due to lack of data in the northern part of the country.) Fluctuations on decadal time scales are evident, however, which led Klugman (1983) to claim "evidence of climatic change" in the U.S. seasonal precipitation during the 19481976 period. Diaz and Quayle (1980) found large-scale changes in contemporary precipitation (1955-1977) as compared to precipitation at the beginning of the century (1895-1920), but these changes were shown to be mainly statistically non-significant. Vining and Griffiths (1985) found an increase over time in the decadal variance of annual precipitation at 10 long-term U.S. stations for the period 1900-1979. Bradley et al. (1987), analyzing the area-averaged percentile of the U.S. annual precipitation, found a precipitation decrease during the period from the 1880s to the 1930s and a general increase thereafter.
There are several reasons for precipitation undercatch by the standard gauges currently being used worldwide (Bogdanova, 1966; WMO, 1991; WWB, 1974). The main factor that reduces the amount of measured precipitation as compared to "ground truth" is wind-induced turbulence over the gauge orifice (Sevruk, 1982). The absence of a wind shield for the gauge (or its poor design) makes the problem even more severe. Field experiments show that an appropriate wind shield (such as the modified Nipher shield used in Canada) reduces wind-related bias in the gauge catch to manageable levels that can then be easily adjusted. All elevated snow gauges in Canada have wind shields, but only about 200 of the approximately 6,000 U.S. gauges that report daily precipitation have a wind shield (Karl et al., 1993a). Furthermore, the Alter shield in use at U.S. stations provides less protection than the Canadian shield (Goodison et al., 1981). As a result, the existing U.S. rain-gauge network measures rainfall with a bias of 3 to 10 percent (Golubev et al., 1992; Sevruk and Hamon, 1984) and snowfall with a bias of up to 50 percent or more (Goodison, 1978; Larkin, 1947; Larson and Peck, 1974). Moreover, in Alaska total biases up to 400 percent in measurements of water equivalent of snowfall have been reported (Black, 1954).
Rain gauges at some of the U.S. primary stations were equipped with the Alter wind shields toward the end of the 1940s. The addition of the wind shield introduced an inhomogeneity into the time-series precipitation data at these stations. Although the stations with Alter shields constitute a small percent of the U.S. stations, they comprise 40 percent of the U.S. National Weather Service stations transmitting meteorological observations over the Global Telecommunication System (GTS) for publication in the Monthly Climatic Data for the World.
Canadian methods of making precipitation measurements differ significantly from those used in the United States. This is illustrated by the jump in measured precipitation values often registered by stations on opposite sides of the border between the two countries (WMO, 1979; WWB, 1974). Although the difference in liquid precipitation is small—the Canadian gauge typically measures a few percent more rain than the standard U.S. gauge (Sanderson, 1975)—there is a considerable difference in the solid precipitation measurements (Goodison et al., 1981; Sanderson, 1975). Currently, 85 percent of the Canadian meteorological network uses a 10:1 ratio as a measure of the water equivalence of fresh-fallen snow. Since the 1960s, 15 percent of this primary network has used an elevated snow gauge with a modified Nipher wind shield. This gauge catches solid precipitation effectively even during high-wind conditions (Goodison, 1978; Goodison et al., 1992). Its record differs systematically from the snow-stick measurements continued at other stations (Goodison et al., 1981; Groisman et al., 1993; Karl et al., 1993a). This difference in snow-measurement methods causes both space and scale inhomogeneities, which cannot be avoided by users of such international publications as World Weather Records and Monthly Climatic Data for the World, since Canada transmits only the data from its primary network over the GTS.
When point precipitation measurements are expanded to area-averaged values in rough terrain, the point values are often less than the "ground-truth area mean", because most meteorological stations are located in valleys, but the vertical gradients in precipitation have been neglected. Figure 1 shows two maps of annual precipitation over North America. The first map (la), which was constructed during the World Water Balance studies (1974), incorporates estimates of topographic effects on precipitation. However, the second map (lb) simply uses data from 1,900 stations, without any consideration of topographic effects on precipitation. Precipitation in the mountainous West is obviously underestimated on the second map, and even Appalachian-ridge precipitation estimates are noticeably biased.
The area-averaged annual precipitation value for the contiguous United States derived from Figure 1a is about 880 mm, while the value from Figure 1b is 700 mm. When we use data from all 6,000 U.S. cooperative stations, this latter annual total increases to 735 mm (NCDC, 1991) but still remains quite low compared to the value reflecting topography. The same problem has been revealed to exist for Canada by Hare (1980), who showed that annual totals over the country should be increased by 20 percent, on average, to obtain a reliable water balance (i.e., runoff = precipitation - evaporation).
Precipitation measurements are very sensitive to changes in the environment surrounding the gauge, in the gauge type, and in the methods of measurement. A small move of the gauge can result in a twofold change in the values of measured precipitation if the gauge exposure is changed (Karl et al., 1993b; Sevruk, 1982). After spatial averaging, it can be expected that the random errors connected with such shifts will cancel out. However, when systematic changes occur countrywide, they cannot be assumed to cancel out. Several such changes have occurred during the