469
an ensemble. For pattern identification and hypothesis generation purposes, visual
analytics tools tend to implement a single model for understanding its spatiotemporal characteristics. Inter-comparison of model simulations is helpful in providing
insight into the uncertainty of an ensemble analysis because it can help understand
the impact of model structures on the outputs.
Other research has begun exploring the reliability and quality of climate projections through intercomparison of ensemble model simulations with varied model
parameters. Such intercomparisons can help researchers gain insight into how
model structures and varied parameters might impact model outputs and the way in
which different models develop across scales and evolve. However, little research
has explored the impact of spatial variation and scale on similarity analysis of modeled scenarios. Intermodel comparisons have also been used to address questions of
nutrient imbalanced in the Chesapeake Bay region; these multimodel comparisons
have resulted in significant changes in policy for the Chesapeake Bay Program.
Visual analytics of FEW systems modeling efforts can expand on the state-ofthe-art by developing novel geovisual analytics methods that will enable users to
explore the impact of spatial variations and scales for climate scenario comparisons.
Such work is critical at the intersection of FEW systems: tools need to allow for
regionally integrated assessment linked to nationally or even globally integrated
analysis and in terms of the interactions between ensembles of models of natural
and human systems with varied model parameters. In the Chesapeake Bay, a publicly available program called BAY-CAST now allows anyone to work with the
actual models used by the policy analysis with the Chesapeake Bay Program.
Moving forward, an important goal in FEW systems modeling is to develop
visual analytics methods for cross-scale comparison of outputs from IAMs, with a
focus on the impact of spatial and temporal scales and variations on FEW system
scenarios. One major challenge is exploring the impacts of these scenarios over different spatial scales. Methods for scenario comparison have traditionally explored
using clustering methods to group scenarios with similar outputs. Clusters can then
be compared for commonalities across input parameters, and changes to sensitivity
explored. However, a common issue in data classification is that items near a classification boundary are often mislabeled. As such, the goal is to augment automated
spatial classification schemes by utilizing interactive machine learning as part of the
cluster creation step.
Recently, the exploration of multidimensional similarity and interactive machine
learning has become a major focus in the visual analytics community: this approach
combines ideas from previous investigations and will focus on developing interactive brushing techniques for interactive classification relabeling.
Little work has been done in interactive machine learning with respect to utilizing geographic projections as an underlying representation of data distributions.
Instead, most work has focused on dimension reduction techniques and steering in
the reduced space (e.g., Choo et al. 2010). Key challenges here include the impact
of classification reassignment on distances in multivariate space and visually translating those distances to distances in physical or geographical space.
17 Applying Science to Practice
an ensemble. For pattern identification and hypothesis generation purposes, visual
analytics tools tend to implement a single model for understanding its spatiotemporal characteristics. Inter-comparison of model simulations is helpful in providing
insight into the uncertainty of an ensemble analysis because it can help understand
the impact of model structures on the outputs.
Other research has begun exploring the reliability and quality of climate projections through intercomparison of ensemble model simulations with varied model
parameters. Such intercomparisons can help researchers gain insight into how
model structures and varied parameters might impact model outputs and the way in
which different models develop across scales and evolve. However, little research
has explored the impact of spatial variation and scale on similarity analysis of modeled scenarios. Intermodel comparisons have also been used to address questions of
nutrient imbalanced in the Chesapeake Bay region; these multimodel comparisons
have resulted in significant changes in policy for the Chesapeake Bay Program.
Visual analytics of FEW systems modeling efforts can expand on the state-ofthe-art by developing novel geovisual analytics methods that will enable users to
explore the impact of spatial variations and scales for climate scenario comparisons.
Such work is critical at the intersection of FEW systems: tools need to allow for
regionally integrated assessment linked to nationally or even globally integrated
analysis and in terms of the interactions between ensembles of models of natural
and human systems with varied model parameters. In the Chesapeake Bay, a publicly available program called BAY-CAST now allows anyone to work with the
actual models used by the policy analysis with the Chesapeake Bay Program.
Moving forward, an important goal in FEW systems modeling is to develop
visual analytics methods for cross-scale comparison of outputs from IAMs, with a
focus on the impact of spatial and temporal scales and variations on FEW system
scenarios. One major challenge is exploring the impacts of these scenarios over different spatial scales. Methods for scenario comparison have traditionally explored
using clustering methods to group scenarios with similar outputs. Clusters can then
be compared for commonalities across input parameters, and changes to sensitivity
explored. However, a common issue in data classification is that items near a classification boundary are often mislabeled. As such, the goal is to augment automated
spatial classification schemes by utilizing interactive machine learning as part of the
cluster creation step.
Recently, the exploration of multidimensional similarity and interactive machine
learning has become a major focus in the visual analytics community: this approach
combines ideas from previous investigations and will focus on developing interactive brushing techniques for interactive classification relabeling.
Little work has been done in interactive machine learning with respect to utilizing geographic projections as an underlying representation of data distributions.
Instead, most work has focused on dimension reduction techniques and steering in
the reduced space (e.g., Choo et al. 2010). Key challenges here include the impact
of classification reassignment on distances in multivariate space and visually translating those distances to distances in physical or geographical space.
17 Applying Science to Practice