Detection and attribution of changes in flood risk: ideas from a urban catchment case study

Presentation given at the Symposium on Regional Floods Effects of Changes in the River System in Vienna, Austria - 13 October 2015
Related paper: http://onlinelibrary.wiley.com/doi/10.1002/2015WR017065/full

Transcript

1. Detection and attribution of changes in flood risk: ideas from a urban catchment case study Ilaria Prosdocimi - ilapro@ceh.ac.uk Centre for Ecology & Hydrology – UK Joint work with Thomas Kjeldsen and James Miller FloodChange Workshop Vienna, October 2015
2. Non-stationarity detection (and attribution) • Typically: take some (annual) series and test the correlation against time • We can do better than that! (Merz B. et al., 2012) • Different sources: natural and anthropogenic Merz B., Vorogushyn, Uhlemann, Delgado, Hundecha (2012) More efforts and scientific rigour are needed to attribute trends in flood time series. HESS, (16), 1379-1387
3. Quantifying non-stationarity Available covariates (evidence of consistency): • Time (year) • Catchment averaged daily rainfall – 99th percentile • Urban Extent (URBEXT) • Large natural variability and many potential confounders
4. Changes in Urbanization Extent Miller, Grebby (2014). Mapping long-term temporal change in imperviousness using topographic maps, Int. J. Appl. Earth Obs.
5. Quantifying non-stationarity Paired catchment approach (evidence of inconsistency): • Identify nearby hydrologically similar catchment Available covariates (evidence of consistency): • Time (year) • Catchment averaged daily rainfall – 99th percentile • Urban Extent (URBEXT) • Large natural variability and many potential confounders
6. Paired catchments
7. Quantifying non-stationarity Paired catchment approach (evidence of inconsistency): • Identify nearby hydrologically similar catchment Available covariates (evidence of consistency): • Time (year) • Catchment averaged daily rainfall – 99th percentile • Urban Extent (URBEXT) An adequate statistical model (provision of confidence): • Point process representation of POT • Large natural variability and many potential confounders
8. Point processes for POT nu = 3 • Model frequency and magnitude simultaneously • Easy to incorporate covariates • Same parameters as GEV - μ, σ, ξ (yi -u) nu = 0 nu = 4
9. Peak Flow data • Block maxima • POT • Annual • Seasonal
10. Urban cover impact - models Available response variable: • Annual and seasonal maxima: Q ~ GEV(μ, σ, ξ) • Annual and seasonal POT: Y ~ PP(μ, σ, ξ) μ = β0 +β1 rain μ = β0 +β2 time μ = β0 +β3 urbext μ = β0 μ = β0 +β1 rain+β2 time μ = β0 +β1 rain+β3 urbext
11. Urban/ Case Rural / Control
12. Urban/ Case Rural / Control
13. Urban cover impact - conclusions • Increased flood risk can be attributed to increased urbanisation levels • Point process representation of POT is useful • Method is demanding in terms of data availability and processing (attribution is difficult!)
14. Want to know more? More details (open access): Prosdocimi, I., Kjeldsen, T. R. and Miller, J. D. (2015), Detection and attribution of urbanization effect on flood extremes using nonstationary flood frequency models. Water Resour. Res.. Accepted. doi:10.1002/2015WR017065 ilapro@ceh.ac.uk @ilapros