Forecasting flash floods using data-based mechanistic models and NORA radar rainfall forecasts

Figures & data

Table 1 Occurrence of NORA events summarized by year and month. Zero values are left blank

Month	Year						Total
	2004	2005	2006	2007	2008	2009
January					1		1
February
March		1		1			2
April	2	1	1		4	2	10
May	1	5	1	3			10
June		3		2	1	2	8
July		2		1	1		4
August	2	2	1	2	2	4	13
September	1	2	2	1	2	1	9
October	2	1	1			2	6
November	1		1	1	2	1	6
December			1			1	2
Total	9	17	8	11	13	13	71

Fig. 1 Summary of the performance of the hourly accumulated NORA precipitation forecasts against those generated by persistence and COSMO2 at lead-times of up to 6 h. For a definition of the verification scores presented, see Wilks (1995) and Panziera et al. (2011). Lead-time is given as the start of the accumulation period.

Fig. 2 One year of calibration data showing the observed data (•) and simulated DBM output (––––).

Table 2 Summary statistics of forecasting performance during calibration of the DBM model using data assimilation with the assumption that future rainfall is known. Values in parentheses represent the statistics computed only when discharge >75 m³/s. Results for the model run without data assimilation are shown for comparison

Lead time	RMSE	Bias	Kendall tau
(h)	(m^3/s)	(m^3/s)
No data assimilation	7.94 (48.6)	1.56 (10.7)	0.76 (0.55)
1	2.92 (19.6)	0.02 (0.45)	0.98 (0.83)
2	4.59 (30.9)	0.04 (1.04)	0.96 (0.73)
3	5.74 (38.2)	0.05 (1.50)	0.95 (0.66)
4	6.64(44.0)	0.06 (1.80)	0.93 (0.61)
5	6.79 (44.6)	0.30 (5.26)	0.92 (0.60)
6	7.27 (47.6)	0.35 (6.14)	0.91 (0.58)

Table 3 Summary statistics of forecasting performance during validation of the DBM using data assimilation with the assumption that future rainfall is known. Values in parentheses represent the statistics computed only when discharge >75 m³/s

Lead-time	RMSE	Bias	Kendall tau
(h)	(m^3/s)	(m^3/s)
1	3.32 (29.1)	0.02 (0.02)	0.98 (0.82)
2	5.46 (47.9)	0.04 (0.24)	0.96 (0.71)
3	7.10 (62.2)	0.05 (0.69)	0.94 (0.64)
4	8.22 (71.8)	0.07 (1.21)	0.93 (0.60)
5	7.91 (67.8)	0.32 (9.75)	0.92 (0.57)
6	8.23 (69.9)	0.38 (11.9)	0.91 (0.54)

Fig. 3 Calibration forecast (made using observed rainfall) for a large event in the autumn of 1993. Observed data (•) and forecasts using data assimilation at lead-times of 1 h (–––), 3 h (......) and 5 h (- - - -) are shown together with the model output using no data assimilation (- . - . -).

Table 4 Summary statistics of forecasting performance using predicted precipitation during time periods where NORA forecasts are available. Data for the first 2 h lead-time where NORA forecasts are not used are shown for completeness. Performance when it is assumed the observed rainfalls are known is shown in parentheses

Lead-time	RMSE	Bias	Kendall tau
(h)	(m^3/s)	(m^3/s)
1	3.33	0.02	0.98
2	5.46	0.04	0.96
3	21.9 (21.7)	−1.15 (0.07)	0.86 (0.87)
4	28.0 (25.0)	−2.33 (0.10)	0.81 (0.84)
5	29.1 (23.9)	−2.83 (0.89)	0.77 (0.82)
6	34.5 (24.8)	−3.96 (1.03)	0.74 (0.81)

Fig. 4 Scatter plots of the observed discharge vs that forecast with lead-times of (a) and (b) 1 h; and (c) and (d) 2 h. For each lead-time, the dataset is separated into forced (left-hand panes) and unforced (right-hand panes) time- steps to show the difference in distribution.

Fig. 7 Forecasts for lead times of up 6 h issued for the largest of the validation events in 2009. The shaded area represents the 95% prediction interval, solid lines the deterministic model forecast and points the observed discharge.

Fig. 5 Scatter plots of against used in the estimation of non-parametric error distribution which is represented by the 95% prediction interval (shaded area) for various forecast lead-times.

Fig. 6 Scatter plots of against for the validation storms in 2009 overlaying the non-parametric error distribution which is represented by the 95% prediction interval (shaded area) for various forecast lead-times.

Wilks, D.S., 1995. Statistical methods in the atmospheric sciences. New York: Academic Press.

 Google Scholar

Panziera, L., et al., 2011. NORA—Nowcasting of Orographic Rainfall by means of Analogues. Quarterly Journal of the Royal Meteorological Society, 137, 2106–2123.

 Web of Science ®Google Scholar