Transcript Swedish catchment

BACKGROUND SMHI
Statistical Downscaling
work
• Large scale difference between NWP models and
Hydrological models
• A way to address this is to statistically downscale the
NWP precipitation forecasts through Statistical PostProcessing
POTENTIAL OF STATISTICAL DOWNSCALING
Improvement of QPF forecast skill due to Statistical PostProcessing
SMHI application of
Model Output Statistics (MOS)
•
•
•
•
•
•
Predictors taken from ERA 40 forecasts over Torpshammar area
Predictands taken from SMHI climate data base
Multiple linear regression, Forward selection, Stepwise
Predictors are only taken from a column at and above the station
Binary as well as continuous Predictors / Predictand
Different MOS-Equations for each
Variable
Cycle [ 00, 06, 12, 18 UTC ]
Forecast lead time
Season
Station
Torpshammar catchment
750 m
700 m
650 m
600 m
550 m
500 m
450 m
400 m
350 m
300 m
250 m
200 m
150 m
100 m
50 m
0m
0
10 20 30 40 50 60 70 80 90 100 km
Catchment area:
4300 km2
Mean Annual Precipitation: 700 mm
Mean altitude:
340 m

Location of rainfall
stations
PREDICTAND
The predictand actually used is a set of Binary Precipitation accumulation amount predictands.
They represent the Occurance (1) or Non-Occurance (0) of events which are defined to be the accumulation of
PREC ≥ PRECC
Binary Precipitation
Predictand Number
PRECC
[mm]
EXAMPLE
PREC = 8 mm
1
0.3
1
2
1
1
3
2
1
4
5
1
5
10
0
6
25
0
The predictands are treated cumulatively, i e one observation may be a simultaneous “occurance” of several “events”
Regression technique yields a set of forecast equations for the probability of accumulated precipitation in 24 hours ≥ PRECC
To promote consistency of the forecasts the Regression Equations are developed simultaneously for all predictands.
The same predictors are offered and are, if selected by any one predictand, used for all predictands.
This is referred to as REEP = Regression Estimation of Event Probabilities