A Validated Fire Model Can Still Be Outside Its Domain of Applicability
Validation is always against a particular set of fires: certain fuel types, certain slopes, a range of wind speeds, a scale, and usually a set of conditions mild enough to have been safely instrumented. Outside that envelope the model is extrapolating, and its agreement with data says nothing about the case in hand. A surface spread model that reproduces grass-fire experiments beautifully contains no crown fire, no ember transport ahead of the front, and assumes the fire has settled into a steady rate in continuous fuel — so it is silent about exactly the fires that do the damage. The rule is mechanical and worth doing in writing: list your case's conditions in one column and the validation set's conditions in the other, and mark every row where yours falls outside. Each mark is a reason for caution, not necessarily a reason to stop, but it must be declared with the result. The confusion this resolves is treating validation as a property of a model rather than of a model-and-conditions pair. After this Concept you can decide whether a validated model's envelope actually covers the fire in front of you.
This Concept is waiting for its first lesson!
Validation is always against a particular set of fires: certain fuel types, certain slopes, a range of wind speeds, a scale, and usually a set of conditions mild enough to have been safely instrumented. Outside that envelope the model is extrapolating, and its agreement with data says nothing about the case in hand. A surface spread model that reproduces grass-fire experiments beautifully contains no crown fire, no ember transport ahead of the front, and assumes the fire has settled into a steady rate in continuous fuel — so it is silent about exactly the fires that do the damage. The rule is mechanical and worth doing in writing: list your case's conditions in one column and the validation set's conditions in the other, and mark every row where yours falls outside. Each mark is a reason for caution, not necessarily a reason to stop, but it must be declared with the result. The confusion this resolves is treating validation as a property of a model rather than of a model-and-conditions pair. After this Concept you can decide whether a validated model's envelope actually covers the fire in front of you.
Are you a teacher? Sign in to start contributing.
Sign In