Precipitation
Documentation written by F. Fetterer; edited by M. Serreze
The Atlas gridded precipitation fields were compiled using data from North Pole drifting stations, and from Canadian and Eurasian data sets that have been corrected for biases by the data providers. An iterated Cressman interpolation, including a first guess field, is used in creation of the gridded fields.
Measuring precipitation is particularly difficult because the measured value depends on gauge type and placement and is subject to errors from a number of sources. Blowing snow and the overall low annual precipitation make obtaining accurate precipitation values especially problematic in the Arctic. Groisman et al. [1991] were speaking of the precipitation record in the former Soviet Union, but his comments hold true throughout the Arctic:
"The development of suitable methods to homogenize the precipitation measurements in the Arctic regions of the USSR remains an elusive task. In general, the measurements obtained in the Arctic are a complex function of precipitation type and intensity, wind speed, the degree of protection from the wind at the gauge site, and the gauge type.... In specific episodes of solid precipitation, the measured precipitation can vary from the true precipitation by 100 percent or more."
We have created gridded monthly fields from data that have been corrected by the data providers as a result of painstaking study, but it is important to remember that errors in the fields are likely to be significant. This is true not only because of the biases in precipitation measurements, but because of the sparse data coverage in the Arctic. A study cited by the WMO/World Climate Research Program Arctic Climate System Study (ACSYS) Arctic Precipitation Data Archive (APDA) project finds that to achieve a relative error of about 10 percent when computing areal mean precipitation on a 100 km grid for southeast Canada, the density must be four to 10 stations per grid cell [Rudolf et al., 1994]. Higher density is needed where precipitation is more variable. This density is not available in the Arctic. The ACSYS APDA project (http://www.dwd.de/research/gpcc/acsys/) will address this problem by blending model output with observations. Here, we provide monthly climatological fields by blending bias-corrected station data with a first-guess field.
Before discussing data sources, it is helpful to review bias corrections to precipitation data. Corrections for four effects can be made:
Wind-induced undercatch
Bruce and Clark [1966], in their Introduction to Hydrometeorology, cite studies from as early as 1884 showing that as wind speed increases, measured precipitation is reduced. This effect is much greater for snow than for rain. Gauges are usually shielded to reduce wind undercatchment, but wind effects are still the largest source of error in precipitation measurements. Groisman et al. [1991] note that undercatch can be as much as 50 percent of measured precipitation. Wind loss depends on gauge type (see for example Goodison et al. [1998]), where the gauge is located relative to obstacles, how high the gauge is mounted, and on the type of precipitation.
Trace precipitation amounts
Trace precipitation is precipitation in amounts too small to be resolved by the collecting gauge. For example, a measure on the Tretyakov gauge of less than 0.1 mm is recorded as "trace". Corrections for trace are usually made by adding in a set amount for each day on which trace precipitation was recorded. Trace precipitation can contribute a significant amount to monthly or annual precipitation totals in regions of little precipitation. For instance, Yang et al. [1999] found that the yearly correction for trace precipitation was 5 percent to 11 percent of the gauge-measured annual precipitation for northern Greenland, but was a negligible 3 percent or less for southern Greenland where precipitation is much higher.
Wetting loss
Wetting loss occurs when a gauge is emptied into a measuring device to obtain a precipitation total. The small amount of precipitation that remains behind in the gauge (sticking to its sides) is the wetting loss. The size of this loss depends on how often the gauge is emptied, as well as on the type of gauge and the type of precipitation. Yang et al. [1999] found that for Greenland, the wetting loss is 5 percent to 6 percent of gauge-measured annual precipitation for northern Greenland, and 2 percent to 3 percent for southern Greenland.
Evaporation loss
Evaporation loss is the amount of precipitation lost from the gauge by evaporation between measurements. It depends on gauge type, the frequency of measurements, and weather conditions. While evaporation can be significant (up to 0.8 mm per day for one Finnish site for the Tretyakov gauge, it is not generally possible to apply a general correction due to the site-specific nature of the loss [Yang et al., 1995].
In 1985, the WMO undertook a project to help quantify these biases, and to identify standard corrections [WMO/CIMO, 1985]. The data sets used for this Atlas make use of the results of this project, along with earlier work on standardizing precipitation data in the USSR documented by Groisman et al. [1991].
Eurasia
Monthly precipitation totals for Eurasia were taken directly from the data set Former Soviet Union Monthly Precipitation Archive, 1891-1993. This archive contains data from 622 stations. Work on this archive began at the State Hydrological Institute in St. Petersburg in 1977. The data are available on-line at NSIDC (http://nsidc.org/NSIDC/CATALOG/ENTRIES/nsi-0059.html), along with a file containing the correction factors so that it is possible to reconstruct the uncorrected data. The stations used for the Atlas products, as well as stations from other data sources, are shown in Figure 2.
Pavel Ya. Groisman (NOAA National Climatic Data Center, Asheville, NC, and the State Hydrological Institute) and others compiled station histories and used instrument intercomparisons and field studies to arrive at methods of correcting the original data. Three corrections were applied. The following description is drawn from Groisman et al. [1991]. Note that all gauges in this data set were mounted two meters above ground.
In the late 1940s and early 1950s, Tretyakov gauges replaced Nipher shielded gauges to reduce wind-induced undercatch. A correction factor (K1) was determined from parallel measurements between the two gauge types, and was applied to measurements made with the Nipher gauge in order to make them compatible with the Tretyakov gauge measurements.
In 1966, the number of observations per day was increased from two to four at most stations. A correction for wetting was instituted at the station prior to archiving the data. In 1967, the method of calculating the wetting correction was changed to differentiate between solid, mixed and liquid precipitation. Coefficient K3, the wetting correction, is well correlated with the number of observations per month with solid, mixed, and liquid precipitation. Monthly values for these coefficients are recorded for each station.
In the 1960s, a field experiment was conducted to assess the absolute accuracy of precipitation measurements made with Tretyakov gauges. Based on results from the experiment, a scale correction, K2, was derived. The value of K2 is a function of climatological wind speed and temperature at the gauge site for monthly snowfall, and climatological wind speed and precipitation intensity for monthly rainfall. Groisman et al. [1991] caution that the values for K2 should not be applied to data from individual stations for individual years or months, since the long-term monthly means for wind speed and temperature from which K2 is derived are subject to large deviations in any given year or month.
Canada
Data for Canada were taken from NCDC data set TD-9816 Canadian Monthly Precipitation [Groisman, 1998]. The data set was prepared by P. Groisman. The original data were purchased by NCDC from the Canadian Atmospheric Environment Service (AES) in the early 1990s and then subjected to bias corrections. A total of 6692 stations are available, extending from the beginning of record to 1990, with the earliest station starting in 1874. We use data north of 60 degrees North for the period 1951 through 1990.
Earlier studies [Metcalfe et al., 1997] indicate that before 1975, Canadian gauges had wetting losses of approximately 0.16 mm per measurement. It was also recommended to multiply rainfall measurements by 1.02 to account for wind-induced undercatch.
Information on the number of measurements per day (that would would allow a systematic wetting correction) is not available at all stations. To provide a more homogeneous time series, the wetting correction before 1975 made use of the mean number of days per month with rainfall at the site or, if not available, a value interpolated from nearby locations. The mean values are based on data from the early 1980s onwards, when total rainfall days began to be inserted in the original archive. A total of 2172 stations have this information for at least five years.
The total monthly rainfall prior to 1975 was taken as the measured rainfall plus the mean number of days with rain multiplied by 0.2. This adjusted total was then multiplied by 1.02 to account for wind undercatch. Subsequent to 1975 and with the adoption of the improved Type-B gauge [Metcalfe et al., 1997], a wetting correction was not deemed necessary and the monthly rainfall was adjusted only for wind undercatch. These adjustments increase rainfall by approximately 5 percent before 1975 and 2 percent thereafter. There are no corrections for trace rainfall events.
To obtain total precipitation, rainfall totals were added to the water equivalent of snowfall. Two instruments are used in Canada to assess snowfall water equivalent. At 85 percent of stations, a snow ruler is used to measure the depth of freshly fallen snow, which is then converted into water equivalent using a 10:1 ratio. Starting from the early 1960s, some stations were equipped with Nipher-shielded elevated snow gauges that directly measure the water equivalent of snow [Groisman and Easterling, 1994]. However, these measurements are prone to wind-induced error. Errors appear to be on the order of 15 percent [Golubev et al., 1995], but Groisman [1998] notes that the errors will be site specific. While accurate corrections for gauge undercatch require wind and site exposure information, this information was not available, requiring the use of climatological adjustments.
Climatological ratios (RAT) of monthly snow water equivalent from the Nipher gauge and the snow ruler (Nipher/ruler) were computed. RAT is generally less than 1.0, and in cold climates as low at 0.6. The RAT values were increased by a factor of 10/9 to account for an average snow undercatch by the Nipher gauge (that is, the undercatch is assumed to be systematic). The RAT values were then multiplied by the water equivalent as determined from the 10:1 conversion of the ruler measurements. This procedure amounts to an adjustment of the assumed snowfall density. The adjustments were only performed where the mean monthly snowfall exceeded 3 cm. For cases in which a RAT value could not be determined due to insufficient data, a value of 1.0 was assumed. Deriving the RAT values required identification of those sites and periods for which frozen precipitation was measured by the Nipher gauge. These procedures are outlined by Groisman [1998].
Arctic Ocean
Monthly precipitation totals for the Arctic Ocean are North Pole drifting station data corrected for biases by Daqing Yang (currently with the Institute for Global Change Research, Tokyo, Japan). North Pole drifting station data from 1957 through 1990 from NSIDC data set Arctic Ocean Snow and Meteorological Observations from Drifting Stations were used. Figure 2 shows the mean monthly positions for these stations. The following description of bias corrections to this data set draws from Yang [1999]. For information on the Soviet "North Pole" drifting station program, see the Atlas documentation for the North Pole drifting stations, and the history of the program in "A Look Back". For additional information on precipitation and snow measurements on the stations, see Colony et al. [1998].
Precipitation was measured twice daily with a Tretyakov gauge at two meters height. To correct for trace precipitation, a value of 0.1 mm was added to the monthly total for every day on which trace was recorded. No correction for wetting was needed, because the data had already been corrected based on earlier study [Colony et al., 1998]. No correction was made for daily evaporation loss.
To correct for wind-induced undercatchment, a daily catch ratio was calculated as a function of wind speed and temperature. The functional relationship was developed based on results from the WMO Solid Precipitation Measurement Intercomparison [WMO/CIMO, 1985]. The ratio of Tretyakov gauge catch to "true" precipitation from a reference was regressed against mean daily wind speed at the gauge height and maximum, minimum and average daily temperature. Different relations for snow, snow with rain, rain with snow, and rain were derived [Yang et al., 1995].
The bias corrections adjust the overall mean monthly gauge-measured precipitation upward by from 30 percent to 100 percent (see Yang et al. [1995] Figure 2). Yang notes that the corrections shift the monthly maximum from July to September, which is in agreement with GCM simulations [Walsh et al., 1998].
Fig. 2. Locations of stations with precipitation data that were used in constructing the interpolated precipitation fields (red crosses): Eurasian stations, Canadian stations, and all mean monthly positions for North Pole drifting stations. For comparison, the location of stations in NCAR data set 570, World Monthly Surface Station Climatology (blue squares) and the data set of 65 Russian coastal stations on this Atlas (green squares) are shown as well. All stations north of 50 degrees North are shown, although not all of these stations lie within the Atlas EASE-Grid.
Climatological monthly means of the station data (all data from stations above 55 degrees North acquired between 1951 and 1990, and with at least ten years of data) were first computed. Data from the North Pole drifting stations are from a different position each month. For compatibility with the land-based records, 11 points in the Arctic Ocean were defined. Using a drop-in-the-bucket approach, monthly precipitation means at each of these points were computed by taking any drifting station observation within 500 km of each point. If the resulting mean was based on less than three years of data, it was treated as missing.
The station data were then interpolated to the EASE-Grid using a Cressman interpolation with an iterative correction of a first guess field. For the first guess, we used the Legates and Willmott monthly climatology [Legates and Willmott, 1990], originally provided on a 0.5 degree by 0.5 degree grid, and interpolated to the EASE-Grid (using the Cressman equations described below, with a single 500 km search radius).
In its simplest form, the Cressman routine is given by:
where xi is the precipitation observation at location i within a radius R of the grid location, and n is the number of stations found within the search radius. The weights wi are given by
where d is the distance from the grid cell to the observation location.
The more complex iterative correction scheme used here is an iterated two-step process repeated using successively smaller search radii. In the first step, the operation of the Cressman routine is reversed to obtain values of the first guess at the observation locations. The difference between each observation and the first guess is then computed. In the second step, the difference values are interpolated back to the EASE-Grid using the form of the equations given above, with the gridded difference field added to the first guess field.
We now have an adjusted first-guess field. Using this adjusted first guess field, the two step process is repeated using a smaller search radius. Here, we modify the first guess four times using successively smaller search radii of 750 km, 650 km, 550 km and 450 km. The use of successively smaller search radii results in the final field being more strongly weighted by the observations in data-rich areas, while placing more reliance on the first guess where observations are sparse. More information on this method and on the Legates and Willmott climatology can be found in Serreze and Hurst [2000].
Precipitation products are monthly mean fields for 1951 through 1990. The browse versions of these files are shown as .gif format images with a color bar and contours. These browse files are for the purpose of quickly visualizing the content of the corresponding ASCII data files. The IDL routine used to color-map the images gives a smooth and visually pleasing result, but keep in mind that the gridded ASCII files have one value only for every grid cell. The grid cell centers are shown as red dots.
For information on the structure of the gridded files, see the "EASE-Grid" section.