Natural Risk Management in a changing climate: Experiences in Adaptation Strategies from some European Projekts Milano - December 14 th, 2011 FLORA: FLood estimation and forecast in complex Orographic areas for Risk mitigation in the Alpine space Dr. Ing. Andrea Salvetti - Ufficio dei corsi d acqua Dip.Territorio Cantone Ticino - CH with essential contributions of: Dr. Massimiliano Zappa and co-workers, Swiss Federal Institute for Forest, Snow and Landscape Research Dr. Urs Germann and co-workers, MeteoSwiss Locarno Monti - CH Ing. Secondo Barbero and co-workers, Arpa Piemonte - I
Project Rationale and Relevance River floods are the most common natural disaster in Europe. Several Climate Change impact studies show that in the coming decades global warming is projected to increase the magnitude and frequency of floods in the Alpine region. Floods can result in huge economic losses due to damage to infrastructure, property and agricultural land (and indirect losses ). Estimation of design floods in ungauged catchments is frequently required in hydrological practice and is of great economic significance. A warning system can empower individuals and communities to respond timely and appropriately to hazards in order to reduce the risk of death, injury, property loss and damage. Introduction
FLORA Project: Key Tasks Forecast of precipitation with high-resolution observations assimilation Estimation of rainfall fields from radar observations Action 1 Action 2 Action 4 Action 3 Regional statistical estimation of floods Flash flood nowcasting by means of radar ensemble Introduction
Action 1: activities Rationale Final Product 1.1 Exchange of observed data between ARPA Piemonte and MeteoSwiss Understanding and the description of the phenomena Database creation 1.2 Quantitative Precipitation Forecast (QPF) verification of high-resolution numerical models in use at ARPA Piemonte and MeteoSwiss (COSMO Model) 1.3 Sensitivity study on the parameters assimilated into the meteorological models 1.4 Implementation of new post-processing methodologies for the enhancement of QPF When and where do models need major improvements? Improving the initial conditions by assimilating surface observed parameters Correction of model output with statistical techniques Guidelines for the definition of a verification method and for the interpretation of the direct model output Implementation of a new algorithm for the complete balance of soil-atmosphere interaction (T2m assimilation) Implementation of linear regression techniques with a real improvement of QPF Introduction Action 1
Action 2: activities Final Products 2.1 Off-line estimation of rainfall fields Revision of method for rainfall estimation Characterization of data quality 2.2 Real-time estimation of rainfall fields Pilot product of rainfall uncertainties Errors on Swiss and Italian composite Introduction Action 1 Action 2
Action 2: activities Investigation on radar data quality using visibility maps and ground truth (279 rain gauges) Period: 16/09/2009-16/02/2011 Overall radar rainfall correction based on excellent visibility for mean bias assessment. The area used for the correction ranges to 50 km far from radar and gauges altitude less than 1000 m a.s.l. Absolute error: ass P F P con F 1. 71 pluvio radar Relative error: rel P ass pluvio Introduction Action 1 Action 2
Action 2: activities Investigation on radar data quality using visibility maps and ground truth (279 rain gauges) Period: 16/09/2009-16/02/2011 Estimation of relative error: ε rel = 0.05+ 0.136 1000 quota RA Red: acceptable relative errors: quota RA < 3000m a.s.l. Radar accumulation over one year Introduction Action 1 Action 2
Action 4: activities Rationale Final Product 4.1 Knowledge updating of flood data Collection of information Database creation and report 4.2 Application and extensions of statistical procedures for flood risk assessment Improved use of observed data for both at-site and regional flood estimation Statistical analysis 4.3 Development of statistical methods for design flood estimation Flood estimation along the river network Statistical analysis 4.4 Increasing the robustness of estimation methods Statistical evaluation of estimation uncertainty Statistical analysis Introduction Action 1 Action 2 Action 3 Action 4
Action 3 Rationale of Action 3 Unlike riverine floods, flash floods are rapid-onset events that are often unpredictable, or predictable with little lead time Very often flash floods carry high sediment and debris loads very high hydraulic force and erosive power The affected areas are alluvial fans, mostly local extent Potential mitigation measures are: early warning systems community preparedness and awareness appropriate emergency measures Introduction Action 1 Action 2 Action 3
Action 3: activities Rationale Final Product 3.1 Exchange of observed data (at-site measures and radar estimate) Description of the phenomena Datamanagement and Processing 3.2 Radar ensemble verification and coupling with COSMO 2 and COSMO 7 weather forecast Performance assessment of the proposed procedure Continuously upgraded meteorological forecast input 3.3 Coupling of radar ensemble with semi-distributed hydrological model PREVAH Probabilistic hydrological nowcasting Meteo-hydrological nowcasting chain 3.4 Results analysis and on-line implementation Verification of the whole procedure Operative Early Warning System Introduction Action 1 Action 2 Action 3 Action 4
Accuracy vs. Lead time Action 3 Accuracy Alarm at Station Nowcasting COSMO2 COSMO7 COSMO-LEPS warning lead time Nowcasting: Forecasts for the following few hours via the analysis and extrapolation of weather systems as observed on radar and in situ sensors, and via the application of short-range numerical prediction. Radar is a particularly valuable tool since it provides the size, shape, intensity, speed and direction of movement of individual storms on a continuous basis. This ability to forecast precipitation amount is particularly useful for the development of early warning systems for intense convective systems, which often result in a lot of damage. Introduction Action 1 Action 2 Action 3 Action 4
Why not deterministic radar estimate? Solution 1: Detailed information on error sources Sophisticated error correction algorithms Introduction Action 1 Action 2 Action 3 Action 4
Action 3 Introduction Action 1 Action 2 Action 3 Action 4
What s REAL Radar Ensemble generator for usage in the Alps using LU decomposition Deterministic Radar QPE (Quantitative Prec. Estimate) Stochastic Perturbation Radar Ensemble REAL consisting of 25 Members
Why such an effort?????? Radar observations have errors and uncertainties, especially in mountaineous regions (scatter, shading )??? Ensembles account for those uncertainties Ensemble Straight foreward way to propagate this uncertainty through a hydro-meteorological model chain
Solution 2: Generate an ensemble of radar fields Introduction Action 1 Action 2 Action 3 Action 4
Recipe Introduction Action 1 Action 2 Action 3 Action 4
REAL hydrological forecasting Hydrological Model REAL PREVAH (Precipitation Runoff Evapotranspiration HRU related model) Input Parameters (hourly): air temperature, water vapour pressure, global radiation, wind speed, sunshine duration and precipitation (e.g. REAL). PREVAH Introduction Action 1 Action 2 Action 3 Action 4
REAL catchments Introduction Action 1 Action 2 Action 3 Action 4
Introduction Action 1 Action 2 Action 3 Action 4
REAL catchments Introduction Action 1 Action 2 Action 3 Action 4
First evaluation of results Accuracy of radar-driven runoff is comparable to that of rain-gauge-driven runoff. We expect radar to outperform rain-gauges for convective situations with strong spatial rainfall variability. The scatter between runoff and runoff driven by radar ensemble (F t,i ) is only slightly larger than the scatter between observed runoff and runoff drived by deterministic radar field (R t ) In terms of runoff the ensemble generator does not overstate the uncertainty that is already present in the deterministic component. F t,i has same bias as R t The generator is bias-free in terms of water amount
Communication of probabilistic forecast The hydrological community in generally moving towards the use of probabilistic estimates of streamflow (Ensemble Streamflow Prediction ESP) Model evaluation is still performed using standard deterministic measures Goals of forecast verification: evaluating the value and the skills of predictions, performing quality control on the forecast and investigating the cause(s) of prediction errors Consistency vs. Uncertainty Introduction Action 1 Action 2 Action 3 Action 4
Runoff Communication of probabilistic forecast Visualisation of Percentile (Exceedance probabilities) Time Probability that the discharge or water level at a gauging level will be reached or exceeded Introduction Action 1 Action 2 Action 3 Action 4
Communication of probabilistic forecast Visualisation of probabilities in graphical form Warning level 4 Warning level 3 Warning level 2 1. Step 2. Step 3. Step 4. Step Warning level 1 Exceedance probability Introduction Action 1 Action 2 Action 3 Action 4
Communication of probabilistic forecast Visualisation of probabilities in graphical form Introduction Action 1 Action 2 Action 3 Action 4
Conclusions - Outlook Flora research activities will officially continue until June 2012 The final conference of the project will take place in spring 2012 The innovative results of this project will we (and are already) applied in the day-to-day operational activities of each agency (ARPA Piemonte, Canton Ticino,...) Further collaboration for integrated research in the field of applied meteorology, hydrology, and flood forecast is needed. Introduction Action 1 Action 2 Action 3 Action 4 Conclusions