news and events

08.05.2014

Our iUSA tool is now available for download.

more ...

07.15.2014

Spatial Accuracy 2014 at MSU (July 2014).

more ...

Contact

Arika Ligmann-Zielinska Department of Geography 673 Auditorium Rd, rm 121 Michigan State University East Lansing, 48824-1117

ligmannz (at) msu.edu

Models of coupled human and natural systems (CHANS) require a large number of drivers that describe the economic, social, political, and environmental components of the system. Because of the variability of a number of CHANS model inputs, the simulations result in many output values. In addition, given that CHANS are inherently geographical, their models exhibit great variability over space and time. Thus, CHANS models produce multiple model outputs that represent different aspects of the system, from scalars that summarize the whole study area (e.g. the cost of land conservation, the area of residential land use) to maps that depict spatially-heterogeneous information (e.g. water quality indices for every lake in a given watershed, or a land use map of new residential development):

Simulation Framework

The results are combined into probabiliy distributions:

Probability Distributions

To delve into the causes of model output variability, we use the distributions in a comprehensive spatiotemporal sensitivity analysis (SA) based on variance decomposition. The SA starts from calculating variance of the scalar, or of every geographical(temporal) location within the study area, which is then broken down an apportioned to inputs:

SA Outputs

The framework has very practical implications. By comparing sensitivities among inputs, modelers can prioritize which inputs have a negligible effect on output variability (leading to model simplification), and which inputs are critical in shaping this variability (leading to prioritization of data collection). In addition, the results of spatiotemporal SA, which we call maps of input dominance (Ligmann-Zielinska and Jankowski 2014), partition the study site into zones of the most influential inputs. This enhancement reveals spatial clusters that are more prone to input uncertainty, allowing for localized data collection or geographically focused process investigation.

Funded by NSF Financial support for this work was provided by the National Science Foundation Geography and Spatial Sciences Program Grant No. BCS 1263477. Any opinion, findings, conclusions, and recommendations expressed in this material are those of the authors(s) and do not necessarily reflect the views of the National Science Foundation.