Hydro Group Projects Deliverables to RFCs

1
Development of Gridded QPE Datasets for
Mountainous Area Distributed Hydrologic
Modeling Mike Smith1, Feng Ding1, 2, Zhengtao
Cui1, 3, Victor Koren1, Naoki Mizukami1, 3, Ziya
Zhang1, 4, Brian Cosgrove1, David Kitzmiller1,
and John Schaake1,5 1Office of Hydrologic
Development, National Weather Service National
Oceanic and Atmospheric Administration 2Wiley
Information Systems Group 3MHW 4University
Corporation for Atmospheric Research 5Riverside
Technology, Inc.
2
Overview

• Purpose
• Methodology
• Data QC Issues
• Results
• Conclusions

3
Purpose

• Develop and test a method to generate gridded
  gauge-only quantitative precipitation estimates
  (QPE) to support NWS RD and operational river
  forecasting
• Leverage RFC tools and data
• Multi-year duration
• Hourly time step
• 4km scale
• Data QC

4
Methodology for Gauge-Only Gridded QPE

• Data Analysis
• 1. Check data consistency double mass analysis
• 2. Generate monthly station means
• 3. Estimate missing data using station means
• Disaggregate all daily data to hourly values
• Use surrounding hourly stations
• Identify values that cant be disaggregated
• Manual QC Fix non-disaggregated values
• Uniformly distribute remaining daily values

SNOTEL Daily

• Generate QPE Grids
• - Use NWS Multi-Sensor Precip. Estimator (MPE)
• Gauge-only option
• Uses PRISM monthly climatology grids
• Uses single optimal estimation (Seo et al.,
  1998, J. Hydrology)

Hourly Point Time Series
5
Comptonville
Methodology 2
North Fork American River
Bowman Dam 67.5
N. Bloomfield 54.6
Ind. Cr. 33.8
Deer Cr. Forebay 72.6
Ind. Lake 47
Ind. Camp 34.67
Lake Spaulding 75.6
Blue Canyon 64
Grass Valley
Sagehen Cr. 32.5
CSS Lab 70.7
Donner 38.9
Gold Run 55.3
Colfax 48.3
Truckee 33.1
Soda Springs 60.7
Truckee 2 34.8
Iowa Hill 59.5
Squaw Valley 69.4
Forest Hill 55.6
Hell Hole 47
Ward Cr. 70.7
Georgetown 54.5
Auburn 37
Blodget Ex. Forest 64
Robs Peak 56.3
Legend
NCDC Hourly
NCDC Daily
CSS Lab
SNOTEL
Donner
Soda Springs
20K30
48332
42467
6
QPE Derivation North Fork American River
Methodology 3

• Generate hourly 4km QPE grids 1980 2006
• Use PRISM 1961-1990 gridded monthly climatology
• Based on 36 NCDC and SNOTEL stations
• Three cases (227,760 grids each case!)
• No correction of non-distributed daily
  observations (312 cases gt 0.5 in)
• Correction of non-distributed daily observations
  and other errors
• Repeat No. 2 with 1971-2000 PRISM climatology
• Hydrologic analysis
• Run distributed model for 1988 to 2006
• Generate hourly streamflow simulation for each
  case
• Compute statistics compared to observed
  streamflow
• Water balance analysis

7
Example of Data Errors
Data QC Issues 1
Missing Flags Foresthill changed from zero to
-998 to agree with Georgetown
Missing accumulation wrongly coded as -999
in data file should be -998
8
Impact of Data Errors on Hourly Gridded QPE
Non-disaggregated daily value at Lake Spaulding
station
Max grid value 4.59 in
00Z 1/22/2000
Snotel
D
Daily
H
Hourly
9
Results 1
Distributed Model Hourly Streamflow Simulation
Statistics Compared to Observed Flow 10/1988
9/2006
Case Bias Hourly RMS Error (cms) Hourly Modified Correlation Coefficient
1. No data QC 61-90 PRISM 8.2 17.3 0.90
2. Data QC 61-90 PRISM 6.2 16.9 0.88
3. Data QC 71-00 PRISM 3.1 16.0 0.89
10
Results 2
Accumulated Streamflow Simulation Error, mm
Monthly Cumulative Error, mm
11
Results 3
Jan 22, 2000 4.59 in
Hydrographs for 3 Cases
1. No Data QC 61-90 PRISM
2. Data QC 61-90 PRISM
3. Data QC 71-00 PRISM
Observed Flow
Time January 16-30, 2000
12
Results 4
Water Balance Analysis
13
Conclusions

• Methodology is sound
• Hourly time step simulations require intensive
  data QC
• Data errors not readily seen in streamflow
  simulation statistics
• Automated procedure to correct wrong data flags
  would streamline the process

14
Thank you!
15
Extra slides
16
DMIP 2 Western Basin Experiments

• NCEP/EMC J. Dong
• HRC K. Georgakakos
• U. Washington J. Lundquist with DHSVM
• CEMAGREF V. Andreassian
• UCI Sorooshian
• U. Illinois Sivapalan
• U. Bologna E. Todini

17
HMT QPE Data Processing for Use in DMIP 2
Advanced DMIP 2 Data Multi-year time series of
gridded data comprised of 1) Basic data and 2)
Processed and gridded HMT data for each IOP
Step 2 Extend Basic Data gridded precip. and
temp. from NCDC, Snotel sites
Step 1 Basic DMIP 2 Data Time series of
gridded precipitation and temperature from NCDC,
Snotel sites to Dec. 2002
-Represent what the RFC uses for current
Forecast operations. -Used for the initial
lumped and distributed DMIP 2 simulations in
the western basins.
Gridded Precipitation for each IOP replaces Basic
Data
Analysis of Data ESRL, NSSL, OHD
Step 3
Note the time scale describes the attributes of
the time series, not the schedule for processing
the HMT data. The HMT observations will be
processed after each campaign and inserted
into the Basic Data time series.
HMT-West Observations Gathered
1
2
3
Year
18
(No Transcript)
19
North Fork American River
20
Methodology for Gauge-Only Gridded QPE
Precipitation Preprocessor -Data QC
-Double mass analysis -Suspect
values -Generate monthly station means
Mean Areal Precip. Processor - Generate mean
areal precip time series - Check data consistency
double mass analysis - Estimate missing data
using station means - Disaggregate all daily data
to hourly values - Non-disaggregated daily obs
put into one hour - Write out hourly time series
for all stations
SNOTEL Daily
-Manual QC Fix non-disaggregated daily
precipitation values -Script to uniformly
distribute remaining daily values
Hourly Point Time Series

• Multi-Sensor Precip. Estimator (MPE)
• Uses PRISM monthly climatology grids
• Uses single optimal estimation in interpolation
• Generate gauge-only 4km gridded QPE

21
00Z 1/22/2000
Snotel
D
Daily
H
Hourly
22
MAP3 Computational Sequence

1. Read in data and corrections
2. Applies consistency corrections to observed data
3. Estimates missing hourly data using only other
   hourly stations.

23
MAP3 Computational Sequencecontinued

• Time distribute observed daily amounts into
  hourly values based on surrounding hourly
  stations.
• Procedure uses 1/d2 weighting for surrounding
  hourly stations.
• If all hourly stations 0, then all
  precipitation is put in last hour of the daily
  station. Hour of the observation time. NFAR
  example
• Estimate missing daily amounts using both hourly
  and daily gages time distribute these amounts
• -If all estimators are missing, then uses 0.0
• Generates file of station and group accumulated
  precipitation for IDMA
• IDMA
• -Compute correction factors
• -Preliminary check of correction factors
• -Insert correction factors into input file
• -Re-run MAP3 for final check of consistency
• Applies weights to station for each area
• Computes hourly MAP time series
• Sums to selected time interval, e.g., 3hr, 6hr.

24
Spatial Extent of DMIP2 American Precipitation
Grid
25
Jan 22, 2000 Corrected 116.58 mm in one hour at
Lake Spaulding.
Corrected Foresthill changed zero to -998 Jan
18 to agree with Georgetown. Corrected Georgetown
data to agree with NCDC paper records (-998 not
-999 on Jan 15-17)
26
DMIP 2 Western Basin Experiments

• HMT experiments 2005-2006 data
• Freezing level, precipitation type
• Value of gap filling radar QPE.