Spatial variability of the correction ratio across each of the gauged basins was constructed by developing a relationship between the correction ratio and the 5min correction bands (discussed in Box 1). PRISM (Daly et al. 1994) for the US was used to determine the form of this relation by assuming that the variability of PRISM precipitation with elevation is correct. 33 basins in the US were used for this regression. A quadratic expression was used in which two constraints were imposed (see box).

San Joaquin(Vernalis, CA)

Area = 35,058 km2

Rave = 1.29A = 0.026B = -0.022C = 0.995

Correction of Global Precipitation Products for Orographic Effects Jennifer C. Adam1, Elizabeth A. Clark1, Dennis P. Lettenmaier1, and Eric F. Wood2

1. Department of Civil and Environmental Engineering, Box 352700, University of Washington, Seattle, WA 981952. Department of Civil Engineering, Princeton University, Princeton, NJ, 08544

8th International Conference on Precipitation (August, 2004) Vancouver, British Columbia, Canada

ABSTRACTUnderestimation of precipitation in topographically complex regions is a problem with most gauge-based gridded precipitation data sets. Gauge locations tend to be in or near population centers, which usually lie at low elevations relative to the surrounding region. For example, past modeling studies have found that simulated mean annual Columbia River streamflows using gridded precipitation based on Global Precipitation Climatology Center (GPCC) precipitation products is about one-third of the observed discharge. In an attempt to develop a globally consistent correction for the underestimation of gridded precipitation in mountainous regions, we used a hydrologic water balance approach. The precipitation in orographically-influenced drainage basins was adjusted using a combination of water balance and variations of the Budyko ET/P vs. PET/P curve. The method is similar to other methods in which streamflow measurements are distributed back onto the watershed and a water balance is performed to determine “true” precipitation; but instead of relying on a modeled runoff ratio, evaporation is estimated using the ET/P vs. PET/P curves. This approach requires annual time-series of hundreds of historical discharge records world-wide which were obtained from the Global Runoff Data Center (GRDC) and the Global River Discharge Database (RivDIS v1.1). The correction ratios from each of the gauged basins were interpolated to the rest of the orographic domain using dominant wind direction and fine-scale elevation information. These ratios were applied to an existing global precipitation data set (1979 through 1999, 0.5˚ resolution), following application of adjustments for precipitation catch deficiencies.

Definition of Correction Domain1

Ratio/Correction Band Relation3

5

Comparison to PRISM over the Contiguous USA5 Application of Correction Ratios

2 Determination of Average Ratios for Selected Basins

CONCLUDING REMARKS

• This work satisfies a need for global gridded precipitation data that account for orographic effects (Nijssen et al. 2001). • This approach (using water balance and a variation of the Budyko ET/P vs. PET/P curve) was implemented over the globe and found to have realistic results in most places. • Some continents (e.g. Africa, Europe, and Asia) have large isolated areas of low correction ratios which we suspect to be due to the use of un-naturalized streamflow. Therefore, as a simple solution, we recommend applying only correction ratios that are greater than one.• PRISM precipitation (Daly et al. 1994) is an independent estimate of precipitation magnitude (we used PRISM to aid in constructing the spatial variability of the correction ratios). Although our corrections result in somewhat higher precipitation values, these magnitudes are comparable and realistic.

Note: See the author for a list of references.

0.0

0.2

0.4

0.6

0.8

1.0

1.2

0 1 2 3PET/P

E/P

Budyko (1974) S&V, gamma=1.5S&V, gamma=1.0 S&V, gamma=0.5Physical Limits

EnergyLimited

MoistureLimited

All 5min cells within the correction domain were assigned to a correction band ranging from 2 (lowest elevations) to 7 (highest elevations). The bands were assigned by determining the maximum and minimum elevations within a specified radius of the cell and evenly dividing the elevations between minimum and maximum into the six correction bands.

UpslopeDownslopeCross-Wind

Annual dominant wind direction at 0.5˚ resolution was determined by using the NCEP/NCAR Reanalysis (Kalnay et al. 1996) daily meridional and zonal wind speeds. Using the 0.5˚ dominant wind direction data (below), and a 5min DEM, slope type at a resolution of 5min was determined by finding the direction of steepest slope from the DEM (over a scale of approximately 50 km), and comparing to the dominant wind direction of the overlying 0.5˚ grid cell. Correction ratios were interpolated from 5min grid cells in gauged basins to

2 3 4 5 6 7

Correction Band

The ET/P vs. PET/P curves of Budyko (1974) and Sankarasubramanian and Vogel (2000). The curves are semi-empirical: the limits reflect physical constraints, but the curves are developed from observations. PET was calculated at 0.5˚ for each year between 1979 and 1999 using the Droogers and Allen (2002) method. The 0.5˚ dataset of Dunne and Willmott (2000) was used for maximum soil moisture storage capacity.

GQEPdt

dS QEP

γφ,fP

E

P

PETφ Where = Aridity Index

P

bγ Where = Soil Moisture Storage Index

γφ,fP

Q-P

max(S)max(E)bmax

max(S) = Soil Moisture Storage Capacity

PETP PET,

PETP P, max(E)

Equations

1

2

In an attempt to develop a globally consistent correction for the underestimation of gridded precipitation in mountainous regions, an approach is used in which streamflow measurements are distributed back onto the watershed and a water balance is performed for that watershed (equation 1). Because evaporation is also an unknown, a second equation is needed. We used the ET/P vs. PET/P curves, discussed in depth by Budyko (1974). The equation of Sankarasubramanian and Vogel (2000) was applied because it also takes into account the soil moisture storage capacity (equation 2). A basin-average correction ratio (Rave) was determined by dividing the “true” precipitation for that basin (calculated from equations 1 and 2) by the precipitation described in Adam and Lettenmaier (2003).

The gridded 0.5˚ correction ratios (left) were applied (via multiplication) to the annual climatology of the dataset described by Adam and Lettenmaier (2003). This is a dataset in which monthly climatological corrections for gauge-catch deficiencies are applied to the monthly 0.5˚ time-series (1979-1999) of Willmott and Matsuura (2001). There are large isolated areas of low correction ratios in most of Africa, Europe, and Eastern Asia (and a few smaller areas elsewhere) which we suspect

may be due to the use of un-naturalized streamflow. It is reasonable to expect correction ratios less than one on the lee side of major divides, but large isolated areas of low correction ratios suggest a limitation in the method. A simple solution is to truncate correction ratios that are less than one to one. We note that orographic corrections in Europe are not as important as in other

the 5min gridded data of slope types (the correction ratios should be affected by the type of slope, e.g. if the grid is on an upslope or downslope, where the rain shadow occurs). An simple distance weighting scheme was used in which the correction ratios were interpolated from grids with the same slope type and the same correction band. A radius of 500 km was used for the interpolation, but this radius was increased if the minimum of 10 data points was not met. The 5min interpolated correction ratios were aggregated to 0.5˚ (as shown in Box 5 for the globe).

In this figure, streamflow stations are overlaid onto the correction domain. Data sources include: RivDIS v1.1, GRDC, and HCDN (United States only).

aveRA 027.0061.0

)(bandrCbandBbandA 2

Equation

Constraints:1. r=1 for band=12. Rave is conserved

From PRISM:

The parameter, A, was found to have a slight dependency on Rave, and therefore is calculated as a function of Rave.

Before Orographic Correction

After Orographic Correction

2 3 4 5 6 7

Correction Band

4 Interpolation of Ratios to Ungauged Basins

grid cells in the rest of the correction domain using

Continent Precipitation Increase: Entire Continent

Precipitation Increase: Correction Domain Only

Africa -3.7% / 0.9% -19.5% / 4.5%

Australia -2.2% / 2.4% -7.3% / 8.1%

Eurasia 6.8% / 10.4% 13.9% / 21.2%

North America 5.6% / 6.4% 19.9% / 22.8%

South America 1.8% / 3.3% 13.2% / 24.0%

Global 3.0% / 5.8% 9.9% / 19.0%

6

Africa

Asia

Australia

Europe

N. America

S. America

Africa

Australia

Eurasia

N. America

S. America

Globe

All Corr.Corr. > 1

continents because the distribution of precipitation stations with elevation matches more closely the hypsometric curve. (See figure at the bottom left which shows the difference between the percent of stations and the percent of area for each elevation band; i.e. the differences are

lowest in Europe.) The percent increase in precipitation due to the application of the correction ratios was computed for each continent and globally (See table: blue values are for all corrections; red values are for corrections greater than one only).

PRISM climatology is an independent estimate of precipitation magnitude and provides a comparison for our corrected precipitation data. We compared the percent increase in precipitation due to our orographic corrections to the percent increase inferred by PRISM

PRISM-Inferred Corrections

Adam et al. Corrections

(Daly et al. 1994) by using the Willmott and Matsuura (2001) 0.5˚ data as the base-line precipitation (the PRISM climatology was first aggregated to 0.5˚).

Percent Increase

PRISM Adam et al.

Entire USA 3.2 % 6.0 %

Inside Corr. Domain

20.9 % 28.4 %

Outside Corr. Domain

1.4 % 0.0 %

The correction domain was defined as all 0.5˚ slopes (aggregated from 5min) greater than a threshold of 6 m/km, the approximate slope above which the Willmott and Matsuura (2001) data differs by more than 10% from PRISM (Daly et al. 1994).