Evaluation of Experimental Streamflow Forecast from The Global Ensemble Forecast Sysyem GEFS

1
Evaluation of Experimental Streamflow Forecast
from The Global Ensemble Forecast Sysyem (GEFS)

• Dingchen Hou, Kenneth Mitchell, Zoltan Toth,
• Dag Lohmann and Helin Wei
• Environmental Modeling Center/NCEP/NOAA, USA
• 5200 Auth Road Camp Springs, MD 20746
• EMC/NCEP/NOAA and SAIC
• Risk Management Solution Ltd. UK
• Acknowledgement
• Dongjun Seo, Pedro Restrepo and John Schaake,
  OHD/NOAA
• George Gayno, Yuejian Zhu, Bo Cui, Jesse Meng,
• Youlong Xia and Micheal Ek EMC/NCEP/NOAA
• 4th NAEFS Workshop,
• October 6-8, 2008, Camp Springs, MD, United States

2
STREAMFLOW FORECAST SCHEME
Atmospheric Model
Statistical Correction of Precipitation
Coupled Atmosphere Land Surface Model
Precipitation
Fluxes

Observed Precipitation
Land Surface Model
Hydrological model
Runoff
Hydrological model forced with observed
precipitation simulates streamflow
River Routing Model
Streamflow
Streamflow Analysis
Post Processor
Streamflow Forecast
3
APPROACH

• STUDY SKILL OF RIVER FLOW ENSEMBLE THROUGH
• PERFECT LAND/RIVER MODEL SCENARIO
• Forced with atmospheric ensemble
• NCEP operational global ensemble (1x1 lat/lon
  resolution, out to 16 days)
• 10 perturbed equivalent higher resolution
  control members
• With perfect land/river model (1/8x1/8 lat/lon
  resolution)
• NOAH land model (Mitchell et al)
• Linear distributed river routing model (Lohman et
  al)
• D8 river flow direction mask
• Tested over continental US
• With perfect river initial state
• Derived/simulated by forcing same perfect model
  with observed precipitation
• NOAH Land surface Data Assimilation System
  (NLDAS)
• Observed precipitation, other variables analyzed
• River model
• Verified against simulated streamflow (analysis)

4
Offline streamflow 2-year simulation
Noah
other land models
East Fork of White River at Columbus, IN
5
STREAMFLOW FORECAST EXAMPLE12-DAY LEAD-TIME
(APRIL 1, 2006)
Analysis (NLDAS)
Ensemble Mean
Ensemble mean similar to analysis Error 10
of flow Ensemble spread comparable to error
in ensemble mean
ErrorEns. Mean - analysis
Ensemble Spread
6
Merrimack-Concord River Lowell, MA A Medium Sized
Basin Major Problem Underdispersive ensemble in
grid and subgrid scale precipitation. Mid-May
Flood Event Compared with the Early-April event,
the Mid-May event is harder for the model to
simulate. Nevertheless, the ensemble shows some
skill indicating a major event with 10 day lead,
various amplitude and timing. Early April Major
event forecast despite short range over- forecast
----- GEFS members ----- GEFS ens. mean -----
GEFS control ----- GFS high resolution ----- NLDAS
May 4th
0 2 4 6
8 10 12 14
16
Lead Time (days)
0 2 4 6
8 10 12 14
16
Lead Time (days)
April 1st
7
FORECAST ANALYSIS TIME SERIES
Positive Correlation between Forecasts and
Analysis for all Lead times
Lower Mississippi River Very Large Basin
Trend is predicted well even at 15-day lead
15-day lead
May 2006 New England Flood correctly predicted,
some minor events indicated 5 days in advance
Merrimack- Concord River, Lowell, MA Medium Basin
----- GEFS members ----- GEFS ens. mean -----
GEFS control ----- GFS high resolution -----
NLDAS Analysis
5-day lead
8
USGS River Station
Mean Streamflow Category NEH
Nehalem River, FOSS OR
40.3 9 POT
Potomac River, Washington DC
110.7 13
MER Merrimack-Concord River,
Lowell MA 499.4
17 MIS Mississippi
River, Vicksburg MS 14788.3
20
9
Table 1. The 20 categories of grid points,
grouped based on the analysed streamflow avegaed
over the period of experiment.

• Mean Streamflow (m3 s-1) Number of Grid Points
  Percentage of Grid Points
• 0.0 1.0 I
  71473 68.77
• 1.0 10.0
  18481 17.78
• 10.0 15.0
  2454 2.36
• 15.0 20.0
  1558 1.50
• 20.0 25.0
  985 0.95
• 25.0 30.0
  800 0.77
• 30.0 35.0
  621 0.60
• 35.0 40.0
  506 0.49
• 40.0 45.0
  433 0.42
• 45.0 55.0
  723 0.70
• 55.0 70.0
  665 0.64
• 70.0 90.0
  667 0.64
• 90.0 120.0
  593 0.57
• 120. 150.0
  533 0.51
• 150.0 200.0
  634 0.61
• 200.0 300.0
  681 0.66
• 300.0 500.0
  592 0.57
• 500.0 1000.0
  557 0.54

10
TEMPORAL CORRELATION BETWEEN FORECASTS AND
ANALYSES
Corr., GEFS Control Fcst
Nehalem River, FOSS OR A Small Basin in the
West High Corr. for all lead times
0.5
Potomac River, Washington DC, Medium Basin Corr.
close to 1 for 1-2 day lead, Decreasing to 0 at
day 10
Corr., GEFS Ens. Mean Fcst
----- GEFS members ----- Mean of GEFS mem. -----
GEFS control ----- GFS high resolution ----- GEFS
Ens. Mean
0.0
11
TEMPORAL CORRELATION VS. LEAD TIME AND VOLUME
Ranges m3/s gt2000 1000-2000 500-1000 300-500 200
-300 150-200 120-150 90-120 70-90 55-70 45-55 40-4
5 35-40 30-35 25-30 20-25 15-20 10-15 1-10 0-1
CTL
GFS-CTL Difference
Predictability out to 15 day
Forcing from higher resolution control does not
help
Ranges m3/s gt2000 1000-2000 500-1000 300-500 200
-300 150-200 120-150 90-120 70-90 55-70 45-55 40-4
5 35-40 30-35 25-30 20-25 15-20 10-15 1-10 0-1
ENSEMBLE Mean -CTL Difference
Mean Score of GEFS Members -CTL Difference
Positive effect of ensemble
12
EFFECT OF INSUFFICIENT SPREAD IN FORCING ON CRPSS
With Bias-correction
Without Bias-correction
gt2000m3/s 1000-2000
gt2000m3/s 1000-2000
500-1000 300-500
500-1000 300-500
0
0

• Observations
• Positive skill for the large river basins in raw
  forecast.
• Improvement due to bias-correction.
• Positive skill for (almost) all river basins
  after bias correction
• lower skill for 3-7 day lead, small and medium
  basins.

200-300 70-90 35-45 15-20
Ranges (m3/s) gt2000m, 1000-2000, 500-1000,
300-500, 200-300, 70-90, 35-45, 15-20
200-300 70-90 35-45 15-20
Medium-small rivers negatively affected in 2-12
days range
Negative effect remains even after 1st moment
bias correction gt Forcing underdispersive on
small scales
13
INSUFFICIENT SPREAD IN FORCING Very small
spread / Large error in small basin, short range
forecast example Typical for all
cases Related to dip in predictability even
after 1st moment bias correction
----- GEFS members ----- GEFS ens. mean -----
GEFS control ----- GFS high resolution ----- NLDAS
Single Case Ensemble April 1st, 2006
Average over 60 cases Ordered Ensemble
May 4th
14
Summary

• The results from the experiment suggest that the
  coupled system is capable to generate useful
  gridded streamflow forecast.
• The uncertainties represented in the initial
  conditions of the atmospheric model helps to
  improve streamflow forecast.
• The expected forecast skill increases with
  increasing size of the river basin or streamflow
  intensity.
• With the current GEFS system, positive skill in
  short range (1 to 3 days) predictions can be
  expected for all significant river basins, and
  for the major rivers with mean streamflow over
  500 m3 s-1, significant forecast skill can be
  expected from extended range (the second week)
  predictions.
• Forecast can be improved by bias correction,
  increase ensemble spread and downscaling of
  precipitation (runoff).