From Models To Decisions (M2D)

如今，许多决策都是利用来自数值模型的证据做出的。这样的模型往往又大又复杂，而且往往看起来没有不确定性。然而，即使模型本身是确定性的，因为它对相同的输入集给出了相同的结果，输入几乎总是不确定的，模型结构本身也不完全已知。因此，任何模型的结果都是不确定的。确定模型中的不确定性通常被称为不确定性量化。这是一个困难的问题；即使对于前向问题，我们知道模型输入中的不确定性，并希望将其传播到模型输出。当(不确定的)观测被用来减少输入中的不确定性时，更难的是反向建模。一旦我们完成了反向和正向建模，我们就可以估计用于帮助做出决策的模型输出中的不确定性。然而，在任何决策中，来自模型的证据很少是唯一的考虑因素。决策者有他或她自己的不确定性和偏见，必须考虑和理解这些。给出相同证据的两个决策者往往会做出不同的决定。向决策者(通常没有接受过数学和统计学方面的培训)提供证据是困难和复杂的。该网络将着眼于向决策者展示不确定性的叙述和可视化的新方法。

Today many decisions are made using evidence from numerical models. Such models are often large and complex and often appear to be without uncertainty. However, even if the model is itself deterministic in that it gives the same result for the same set of inputs, the inputs are almost always uncertain and the model structure itself is not fully known. Thus any model result is uncertain. Determining the uncertainties in models is often referred to as uncertainty quantification. This is a hard problem; even for the forward problem where we know the uncertainty in the model inputs and we want to propagate it to the model outputs. Even harder is inverse modelling when (uncertain) observations are used to reduce the uncertainty in the inputs. Once we have done both inverse and forward modelling we can estimate the uncertainty in the model output that is being used to help make the decision. However, in any decision the evidence from the model is rarely the only consideration. The decision maker has his or her own uncertainties and biases and these must be taken into account and understood. Two decision makers given the same evidence will often make different decisions. Presenting the evidence to decision makers (often untrained in mathematics and statistics) is difficult and complex. The network will look at new ways of presenting narratives and visualisations of uncertainty to decision makers.