Input data for spatial environmental models may have been measured in the field or laboratory, spatially interpolated, derived from remotely sensed imagery or obtained from expert elicitation. In all these cases errors are introduced. Although users may be aware that errors propagate through their models, they rarely pay attention to this problem. However, the accuracy of the model output may be insufficient for the intended use, causing inaccurate model results, wrong conclusions and poor decisions. The purpose of this course is to familiarize participants with statistical methods to analyse uncertainty propagation in spatial environmental modelling, such that they can apply these methods to their own models and data. It also teaches methods that quantify the contribution of individual uncertainty sources and statistical validation methods to assess the accuracy of spatial model outputs with independently sampled data. Quantification of model parameter uncertainty is covered using Bayesian calibration techniques. The methodology is illustrated with real-world examples. Computer practicals make use of the R language for statistical computing.
This course differs from the “Statistical Uncertainty Analysis of Dynamic Models” (SUADM) course, in that it:
- focuses on uncertainty propagation in spatial models, while SUADM concentrates on uncertainty analysis of dynamic models;
- uses basic to intermediate statistical approaches and graphical tools to analyse uncertainty and uncertainty propagation, while SUADM uses more advanced statistical approaches;
- dedicates a full day to statistical validation of outputs of spatial models using spatial sampling theory, while SUADM draws specific attention to stochastic sensitivity analysis.
Day 1, morning: Course overview; lectures and exercises on probabilistic modelling of uncertainty; lecture and exercises on the Taylor series method.
Day 1, afternoon: Computer practical on the Taylor series method.
Day 1, evening: Group dinner in Wageningen centre.
Day 2, morning: Lecture and exercises on the Monte Carlo method and the quantification of uncertainty source contributions.
Day 2, afternoon: Computer practical on the Monte Carlo method and analysis of uncertainty source contributions.
Day 3, morning: Lecture and exercises on Bayesian calibration for the quantification of model structural and parameter uncertainty.
Day 3, afternoon: Computer practical on Bayesian calibration.
Day 4, morning: Lecture and exercises on statistical validation and cross-validation, including spatial sampling for validation and reliability plots.
Day 4, afternoon: Computer practical on statistical validation and cross-validation.
Day 5, morning: Lecture, exercises and computer practical on uncertainty quantification of spatial averages.
Day 5, afternoon: Finish computer practical; course evaluation; uncertainty game; drinks and snacks.
The last 45 minutes of each afternoon (except Friday) are reserved to either continue the scheduled computer practical or apply the methods learnt to your own models and data.
On Monday evening there is a group dinner in town (cost included in course fee).
- Prof. Dr Gerard Heuvelink (Soil Geography and Landscape Group, Wageningen University)
- Prof. Dr Sytze de Bruin (Laboratory of Geo-information Science and Remote Sensing, Wageningen University)
- Dr Erick Chacón-Montalván (Laboratory of Geo-information Science and Remote Sensing, Wageningen University)
| Target Group | The course is aimed at PhD candidates and other academics working with spatial models who want to trace how errors in model inputs and parameters propagate to model outputs and want to learn how to evaluate map accuracy with independent data. |
| Group Size | Min. 15, max. 25 participants |
| Course duration | 5 days |
| Prior knowledge | Intermediate knowledge of statistics, geo-information science and spatial modelling. Familiarity with the R programming language is highly recommended but not required. |
| Fees1 | REGULAR FEE |
| PE&RC/WIMEK/EPS/WASS/VLAG/WIAS PhD candidates with approved TSP and WU EngD candidates | € 385,- |
| PE&RC postdocs and staff | € 770,- |
| All other academic participants | € 810,- |
| Non-academic participants | € 1580,- |
1 The course fee includes a reader, coffee/tea, and lunches and one course dinner on Monday. It does not include accommodation.
Note:
PE&RC Cancellation Conditions
IMPORTANT: ALWAYS read the Cancellation conditions for PE&RC courses and activities.
- PE&RC Office
Email: office.pe@wur.nl
