Pedometrics › Forums › 10 PM challenges › Can we develop communicable measures of uncertainty?
- This topic has 2 replies, 2 voices, and was last updated 3 years, 3 months ago by
Nicolas Saby.
-
AuthorPosts
-
May 30, 2019 at 12:27 pm #2545
Alexandre Wadoux
Keymaster(proposed by Gerard Heuvelink)
One of the things that pedometricians can be proud of is that as a rule we always quantify the accuracy of our products, typically by probability distributions, although sometimes limited to an RMSE or concordance correlation. Among others, we quantify uncertainties because it tells users of our products whether a product is accurate enough for the intended use. But somehow this is where it goes wrong: many users do not seem to care or are not able to comprehend our measures of uncertainty. We have not made a good job of showing why quantified uncertainty is important and how it can be used. For instance, it may be essential for decision making and risk analyses. If we can communicate uncertainty better then we might be more successful in getting users to appreciate and use our measures of uncertainty.October 23, 2019 at 2:23 pm #2595Nicolas Saby
Keymaster[by Philippe Lagacherie]
Actually, the uncertainty of our map has the same status as the special clauses that are written in small letters at the bottom of a selling contract or as the Personal Data Protection protocols that you should read before using any web tool: They are theoretically accessible but nobody cares. Obviously, when setting the GlobalSoilMap specifications, we naively overestimated the ability of the users to get to grips the uncertainty information, yet clearly provided by prediction intervals. My personal experience is that most of them simply dislike any uncertainty and the others do not know how to use it. We should force the users to open their eyes, even if the price to paid is to produce less attractive maps. The solution would be to degrade the map resolution at locations where uncertainty is higher than a given RMSE value or confidence interval width or simply not to provide predictions at these locations (like “terra incognita” of the old maps). I even suggest that the user provides him/herself the error threshold from his/her knowledge of his/her requirement so that the final map could be more acceptable for him (her).
October 25, 2019 at 1:42 pm #2608Nicolas Saby
Keymaster[By Zamir Libohova ]
Can we translate uncertainty at every scale to cost-benefit and risk assessment analysis capable of supporting any decision? “Show me the money” is a movie quote that somehow drives my point. I am not suggesting following the money literally, but to make the case to the decision makers for the need for more accurate predictions of soil properties and functions. If we can quantify in monetary values the impact that poor predictive soil maps have on the society, then maybe decisions makers would open their pockets. Maybe, because at the end of the day more accurate maps may not make a difference. However, it should not stop us from trying -
AuthorPosts
- You must be logged in to reply to this topic.