Bias Correction of Global Gridded Precipitation for Solid Precipitation Undercatch Jennifer C. Adam and Dennis P. Lettenmaier Department of Civil and Environmental Engineering, Box 352700, University of Washington, Seattle, WA 98195 2 4 Methodology Specific to Canada Global Gridded Data Set Comparisons ABSTRACT Systematic biases in gauge-based measurement of precipitation can be particularly severe in the case of solid precipitation. Of these biases, wind-induced undercatch of solid precipitation is by far the most significant. A methodology for producing gridded monthly average catch ratios for the correction of wind-induced undercatch of snow is developed, suitable for application to continental or global gridded precipitation products. Catch ratios (defined as the ratio of measured precipitation to “true” precipitation) are developed for all countries where snow accounts for a significant portion of cold season precipitation and are gridded using a ½º spatial resolution. Catch ratio equations, specific to gauge type and taken from the recent World Meteorological Organization (WMO) Solid Precipitation Measurement Intercomparison, were applied to five years of daily data (1994 through 1998) from which mean monthly catch ratios were estimated. Canadian catch ratios were determined using somewhat more detailed information than for the rest of the domain, and are therefore expected to be more reliable. The gridded gauge correction products are designed to be applicable both to climatological estimates and to individual years during the 1979-98 reference period. As compared with other recent (but more localized) studies that used a similar method to account for wind-induced catch deficiencies; our estimates of mean precipitation tended to be somewhat higher (from 4 to 12%), mostly because we did not attempt to correct for liquid precipitation undercatch. Two recent studies have attempted to create an adjusted precipitation archive for Canada (Groisman, 1998 and Mekis et al., 1999). Both studies were more exhaustive than ours, and in particular, evaluated metadata in more detail than we did. Therefore, we attempted to make use of the results of both of these studies. Groisman (1998) completed a monthly analysis and performed adjustments on 6,692 stations located throughout Canada. Mekis et al. (1999), on the other hand, had access to more detailed metadata for a smaller set of 495 stations. Both data sets make adjustments for wetting losses and wind-induced undercatch for liquid precipitation and Mekis et al. make an adjustment to account for trace precipitation. The catch ratios were applied to twenty years of the Willmott et al. (2001) precipitation data. This adjusted data set was then compared to other existing global precipitation data sets. 5 global monthly precipitation data sets were used in our comparisons (3 time-series and 2 climatologies). All data sets were interpolated to a common resolution before comparison. Climatologies were created from the data sets by averaging the time-series over the period 1979 through 1998. The adjusted Willmott et al. data set approximates the widely used Legates et al. (1990) data during the winter months and especially in the Northern Hemisphere. The adjusted Willmott et al. data set is closer to the unadjusted data sets during the warmer months, because no corrections were made for liquid precipitation. Dataset Details Measurement Biases Corrected For Adjusted Willmott et al. (2001) Time-series 1979-1998 ½º Wind-induced solid precipitation undercatch in snow-dominated regions Willmott et al. (2001) 1950-1999 none Legates et al. (1990) Climatology 1920-1980 Wind-induced undercatch, wetting, and evaporation of liquid and solid precip. GPCC - Rudolf et al. (1994) 1961-1990 1º CRU0.5 - New et al. (2000) 1901-1998 For 485 stations: mean monthly ratios of Groisman to Mekis et al. accumulated monthly precipitation estimates were derived The ratios were gridded to a ½° resolution and applied to the Groisman adjusted monthly station data to create an extensive network of station measurements that reflect the Mekis et al. adjustments. Mean monthly catch ratios were determined by dividing the original unadjusted mean monthly station data by the Mekis et al. adjusted Groisman mean monthly station data, both averaged over the same period. Between 20º and 35º: all data sets are approximately equivalent Above 35º: the adjusted Willmott et al. and Legates et al. data sets are considerably higher than the unadjusted data sets, owing to bias adjustment Between 38º and 45º in Eurasia: the Legates et al. data set is much higher than the adjusted Willmott et al. data set 1 General Methodology The purpose of this study was to develop a ½º global gridded precipitation product suitable for global modeling studies, that reflects as best we could the known effects of measurement biases, using the results from the WMO Solid Precipitation Measurement Intercomparison (Goodison et al., 1998). Our focus is not on the exhaustive reduction or elimination of all sources of error, but rather is intended to reduce the largest component of the net annual biases in estimation of climatological mean precipitation. Therefore, catch ratios were developed to take into account the wind-induced undercatch of snowfall, which generally is the greatest source of error in precipitation measurements. Of particular concern is precipitation estimates in those areas (which constitute approximately ½ of the northern hemisphere land area) where snow accounts for a substantial fraction of the annual precipitation available for runoff. For our study, catch ratios were determined only for the countries that experience at least half of their coldest month’s precipitation as solid precipitation in at least half of their land areas. There were a total of 30 countries selected. Groisman / Mekis et al. Precipitation (mm/month) 5 3 6 Globally Gridded Catch Ratios Comparisons to Yang et al. Comparisons to Canadian and Former USSR Adjusted Data Sets Yang and others performed adjustments on the precipitation measurements from several stations in the USA (1996 and 1998), Greenland (1999a), the Arctic Ocean (1999b), and Siberia (2000). Yang made adjustments to rainfall and snowfall measurements including the systematic biases resulting from wind-induced undercatch, wetting, and the treatment of trace precipitation as zero. Yang’s method to adjust for wind-induced undercatch is similar to our method (e.g. WMO intercomparison results used in both) with the exception that Yang had access to specific station information and therefore did not need to make assumptions regarding gauge type, shielding, gauge height, and wind-sensor height. We determined the Yang catch ratios from their published results by dividing the total gauge-measured precipitation by the sum of the total gauge-measured precipitation and the total depth of wind-induced undercatch. Canada Former USSR The most appropriate regression (based on the WMO study) was assigned to each country. We assumed that a single prevalent type of gauge or shield is representative for a given country. We relied on information in Sevruk et al. (1989) to determine what type of gauge and what type of shielding, if any, was prevalent for a given country. Some countries did not participate in the WMO intercomparison and therefore there were no catch ratio equations derived for their national gauges. In these cases, regression equations for gauges that are similar in design, material, and shielding were applied. Catch Ratio Percent (Measured / True Precipitation) Groisman et al. corrected precipitation data for 622 USSR stations on a monthly basis using Reference Book on the Climate of the USSR (1966 – 1969). These were gridded to ½° resolution and averaged over the 1979 through 1990 period. The Groisman et al. estimates generally are much lower (approximately 20 to 30%) than the adjusted Willmott et al. probably because of the out-dated methods that were used. Monthly catch ratios were then gridded to a ½° resolution using the SYMAP algorithm of Shepard (1984) as implemented by Widmann and Bretherton (2000). The catch ratios were derived using a time-series of daily meteorological values for the period during which all of the necessary variables (mean precipitation, mean wind speed, maximum and minimum temperature) were available (1994 through 1998). Station observations of these variables were obtained from the NOAA Climate Prediction Center Summary of the Day data archived at the National Center for Atmospheric Research. (In the figure, the Canadian stations are not shown because other data were used for Canada.) The Groisman (1998) and Mekis et al. (1999) data were gridded to ½° resolution and averaged over the 1979 through 1990 period. The adjusted Willmott et al. data set captures the approximate values of the more detailed Mekis et al. Corrections. CONCLUDING REMARKS We believe that the adjusted precipitation data set described herein offers an improvement over current global products that are either unadjusted for solid precipitation catch deficiencies, or that use methods that predate the WMO intercomparison (Goodison et al., 1998). Development of high quality gridded global precipitation data sets suitable for large-scale modeling is an incremental process. We believe that the adjustment procedure, and accompanying adjusted (from Willmott et al, 2001) data set is a next step in a progression. Its major desirable features are that it is closely tied to the results of the most recent WMO precipitation measurement intercomparison (Goodison et al., 1998). Note: See the author for a list of references. The adjusted and unadjusted daily precipitation values were summed to provide monthly totals, and mean monthly catch ratios were determined for each station by dividing the unadjusted sum by the adjusted sum. Using the monthly gridded catch ratios to adjust an existing gridded precipitation data set yielded an increase in the global mean annual precipitation of 4.7% over the time-period 1979 through 1998, but the greatest increase of nearly 85% occurs at approximately the 80º latitude during the winter (DJF). Our catch ratios are on average between 4.1% and 7.4% higher than Yang et al.’s The variation among catch ratios is generally inversely related to the mean.