Uncertainty & environmental process modelling

摘要 这项提议很大程度上涉及环境过程，如在海洋等大空间域中演变的天气。已经形成了两种截然不同的建模文化，一种基于试图复制现实的确定性模型，另一种基于试图总结现实的随机模型。该提案探讨了将这些方法结合起来的方法，以克服需要预测的各个方面的不足。事实上，建议的研究将适用于一些不同的确定性模型和测量方法需要整合的情况，如天气预报。该提案的一部分涉及更好的环境过程统计模型，这些模型的参数可以随着时间和空间的变化而逐渐变化。提出了处理所涉及的非常大的域和数据集的方法。(我们讨论的一个应用涉及一个具有超过3-3个参数的统计模型。) 为了保护人类的健康和福利，监测空气污染等环境过程需要设置一个新的监测仪网络。然后，这些可能是专门构建的，以检测不符合法规标准的情况。我们将考虑如何调整测量以消除偏差，以及如何为其他目的定位它们，例如获得对健康影响的准确估计。 气候变化将重点放在开发工具，以更好地管理加拿大农业的风险。保险和期货合约应该如何定价。我们将继续这方面的工作，并制作涵盖空间和时间的模型，以预测当地农场一级干旱和第一次夜间霜冻的持续时间。将在预测作物产量作为土壤水分的函数方面做进一步的工作。 最后，该提案涉及理论问题。特别是，我将进一步探讨信息如何影响不确定性。第一个问题越多，第二个问题就越少？答案通常是否定的，但在什么情况下不确定性会增加呢？这样的问题还没有得到太多的探讨。

Summary Much of this proposal concerns environmental processes such as weather that evolve over large spatial domains such as an ocean. Two very different modelling cultures have developed, one based on deterministic models that attempt to replicate reality and the other, based on stochastic models that try to summarize it. This proposal explores ways of combining those approaches, to overcome their individual deficiencies where forecasts are needed. In fact, the research proposed would apply when a number of different deterministic models and measurement methods need to be integrated, as in weather forecasting. Part of the proposal concerns better statistical models for environmental processes, ones whose parameters can change gradually over time and space. Approaches are presented for handling the very large domains and datasets involved. (One application we discuss involves a statistical model with more than 3mi parameters.) Monitoring environmental processes such as air pollution to protect human health and welfare entails siting a newtwork of monitors. These may then be purpose built to detect non compliance regulatory standards. We will consider both how to adjust the measurements to remove bias and also how to site them for other purposes such as the getting accurate estimates of the processes impact on health. Climate change has put emphasis on developing tools to better manage risk in Canadian agriculture. How should insurance and futures contracts be priced. We will continue work in this area and produce models covering space and time to forecast the duration of a drought and the first night frost at the level of a local farm. Further work will be done on forecasting crop yields as a function of soil moisture. Finally the proposal addresses theoretical issues. In particular, I will explore further how information impacts on uncertainty. Does more of the first always lead to less of the second? The answer is no in general, but under what circumstances does an increase in uncertainty occur. Such issues have not been much explored.