices (e.g., satellitebased) for remote sensing, surveys, citizen science, or other, are inherently imprecise. There is always a limit on the accuracy of the measurements, and the accuracy
can vary based on the existing conditions. For example, at higher temperatures,
there may be a higher error in measuring humidity; clouds may occlude the view of
the satellite, etc. The impact is especially amplified when certain types of actual
measurements must be used for indirectly estimating other values.
In practice, regardless of the underlying technology, data samples are obtained in
discrete time-instants and at discrete geo-locations. However, the data are used to
approximate a continuous behavior of the underlying phenomena: implying that one
must resort to interpolation using models and take this into account in storage and
E. Eftelioglu et al.