IntroductionDecisions involving climatic precipitation analyses often hinge on issues of data quantity and quality. For instance, researchers frequently face questions such as the following two. At what point do the liabilities of questionable historical gauge measurements override the advantages of maximum density of observations? When several independent (and often substantially different) gauge networks are available, what are the characteristics of the data or of the analyses being undertaken that limit the value of their combination?
We address these questions using observations from a network of hourly reporting stations called the Hourly Precipitation Dataset (HPD), which is managed by the National Climatic Data Center. These data span periods of up to 50 years, and although the sites east of the Mississippi are more evenly distributed, the overall station density is fairly good across the contiguous 48 states. Data quality is generally good; however, one unfortunate characteristic of these data is a gradual changeover with time from gauges that report precipitation to the hundredth inch to gauges that are limited to tenth-inch resolution. By treating the two sets of gauges as separate datasets, we are able to investigate the possibility of significant differences between precipitation frequency that are obscured when all gauges are included irrespective of resolution. Longer-term averages and station extreme values are also examined with this resolution difference in mind.
Gauge Data Characteristics
Studies have been done regarding the impact of HPD dataset characteristics on analyses that utilize them. The studies described composite analyses of mesoscale convective complex (MCC) precipitation produced using HPD observations. The primary concern was the possibility that differing resolution of HPD gauges may distort estimates of average MCC rain rates. The conclusion was that the overall trends, at least, were preserved. Other corresponding studies at FSL by Randall Collander and others described a large increase in per-day frequency of hourly rainfall at HPD gauges that measure to hundredth- inch resolution as compared with the same statistics at tenth-measuring HPD sites.
Our interest here is primarily with daily up to monthly precipitation. Thus, we have summed hourly observations from the HPD over daily and monthly time periods. For long period comparisons, observations are examined from a 2o x 2o area in Indiana and Illinois (Region A in Figure 1). Concentrating on this relatively small area reduces data requirements to a reasonable level. The choice of this particular area of flat terrain allows us to assume that the data retain relatively uniform statistical properties from station to station.
Figure 1. Hourly Precipitation Data station sites. Rectangular boxes indicate areas for gauge resolution study (Region A, Indiana and Illinois; Region B, an expanded area.)
Gauge Resolution Effects – As indicated previously, the HPD is in essence a combination of two datasets, one that includes gauges with hundredth-inch resolution and another that includes gauges with tenth-inch resolution. The dataset thus provides an opportunity to determine the effects of different gauge resolutions. There may also be other differences in measurement characteristics that make the two sets of observations inhomogeneous.
Figure 2 shows the changeover from hundredth to tenth gauges in Region A on Figure 1. Although this changeover has occurred on different schedules in various parts of the United States, Region A is representative of most HPD gauge sites. Before 1965 most of the HPD observing sites used Universal gauges, which could resolve precipitation to one-hundredth inch. By 1990, almost all sites were using tenth-measuring instruments. The total number of gauges remained roughly the same during the entire period of record.
Between 1965 and 1986 the set of gauges in Region A included both types of gauges. Thus, observations are chosen from this period to determine how much of an effect the difference in resolution can have on precipitation averages, distributions, extremes, and frequencies.
Figure 2. The number of stations with the indicated gauge type in Region A of Figure 1.
Monthly and Daily Averages – Area-averaged totals constitute a fundamental application of precipitation observations. To see if the two types of gauges give significantly different values when used in this way, we compute separate monthly averages in Region A, one consisting of only hundredth gauges and another consisting of only tenth gauges. The results are displayed on the scatterplot of Figure 3. Although the two sets of averages show some scatter, there are no evident systematic differences between the two gauge types (the one large January anomaly occurred during the first year that the then new tenths gauges were introduced; the small sample of tenths gauges that month (only 1), and possibly trouble with interpreting or reporting the new measurements, might explain this single large outlier). The average absolute standard error between the two sets of averages is .43 inches in January and 0.61 inches in July, whereas the systematic bias between the two is .32 Inches in January and .08 inches in July, with tenths-gauge averages being larger in both cases.
Even for accumulation periods as long as a month, spatial variability within the region, especially during July, is likely to cause much of the difference in the monthly averages in Region A. One possible method to determine how much of the difference is due to instrumentation and how much to natural spatial variability would be to randomly select pairs of gauge sets in the region irrespective of gauge type and compare the resulting area average differences with those computed here.
Figure 3. Simultaneous measurements of monthly total precipitation observed at hundredths- and tenths-measuring HPD gauges. Totals represent individual monthly averages over gauges of each type in Region A of Figure 1, between 1965–1986.
Weekly averages were compared in the same way as the monthly averages shown here. Although somewhat higher variability resulted, as might be expected, there again was no large bias between gauge types. Comparison of the full distributions of daily precipitation (not shown) also revealed no clear gauge- related differences.
Precipitation Frequency – Gauge resolution differ-ences have the most potential impact on precipitation frequency. For instance, in 1993 Collander et al. demonstrated large differences in the distributions of hours of precipitation per day as measured by the hundredths and tenths gauges in the HPD.
We investigate a related issue by examining how the two gauge types measure the overall frequency of nonzero hourly precipitation during individual months. Figure 4 illustrates this comparison for 20 July between 1966 and 1985. In order to include an adequate number of gauges of each type, we have increased the domain size (Region B on Figure 1) and decreased the display period by two years relative to results described earlier.
The gauge-related differences among July frequencies displayed in the figure are substantial, averaging nearly a factor of 2 during the 20 years. Year-to- year differences can also be almost this large. However, the two frequencies track each other closely from July to July, and their relative difference remains fairly constant, suggesting that the instrumentation effect overwhelms the natural temporal variability. Clearly, gauge differences must be considered in studies utilizing long-term time series or averages of hourly precipitation frequencies.
Figure 4. Percentage of hours in July with measured precipitation observed by hundredths- and tenths-measuring HPD gauges in Region B of Figure 1.
Since the frequency of hourly precipitation measured by the hundredth gauges is always larger, an explanation for the results appears straightforward: the hundredth gauges measure precipitation events that are not registered by the higher threshold of the tenth gauge. A related effect is the tendency for small (<.1 inch) observations to be accumulated over several hours in the tenth gauges and finally registered as a .1 inch event some hours later than the event is recorded by a hundredth gauge. For instance, two consecutive hourly rainfalls of .06 inches would be recorded by a tenth gauge as .1 inch at the end of the second hour.
These effects are more prevalent when precipitation rates are small; for example, when accumulations of less than .1 inch are common. Thus, the July differences in rainfall frequency shown in Figure 4 are less than those computed in a similar fashion using observations from January (not shown here) when precipitation rates are often small. The effects are also reduced by accumulating precipitation over longer time periods. The intergauge difference in frequency of nonzero daily precipitation during July, for instance, is of the order of 10%, as compared to the nearly twofold difference for hourly precipitation.
Precipitation Extremes – Now we compare the largest of the monthly and daily precipitation totals computed from the two sets of gauges. The magnitude of the differences between the two sets will indicate whether it would be important to distinguish between gauge types with varying resolution in long-term studies of precipitation extremes.
One approach to the study of extreme values is to examine the highest percentiles of precipitation. Figure 5 shows a seasonal breakdown of the 90th, 95th, and 99th percentiles of daily precipitation broken down by gauge type. Except for the 99th percentile, the curves are essentially identical for the hundredth and tenth gauges. The relatively small variations in the 99th percentile may be as much due to small sample size at these large precipitation rates as to any instrumentation. In any event, seasonal changes in all percentiles dwarf the differences related to gauge type.
Figure 5. Percentiles (90th, 95th, and 99th) for distributions of daily total precipitation between 1965 to 1986 at all hundredth- and tenth-measuring HPD gauge sites in Region A on Figure 1.
We can also compare extreme values by determining the largest precipitation event recorded at both types of gauges during the period when both sets were in operation. The Table lists the number of years (out of 20 possible between 1966 and 1985) in which each gauge type observed the largest precipitation total in that month (yearly maxima, January and July), and the number of gauges of each type that observed the largest monthly total of precipitation during any of the months in the 20-year period (station maxima, January and July). There is a remarkably even distribution of these record values between gauge type, suggesting no discernible difference in the way the two gauges handle heavy precipitation.
Concluding RemarksFor users of the Hourly Precipitation Data, it is a welcome result to find that the existence of multiple rain gauge types in the historical dataset does not appear to adversely affect (at least in any significant way) the use of the full HPD in studies involving precipitation field analyses, area averages, or extreme value studies. This latter finding is particularly encouraging since the search for extreme values is best undertaken with as long and complete a series of observations as possible. On the other hand, for studies that require short-term precipitation frequencies, researchers will clearly need to recognize the differences between the two kinds of gauges.
[Editor's Note: More information on this topic is available on the MAB Website, including bibliographies.]
( Edward Tollerud is a researcher in the Meteorological Applications Branch, headed by Cecilia Girz.)