1. Experimental Real-time Seasonal Hydrologic Forecasting
Andrew Wood
Dennis P. Lettenmaier
University of Washington
presented:
AMS Conference on Applied Climatology, 2002
Portland, OR May 2002
important points:
Arun Kumar works for the NCEP Environmental Modeling Center, Climate Modeling Branch and runs the global spectral model to produce forecasts.
this study period and region are important because there is an evolving drought that we would like to use as a test situation for the method’s skill.
this approach was previously applied to the East Coast during Summer 2000, when there was an evolving drought in that region – a paper evaluating the project is currently submitted to JGR-Atmospheres.

2. Project Overview Research Objective:
To produce monthly to seasonal snowpack, streamflow, runoff & soil moisture forecasts for continental scale river basins
Underlying rationale/motivation:
Global numerical weather prediction / climate models (e.g. GSM) take advantage of SST – atmosphere teleconnections
Hydrologic models add soil-moisture – streamflow influence (persistence)
important point(s):
the approach attempts to make use of forecast skill from 2 sources:
better understanding of synoptic scale teleconnections and the effects of persistence in SSTs on regional climate, as reproduced in coupled ocean-atmosphere models;
the macroscale hydrologic model yields an improved ability to model the persistence in hydrologic states at the regional scale (more compatible with climate model scales than prior hydrologic modeling).
Climate forecasts with monthly and seasonal horizons are now operationally available, and if they can be translated to streamflow, then they may be useful for water management.

3. Topics Approach Columbia River basin (summer 2001) results
Ongoing Work
Comments
important point(s):
the approach attempts to make use of forecast skill from 2 sources:
better understanding of synoptic scale teleconnections and the effects of persistence in SSTs on regional climate, as reproduced in coupled ocean-atmosphere models;
the macroscale hydrologic model yields an improved ability to model the persistence in hydrologic states at the regional scale (more compatible with climate model scales than prior hydrologic modeling).
Climate forecasts with monthly and seasonal horizons are now operationally available, and if they can be translated to streamflow, then they may be useful for water management.

4. General Approach climate model forecast meteorological outputs
~1.9 degree resolution (T62)
monthly total P, avg T
Use 3 steps: 1) statistical bias correction
2) downscaling and disaggregation
3) hydrologic simulation
 hydrologic model inputs
streamflow, soil moisture,
snowpack,
runoff
1/8-1/4 degree resolution
daily P, Tmin, Tmax
important point(s):
the overall forecasting approach involves using forecast model (the global spectral model) T & P output at a coarse timestep & scale as hydrologic model input at a finer timestep and scale.
to make a hydrologic forecast, you need a transformation of the forecasts that first overcomes climate model bias and the scale differences, then simulates the water balance.
also, GSM is really run at very fine timestep (~5-15 minutes) but only the monthly anomalies are archived for our use. most of the signal is at the monthly scale, however, so this is acceptable.

5. Models: 1. Global Spectral Model (GSM) ensemble forecasts from NCEP/EMC
forecast ensembles available near beginning of each month, extend 6 months beginning in following month
each month:
210 ensemble members define GSM climatology for monthly Ptot & Tavg
20 ensemble members define GSM forecast
important point(s):
GSM forecasts take the form of monthly ensembles of length 6 months
we get them early in each month for a start date of the following month.
the climatology ensemble enables us to define the climate model bias and correct it
climatology ensembles run out 6 months just like the forecasts, but use observed rather than predicted tropical Pacific SSTs
also:
210 ensembles for GSM climatology are derived from observed SSTs in each year of the 21 year climatology period ( ) combined with 10 initial atmospheric conditions for each year
GSM is at T42 spatial resolution, but moving to T62 soon (resolution improvement of about 1/3)

6. Models: 2. VIC Hydrologic Model
important point(s):
VIC is a water & energy balance with some subgrid scale approximations for vegetation, elevation and soil dynamics, and has a crude routing that works as a post processor.
VIC has been applied to a number of continental scale river basins around the world and is well documented in the literature.

7. Flow Routing Network domain slide important point(s):
we’re modeling most of the US at 1/8 degree now with the VIC model, but we are performing this forecasting exercise in the Columbia River basin. The plusses show the grid of the numerical weather prediction (forecasting) model that we used (GSM), and the ¼ degree hydrology model resolution can just be discerned in the figure. 24 climate model grid points were used, and 1,668 VIC model cells. We’ve aggregate the VIC model to ¼ degree from 1/8 degree in the Columbia River basin to speed the forecast runs.

8. One Way Coupling of GSM and VIC models
a b c.
TGSM
TOBS
a) bias correction:
climate model climatology  observed climatology
b) spatial interpolation:
GSM ( deg.)  VIC (1/8 deg)
c) temporal disaggregation (via resampling of observed patterns):
monthly  daily
important point(s):
this is our approach to turning GSM forecasts into VIC input: first, bias correction:
1. using the climatologies of the observed precip & temperature to define parallel distributions (for each month in the forecast), we translate each met. value in the GSM ensembles to a quantile value, then retrieve the met. variable value for that quantile from the observed distribution. (at the ends of the empirical distributions, we use fitted theoretical ones if needed).
then downscaling:
2. that was all done at the GSM scale. then we interpolate the anomalies to the to the VIC resolution (nothing fancy here).
then we impose the daily pattern by resampling the historic VIC forcings (P&T for each month taken from the same year to preserve correlations), and then scaling monthly avg. temp and shifting month. tot. P to reproduce the forecast anomalies. after all, when you sample at random, the daily pattern you get won’t have the monthly anomaly you need for the forecast signal.
the bias correction step is critical, as the next 2 plots will show.

9. Bias Example: JFM precipitation from Parallel Climate Model (DOE) climate model vs. “observed” distributions at climate model scale (T42)

10. Dealing with bias using a climatology-based correction
Note: we apply correction to both forecast ensemble and climatology ensemble itself (to use as a baseline)

11. Downscaling: add spatial VIC-scale variability
observed
mean fields
(1/8-1/4 degree)
monthly GSM
anomaly (T62)
VIC-scale
monthly forecast
interpolated to VIC scale
note:
month m, m = 1-6
ens e, e = 1-20

12. Lastly, temporal disaggregation…
for each VIC-scale monthly forecast value, e.g.:

13. Simulations VIC model spin-up Forecast Products streamflow
soil moisture
runoff
snowpack
VIC model spin-up
VIC forecast ensemble
climate forecast information (from GSM)
VIC climatology ensemble
1-2 years back
NCDC met. station obs. up to 2-4 months from current
LDAS/other met. forcings for remaining spin-up
data sources
start of month 0
end of month 6
important point(s):
after bias correcting and downscaling the climate model forecasts, the procedure for producing hydrologic forecasts is as follows:
we spin up the hydrologic model to the start of the forecast using observed met. data (from 2 sources: NCDC cooperator stations through 3-4 months before the start of the forecasts, then LDAS 1/8 degree gridded forcings thereafter).
The GSM forecasts comprise 2 sets of ensembles, one for climatology and one for the forecast. The climatology ensemble yields a distribution of the conditions we’ve seen over the period , while the forecast ensemble yields the distribution of the conditions we might see for the next 6 months. Although the climatology ensemble is nominally unbiased against a simulated climatology based on observed met. data (rather than bias-corrected, downscaled GSM met. forcings), we compare the forecast and GSM climatology so that any unforeseen biases (resulting, perhaps, from the downscaling method) occur in both climatology and forecast. Eventually this cautionary step may be eliminated, and we’ll compare directly to the simulated observed climatology.
at the end of the spin-up period and one month before month 1 (out of 6) of the forecasts, we save the hydrologic model state. The state is then used for initializing the forecast runs. Through the first month, the model runs on observed data to the last date possible, then switches to the forecast data. Usually, we process the observed forcings up through the 15th to 25th of this initialization month, then the forecast forcing data carries the run forward for the remaining days in the month, and throughout the following 6 month forecast period. Note, the state files used for the climatology runs correspond to the spin-up associated with the particular year (out of ) from which the climatology ensemble member is drawn.
the spin-up period captures the antecedent land surface hydrologic conditions for the forecast period: in the Columbia basin, the primary field of interest is snow water equivalent.
forecast products are spatial (distributed soil moisture, runoff, snowpack (swe), etc.), and spatial runoff + baseflow is routed to produce streamflow at specific points, the inflow nodes for a management model, perhaps.

14. Columbia River Basin Application

15. Initial Conditions late-May SWE & water balance
important point(s):
This set of plots shows the initial conditions (starting state, approximately) for snowpack in the basin, in comparison to 2001 SNOTEL data, and the 1988 and 1977 simulated conditions.
The small circles in the plots at left contain the SNOTEL swe estimates for 2001, march 15. There are many more stations, but I just plotted a couple dozen.
The background in the plots at left is the simulated snow water equivalence, for comparison to SNOTEL. In the top plot, it looks like 1977 is a little lower than current SNOTEL. In the middle one, it looks like 1988 is a closer match, and in 2001, it looks like we undersimulate SWE a bit in some locations, compared to SNOTEL (see the sites in northern Idaho).
On the right are the ratios of 1977 to 2001 and 1998 to 2001, which confirm that the 2001 simulation shows deeper snowpack than 1977, somewhat nearer to 1988 (another very low year, mind you).
One observation about the discrepancies between SNOTEL and the simulation is worth making: The VIC grid cells represent areas of about 150 km squared (1/8 degree) and 625 km squared (at ¼ degree), whereas the SNOTEL data are points, so we don’t expect them to match up perfectly. The VIC model, though, adds an estimate of the spatial distribution of the snowpack that is only possible in a more limited way from the point SNOTEL data – so the combination of the two has the potential to yield improved estimates of basinwide snowpack than would be possible without the distributed hydrologic model. Not to mention, retrospective comparisons are possible to years before the SNOTEL network existed, such as 1977…
We could also show soil moisture and runoff starting states, but the snowpack is most critical in this basin.

16. Initial Conditions late-May SWE & water balance (percentiles)

17. May climate forecast
forecast
observed
forecast
medians

18. May snowpack forecast
hindcast
“observed”
forecast
forecast medians

19. May runoff & soil moisture forecast
hindcast “observed”
forecast
medians

20. May streamflow forecast
important point(s):
The streamflow results are shown for The Columbia River at the Dalles as two ensembles (climatology and forecast) plotted alongside each other, for each forecast month.
The clear message from this plot is that the forecast (red plusses) distributions are clustered significantly below the climatology distributions, to a diminishing extent at the end of the forecast period (when runoff is dominated more by precipitation than snowmelt).
Note, however, that the climatology period doesn’t include the recent end point, 1977, so if it did, the distributions might extend further down with respect to the forecast ensembles – not enough to change the significance of the results, however.

21. Ongoing Work: Assessment and Expansion

22. Tercile Prediction “Hit Rate” e. g
Tercile Prediction “Hit Rate” e.g., GSM Ensemble “Forecast” Average, January (based on retrospective perfect-SST ensemble forecasts) Masked for local significance
important point(s):
The streamflow results are shown for The Columbia River at the Dalles as two ensembles (climatology and forecast) plotted alongside each other, for each forecast month.
The clear message from this plot is that the forecast (red plusses) distributions are clustered significantly below the climatology distributions, to a diminishing extent at the end of the forecast period (when runoff is dominated more by precipitation than snowmelt).
Note, however, that the climatology period doesn’t include the recent end point, 1977, so if it did, the distributions might extend further down with respect to the forecast ensembles – not enough to change the significance of the results, however.

23. U.S. West-wide Hydrologic Forecasting
important point(s):
The streamflow results are shown for The Columbia River at the Dalles as two ensembles (climatology and forecast) plotted alongside each other, for each forecast month.
The clear message from this plot is that the forecast (red plusses) distributions are clustered significantly below the climatology distributions, to a diminishing extent at the end of the forecast period (when runoff is dominated more by precipitation than snowmelt).
Note, however, that the climatology period doesn’t include the recent end point, 1977, so if it did, the distributions might extend further down with respect to the forecast ensembles – not enough to change the significance of the results, however.

24. Summary Comments
climate-hydrology model forecasting method has potential
hydrologic persistence was most important in the CRB example
bias-correction of climate model outputs (using a climate model hindcast climatology) is critical
access to quality met data for hydrologic model initialization is also essential