about one-half of this stream is used by pink 

 and chum salnnon. 



Maybeso Creek is 20 to 60 feet wide, and 

 has a watershed of 15 square miles. More 

 than 5 miles of stream are accessible to 

 salmon. Average gradient is probably higher 

 than for the other study streams (James, 1956, 

 gives the gradient for 500 feet as 4.66 per- 

 cent, but this high gradient does not prevail 

 upstream from the area measured). This 

 stream has an obstruction which may block 

 salmon. A stepped falls at the mouth has, on 

 periods of low stream discharge and small 

 high tides, caused pink and chum salmon to 

 die unspawned because they could not enter 

 the stream. 



Twelvemile Creek is 30 to 70 feet wide and 

 has a watershed of 14 square miles. About 

 5 miles are accessible to salmon, but spawn- 

 ing occurs mostly in the lower reaches. 

 Average gradient for 8,750 feet from lower 

 intertidal zone upstream is 0.12 percent (per- 

 sonal communication from G. A. James, Alaska 

 Forest Research Center). 



Old Tom Creek has a watershed of 7.5 

 square miles. It is the only stream of the five 

 that is connected to lakes, of which there are 

 two: one is 85 acres, the other 62 acres. The 

 streann forks about a mile above tidewater. 

 Average gradient is about 0.79 percent (James, 

 1956). 



VARIABILITY IN ESTIMATES 



Counts nnade on foot surveys give rough 

 indices of numbers of salmon using the 

 streams. Foot surveys have usually been 

 conducted in Alaska by making two to five 

 visits to a stream during a spawning season. 

 Stream surveyors walk up a stream and esti- 

 mate total numbers of live salmon in pools 

 and on riffles. Spawners are not a static pop- 

 ulation because they continually move 

 in and die throughout the season. Therefore, 

 although periodic counts give a point estimate 

 in time, a peak count is a reliable index only 

 insofar as escapement curves are the same 

 shape year after year. However, if we know 

 the average duration of life of spawning 

 salnnon and if we confine counts to riffles, 

 by conducting several surveys we can correct 

 for this type of error (Gangmark and Fulton, 

 1952). Counts must be restricted to riffles 

 because (1) nunnbers of fish in deep pools 

 cannot be estimated reliably and (2) variabil- 

 ity in length of life of salmon in pools is 

 probably greater than variability in length of 

 life of salmon after they commence spawning 

 and are on the riffles (we have observed 

 tagged pink salmon in pools for weeks, while 

 average length of time on riffles is 5 to 12 

 days). 



In this paper differences between estimates 

 and the true population are not considered. 

 Discussion is limited to variability between 

 estimates of different observers and between 

 successive estimates by the same observer. 

 Although this type of variability has been 

 discussed by Bevan (1961) for aerial surveys, 

 it has not been evaluated for foot surveys. 



Testing Procedures 



Tests of variability were conducted under 

 two sets of conditions in Mollis area streams 

 in 1956 and 1957. The first set ofconditions is 

 termed "ideal" and the second, "nornnal." 



In the first set counts were made only on 

 shallow spawning riffles where salmon were 

 clearly visible, boundaries were clearly de- 

 fined, and there was little in and out movennent. 

 Counting was done from the bank so that fish 

 would not be disturbed. Each observer made 

 a series of counts, one immediately following 

 the other; each used the same path and had 

 the same angle of vision. Light or other 

 conditions changed little in the short time 

 needed to complete an experiment (usually 

 less than one-half hour). In most cases 

 Polaroid glare shields were worn. 



In the second set counts were made on 

 either a large part or the entire length of a 

 stream that included both pools and riffles. 

 This method is similar to that used on routine 

 foot surveys in the past. 



Observers were coded by number in all 

 tests. Veeder-Root hand counters with the 

 tabulation covered with tape were used, and 

 results were recorded without informing ob- 

 servers of the nunnber on the counter. All 

 observers were briefed on proper procedure. 



Resiilts of First Set of Tests 



Basic data, mean, standard deviation (s), 

 and coefficient of variation (C.V.) for four 

 tests conducted under ideal conditions show 

 the variation to be expected between observers 

 and between successive counts by the same 

 observer when 30 to 250 salmon are on a 

 riffle (table 1). When the number of salmon 

 to be estimated is low, variation is high. It 

 decreases with increasing numbers of salmon 

 up to a point and then again increases with 

 increasing numbers of salmon. 



Analysis of variance (table 2) shows a signifi- 

 cant F value (95 percent) only once out of 

 eight times, and this in relation to difference 

 between observers in test IB. In all four 

 experiments variance ratio was higher between 

 observers than between successive counts by 

 the same observer. 



