441
polygons within a given spatial range) for such models. Users of data-driven
approaches often use higher level languages (e.g., Python, R) that already have
the appropriate scientific tools (coded by someone else) for their tasks as well
as due to their simplicity of use and re-usability of the code for sharing in the
scientific community.
(b) Coupling multiple models: The essential nature of the FEW Nexus requires
developing models that represent human and natural systems (e.g., energy,
agriculture, and land use, water supply and use, economy, and climate).
Furthermore, these models may be mechanistic models, data-driven models, or
some combination of the two. Many FEW models reflect the coupling of
empirical and model-based approaches, often incorporating optimization
methods to assist in identifying the level of process complexity that is consistent
with the observations (Guillibert 2015; Mohtar & Daher 2012; NSF 2015).
Coupled models raise issues of end-to-end testing, debugging, evolution,
uncertainty quantification, and validation. For these tasks, good visualization
tools [45] are essential. Two related important questions are how to enable
modelers and end users to understand the essential effects of the coupling of
multiple models and how does the coupling alter the behavior of the individual
models and how does this affect their validity.
(c) Variable temporal and spatial scale: Both the data and the models in the FEW
Nexus typically involve multiple spatial and temporal scales. For example, an
individual farmer deploying precision agriculture may need a model of soil
moisture and nutrients at the 1-m scale at a weekly timescale. At the same time,
the operator of a system of hydroelectric dams may only need precipitation
estimates at the scale of an entire watershed and yet need to model the hydrograph at 10-min intervals.
The data collected to create and validate such models is similarly characterized by a wide range of spatial and temporal scales Zhang (2010). Human
population, an important input to economic models, is measured in the USA
every 10 years and released at the level of census blocks. The MODIS satellite
instrument scans the entire planet every day at scales of 250 m, 500 m, and
1 km cells, whereas the LANDSAT 7 instrument produces data at 16–60 m
and a temporal resolution of 16 days (Allen et al. 2007; Norman et al. 2003;
Steering Committee for NASA Technology Roadmaps 2016).
Differences of scale raise two important issues: First, when creating or evaluating models, if the spatiotemporal scale at which the model operates is different from the spatiotemporal scale at which the data are observed, then we must
find a way to bridge that gap. Second, when coupling multiple models, if the
models operate at multiple spatiotemporal scales, then we must find a way to
link them properly. Currently, this linking is done using interpolations
(Goovaerts et al. 2010), or estimations which cause results to be inaccurate or
cause missing information at the finer scale.
(d) Spatio-temporal modeling: Understanding the FEW nexus from a computer science perspective requires developing effective and efficient spatiotemporal
models. Observational data on FEW are spatiotemporal in nature. Remote sens16 Computing
polygons within a given spatial range) for such models. Users of data-driven
approaches often use higher level languages (e.g., Python, R) that already have
the appropriate scientific tools (coded by someone else) for their tasks as well
as due to their simplicity of use and re-usability of the code for sharing in the
scientific community.
(b) Coupling multiple models: The essential nature of the FEW Nexus requires
developing models that represent human and natural systems (e.g., energy,
agriculture, and land use, water supply and use, economy, and climate).
Furthermore, these models may be mechanistic models, data-driven models, or
some combination of the two. Many FEW models reflect the coupling of
empirical and model-based approaches, often incorporating optimization
methods to assist in identifying the level of process complexity that is consistent
with the observations (Guillibert 2015; Mohtar & Daher 2012; NSF 2015).
Coupled models raise issues of end-to-end testing, debugging, evolution,
uncertainty quantification, and validation. For these tasks, good visualization
tools [45] are essential. Two related important questions are how to enable
modelers and end users to understand the essential effects of the coupling of
multiple models and how does the coupling alter the behavior of the individual
models and how does this affect their validity.
(c) Variable temporal and spatial scale: Both the data and the models in the FEW
Nexus typically involve multiple spatial and temporal scales. For example, an
individual farmer deploying precision agriculture may need a model of soil
moisture and nutrients at the 1-m scale at a weekly timescale. At the same time,
the operator of a system of hydroelectric dams may only need precipitation
estimates at the scale of an entire watershed and yet need to model the hydrograph at 10-min intervals.
The data collected to create and validate such models is similarly characterized by a wide range of spatial and temporal scales Zhang (2010). Human
population, an important input to economic models, is measured in the USA
every 10 years and released at the level of census blocks. The MODIS satellite
instrument scans the entire planet every day at scales of 250 m, 500 m, and
1 km cells, whereas the LANDSAT 7 instrument produces data at 16–60 m
and a temporal resolution of 16 days (Allen et al. 2007; Norman et al. 2003;
Steering Committee for NASA Technology Roadmaps 2016).
Differences of scale raise two important issues: First, when creating or evaluating models, if the spatiotemporal scale at which the model operates is different from the spatiotemporal scale at which the data are observed, then we must
find a way to bridge that gap. Second, when coupling multiple models, if the
models operate at multiple spatiotemporal scales, then we must find a way to
link them properly. Currently, this linking is done using interpolations
(Goovaerts et al. 2010), or estimations which cause results to be inaccurate or
cause missing information at the finer scale.
(d) Spatio-temporal modeling: Understanding the FEW nexus from a computer science perspective requires developing effective and efficient spatiotemporal
models. Observational data on FEW are spatiotemporal in nature. Remote sens16 Computing