Recent Lessons about Uncertainty and Sensitivity Analysis in Environmental Modeling

Models of environmental systems are infinitely explorable, and exploration can yield 
insight into numerical hazards and solutions, reveal ways in which the model provides an 
inadequate representation of the system, and provide assessments of model utility. The 
recent lessons addressed in this talk include three from hurricane Sandy, including the 
power of (a) being transparent, (b)  demonstrably improving prediction accuracy, and (c) 
understanding how complex environmental systems work through testing hypotheses using data, 
and conducting sensitivity analysis and uncertainty quantification. Additional lessons are 
derived from consideration of the following recently developed capabilities: 
(1)  MODPATH_OBS, software that makes it easier to test models against some kinds of 
transport, water quality, and age data; 
(2) SENSITIVITY GRAPHS stacked to show the contribution of defined observation types, 
(3) ALTERNATIVE MODELS that allow evaluation of model structure uncertainty (demonstrated 
here using the modular FUSE surface-water model and the MODFLOW groundwater model),
(4) DELSA (Distributed Evaluation of Local Sensitivity Analysis) to conduct more insightful 
multi-scale sensitivity analysis and tested using surface-water models constructed using 
FUSE,
(5) COMPUTATIONALLY FRUGAL AND DEMANDING METHODS for quantifying prediction uncertainty 
with tests conducted using MODFLOW.
Attendees can expect to come away with new insights and ideas for their model development 
efforts.