Presentation is loading. Please wait.

Presentation is loading. Please wait.

What forecast users might expect: an issue of forecast performance

Similar presentations


Presentation on theme: "What forecast users might expect: an issue of forecast performance"— Presentation transcript:

1 What forecast users might expect: an issue of forecast performance
Tomas Vlasak, Radek Cekal & Jan Danhelka Czech Hydrometeorological Institute Czech republic

2

3 Hydrological service in the Czech republic
Deterministic forecast (> 100 forecasting points 48hours / 1hour step) Flash flood guidance FFG-CZ Short-range probabilistic forecast (based on ensemble QPF) Medium-range probabilistic forecast (based on historical analogs)

4 Hydrological forecast evaluation
Feedback for FORECASTERS: model parameters or model structure data input human impact strategy Feedback for forecast USERS: realize forecast uncertainty estimate forecast uncertainty avoid false expectation

5 Evaluation methods Comprehensible not only for expert.
Simplification to 1 indicator: threshold exceeding YES/NO total flow volume discharge maximum Evaluation of whole of forecasting process without prior differentiation of the source of uncertainty.

6 Evaluation methods – forecasts selection
Target to flood forecasts Selection conditions: 1) forecasted or observed discharge exceeded flood threshold 2) last measured discharge was smaller then threshold

7 Categorical evaluation
Observed YES NO Forecast HIT FALSE ALARM MISS CORRECT NEGATIVE portion of these category in different aspect: Threshold Basin area Season Year Forecasting office categorical statistics Hit Rate False Alarm Ratio Frequency Bias Critical Success Index ..... HIT HIT HIT MISS FALSE ALARM FALSE ALARM

8 low FLOOD EXTREMITY high
Categorical evaluation – RESULTS – different thresholds Decrease of portion of HITS with increasing of flood extremity is small MISS forecasts prevail to FALSE ALARM low FLOOD EXTREMITY high

9 Categorical evaluation – RESULTS – basin area
Forecasts for big rivers are more successful uncertainty of QPF produce FALSE ALARMS and uncertainty of hydrological modeling MISSES

10 Categorical evaluation – RESULTS – lead time
Decrease of HIT rate is significant to first 24 hours of lead time. The most of forecasting points delimit basins with lag time shorter then 24h

11 Categorical evaluation – RESULTS – lead time
Decrease of HIT rate is significant to first 24 hours of lead time. The most of forecasting points delimit basins with lag time shorter then 24h

12 Categorical evaluation – RESULTS – lead time
Decrease of HIT rate is significant to first 24 hours of lead time. The most of forecasting points delimit basins with lag time shorter then 24h

13 Categorical evaluation – RESULTS - years
Without significant trend

14 Categorical evaluation – RESULTS – years –regional office Plzeň
change of hydrological model performance after model recalibration in 2010

15 Categorical evaluation – RESULTS

16 Categorical evaluation - RESULTS
Forecasting point on big river has more successful forecasts

17 Predominance of FALSE ALARM category
Categorical evaluation - RESULTS Influence of different forecaster‘s strategy Predominance of MISS category Predominance of FALSE ALARM category

18

19

20

21

22

23 THANK YOU


Download ppt "What forecast users might expect: an issue of forecast performance"

Similar presentations


Ads by Google