UncertWeb logo: uncertainty enabled model web

Frequently Asked Questions

The UncertWeb project

Q. What is UncertWeb about?
A. UncertWeb is about how to compose workflows of resources within a GEOSS or web service context accounting for uncertainty. Lets unpack that.

Composing workflows: this is about linking together activities, often to form a chain, to address a specific problem.

Resources: we talk about resources, because these might be data sources, or they might be models. For example we might want to integrate a digital elevation model and a model to predict solar radiance at a given location.

Within a GEOSS or web service context: we are consider resources exposed on the web using standard web service based interfaces, mainly those developed by the Open Geospatial Consortium.

Accounting for uncertainty: all the models and data sources we dealing with have approximations and errors. These mean they are uncertain, so each time we use one of the resources we should account for this uncertainty. If we don't we have no idea of the validity of the output at the end of the chain, and thus no rational way to make decisions.

Put these ideas together and you have UncertWeb.
Q. When does UncertWeb run?
A. UncertWeb started in February 2010 and will run for 3 years until January 2013.
Q. What are the main ouptuts of UncertWeb?
A. UncertWeb will develop the framework to support the "uncertainty enabled model web". This framework will consist of a set of information models and encodings, software to enable the use of these, a workflow discovery and composition system including a client, and possibly most importantly a set of web based tools to support the management of uncertainty by users including tools for elicitation, analysis, transformation and visualisation of uncertainty.
Q. Where are the UncertWeb requirements?
A. During the inception phase of UncertWeb requirements have been captured from the different application domains being considered, as well as the broader requirements for managing uncertainty in distributed, web-based workflows.

D4.1, D5.1, D6.1 and D7.1 (single consolidate document) bring together requirements from the application areas in which the UncertWeb solutions will be deployed and evaluated. In particular the following application domains are considered:
• Biodiversity and climate change (WP4);
• Land-use response to climatic and economic change (WP5);
• Short term uncertainty-enabled forecasts for local air quality (WP6);
• Individual activity in the environment (WP7).
The main aim is to establish the scope of models, input data and output data that will be considered within UncertWeb. Here by scope we mean a number of characteristics such as:
• Types of data being used (vector, gridded, observational, etc)
• Types of values being considered (continuous valued, categorical, binary, etc)
• Spatial and temporal support (domain and resolution) of the data
• How uncertainty is addressed, how important it is
• Physical size / structure of the data payload / time for the model computations
In addition we also explore the actual questions that will be addressed by the model chains constructed within UncertWeb and describe the analysis and post processing that might be employed in each chain. The results are presented in a standard format, describing the models and their inputs and outputs using prescribed tables.

These application domain requirements inform the requirements for UncertML captured in D1.1. These requirements emphasise the need for a coherent conceptual model for uncertainty within UncertWeb. In particular all use cases considered within the project only require probabilistic treatment of uncertainty. While there are alternative models for uncertainty within UncertWeb the scope will be constrained to probabilistic representation of uncertainty. It should be noted that this includes the notion of subjective probability, in the sense of a subjective Bayesian view, where probability is associated with personal belief.

The requirements for the UncertWeb architecture are presented in D2.1. These again emphasise the application domains within UncertWeb, but also consider a wider set of stakeholders such as external users and the existing standards and tools as well as the broader GEOSS context.

D8.1 reviews the requirements for the integration of UncertWeb components. In particular it pays attention to the requirements for information encoding (for spatial, temporal, spatio-temporal and non-spatial data types) and processing (including communication patterns and protocols and security). Within the document a coherent set of profiles are established to facilitate the development of a common UncertWeb framework, allowing the requirements to be realised within the project.

Uncertainty

Q. What is uncertainty?
A. The standard definition says something like:

"The state of not being certain", or "the state of having incomplete knowledge of something of interest". Typically we are always uncertain about the future, and often also about the present (and the past). There are many ways to think about uncertainty, and this has been the subject of intense philosophical debate.

Within UncertWeb the main source of uncertainty can be linked to a lack of knowledge or information. This is typically described as epistemic uncertainty. Within UncertWeb uncertainty is modelled using probabilistic concepts.

Further information can be found at http://understandinguncertainty.org/ or http://en.wikipedia.org/wiki/Uncertainty.
Q. What is probability?
A. Probability is the chance of an event occurring. This has classically been associated with frequentist notions of repeatable experiments. However often events cannot be repeated many times (for example the weather tomorrow will happen only once, but I want to say something about how likely it is that it will rain, given what I know now). Thus Bayesian thinkers tend to provide a broader definition, although this has several flavours and complications. Typically Bayesians consider subjective probability which is often discussed in terms of belief. There are ongoing debates on the merits of the objective and subjective Bayesian views of probability.
Q. Why is UncertWeb only considering probabilistic approaches?
A. UncertWeb considers only probabilistic approaches to uncertainty representation. The main reason for this is that for the use cases we are developing this is the only representation of uncertainty that is required. However, there are also deeper reasons. A probabilistic approach to managing uncertainty predominates in the world of statistics. This is not accidental - Bayesian approaches to probability are natural generalisations of formal logic to reasoning with uncertainty. The mathematical framework provided by probability theory is comprehensive and self consistent. The basic axioms, known as Cox's axioms are very simple, and difficult to argue with.

However the project does not consider deep philosophical issues, and realises that other treatments of uncertainty, such as Bayes Linear approaches and imprecise probability theory provide interesting avenues for further developments in the future, but are beyond the resources and scope of UncertWeb.
Q. Is probability subjective or objective?
A. Historically there has been a war of words and ideas between different ideologies within statistics. This used to consider the merits of frequentist versus Bayesian views on how to conceptualise uncertainty. UncertWeb remains neutral to these philosophical matters, focussing only on a probabilistic approach to representing uncertainty.

However, it should be acknowledged that there are different views on how to interpret and assess probabilities. Even within the Bayesian viewpoint there are objective Bayesians, who essentially identify with a notion that probability is something that can be defined outside of the self, and subjective Bayesians who believe that probability can also be the construct of an individual and equate probability to a degree of belief. UncertWeb accepts all these views, however some tools, for example the Elicitator will appeal more to those who are comfortable with the subjective Bayesian view of probability.

UncertML

Q. Is there an UncertML discussion list?
A. Yes there is - it can be found at http://uncertml.forum.52north.org/ - plese feel free to join and add your comments, requests and questions there.
| Contact us | Advisory panel | Sponsor | Disclaimer | Member area |

The UncertWeb project has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° [248488].