minimum used in the excessive precipitation sheets. The smallest number of stations for one storm, used in plotting the rainfall maps, was 33 and the greatest number for one storm was 965.

As shown in figure 104, the stations were first listed in alphabetical order, and then opposite each station in the successive columns were copied the daily rainfall records during the storm period considered. The appropriate dates are entered at the tops of the respective columns.

Next, each column was totaled. The day with the greatest total was considered the date of maximum 1-day precipitation. In most of the 33 storms thus studied in detail no doubt exists as to the date of maximum 1-day precipitation. In some cases, however, usually in storms of long duration and covering a large extent of territory, there was some uncertainty as to what should be considered the maximum 1-day period. The date of maximum precipitation over one center of heavy rainfall, for instance, is sometimes different from the date of maximum precipitation over another distinct center of sufficient importance to merit separate consideration.

The period of maximum 2-day precipitation was determined by joining to the date of maximum 1-day precipitation either the day preceding or the day following, using the day on which occurred the greater total sum of precipitation at all the stations. The dates for the 3-day and successive maximum periods were similarly determined. After the dates were fixed, the total amounts of precipitation at each station for the maximum 1, 2, 3, 4, and 5-day periods were entered in the last columns of the form as illustrated in figure 104.

MAPPING THE STORMS

On outline maps showing the principal geographical features and state boundaries within and adjacent to the area of the storm, each rainfall station within the storm area was located and its precipitation for the period considered was noted beside it. Isohyetal lines were then drawn at 1-inch rainfall intervals, down to and including the 2-inch line.

In appearance these rainstorm maps, figure 105, resemble topographic maps. If data were available to plat rainstorm maps with the same degree of accuracy with which good topographic maps are platted, there could have been no necessity for selecting one from several methods of platting the former. Unfortunately, however, several factors operate to reduce the accuracy of rainstorm maps, making it advisable to devise some means, if possible, to eliminate these inaccuracies. They may be divided into four classes:

FIG. 104.-FORM ILLUSTRATING THE METHOD OF COMPILING STORM DATA.

1916.

SOURCES OF ERROR IN ORIGINAL DATA

Class 1. Arbitrary Location of Observing Stations

Rainfall recording stations, being permanently located, rarely furnish data sufficient to develop precisely the shape of a rainstorm area. A rainstorm map platted from the available records depicts the storm no more accurately than a topographic map platted from random stadia observations would represent actual topography. In comparatively level country the difference between two topographic maps, one from well chosen stadia observations and the other from random stadia observations, would not be great. This also holds for the comparatively regular edges of a rainstorm map. In hilly country, however, the two topographic maps would probably not resemble each other closely; and it is very likely that this can be truly said of two rainstorm maps platted under analogous conditions. In recent years, and particularly in the more densely inhabited parts of the country, stations are comparatively close together, so that storms platted from their records more nearly represent the true conditions. It is noticeable that the closer stations are together, the more irregular appears to be what may be called the topography of the storm. Particularly is this so in the vicinity of the storm centers. For this reason the percentage of error in areas on which great depths fall is larger than in areas covered to less depths. Other things being equal, also, the percentage of error will be greater as the stations are farther apart.

Class 2. Inaccuracy of Records

In some cases the accuracy of the data is questionable. It is subject to two kinds of errors. First, errors due to failure to catch the precipitation properly; and second, personal errors, due to failure to read, record, transcribe, or tabulate correctly. By far the greatest number of observers are carrying on the work voluntarily, and hence it is not probable that the close attention is given to accuracy which is to be expected from paid observers. Occasional inconsistencies in the records were found, indicating errors from these sources, and it was difficult at times to decide whether or not to alter or discard certain data.

Class 3. Arbitrary Division of Time

Another and very important source of inaccuracy in determining the greatest rainfall for a given period is due to the fact that readings at each station are recorded at a stated time each day without regard to the time of beginning or ending of the storm. Under these cir

cumstances it is almost certain that a record of the greatest rainfall within a given length of time is seldom obtained. What is recorded as a 2 days' rainfall may be, and often is, found to be actually a storm of less than 24 hours' duration. A 24-hour record may likewise cover only a few hours of rainfall. Errors of this kind obviously are always in the same direction, and arithmetical averages derived from them, therefore, cannot be compensating. They are the more serious because they always indicate a greater length of time for a given precipitation, and therefore a smaller storm intensity, than the true one. The effect of this error, however, tends to be less for two days than for one, less for three days than for two, and in the more serious storm type extending over a period of several days the error is of little consequence.

Another factor which tends to reduce this error is the area over which a given storm occurs. Let us assume that the storm is moving in an easterly direction and causes a precipitation of 5.2 inches at station a lasting from 11 p. m. on May 18 to 11 p. m. May 19. In its eastward movement it reaches station b and causes a precipitation of 4.5 inches, lasting from 2 p. m. May 19 to 2 p. m. May 20. Although the precipitation lasts only 24 hours at each station, as stated above, it is recorded as a 48-hour storm at both stations if the readings are taken about 6 p. m. But from the time rainfall began at station a until it ceased at station b was 39 hours, and the runoff for the storm area including stations a and b was in many respects similar to that which would have been caused had the precipitation at each station been distributed uniformly throughout the 39 hours.

Class 4. Variations in Time of Observation

Another class of inaccuracies is brought about by the fact that readings at all stations are not taken simultaneously. Most of the readings are taken about sundown, but a considerable number are taken in the morning. A few, the so-called regular Weather Bureau records, are computed from continuous automatic records and cover the period from midnight to midnight. It was found that the last class agrees fairly closely with the evening readings at surrounding stations. No semblance of agreement could, however, be found to exist between evening and succeeding morning readings. As the evening readings are in the preponderance, they necessarily constitute the controlling data for the maps. The morning readings would have been discarded altogether had it not been that without them the remaining material was too scant to be useful.