WMOD00519

• Project Name: Weather Modification Programme - Precipitation Enhancement Project
• Title: A Review of the Hydrological Aspects of Evaluation of Precipitation Enhancement
• Date: 5/1/1977
• Weather Modification - Doc Type: Report

7

31. For the Skagit River at Newhalem, six explanatory variables, (three stream-
floi'l, h,o summer precipitation, and one temperatm:e) survived the 0.10 significance
level screening. The resulting prediction equaiion produced estimates of streamflow
that were highly correlated with observed values, When applied to the seeded years
of 1963 and 1964, the indicated increase in streamfloi'l due to seeding Has significant
at the 0.01 level. This ,-,as largely due to the 1964 record exhibiting a significance
level of 0.005 since the 1963 record "laS not significant at the 0.05 level. The
estimated increment of streamflow in 1964 was 352,000 acre-feet (4.34 x 108m;>). The
nearly ideal conditions encountered in the Skagit River evaluation are rare in the
history of cloud-seeding.

v

TESTS EHPLOYED IN STATISTICAL EVALUATION

32. This reviev' is not intended to cover the complex subject of the statistical
design of experiments, but some consideration of tests suited for use on strE~amflo'.,
data may be helpful.

33. Harkovic (1966) offers six tests for statistical evaluation using river
flovr. Annual river flov' was the only variable used in his studies. Discrimination
of change was accomplished. by using the mean and variance. An increase in flow due
to effective cloud-seeding vrill, if repeated over a period of years, b'a refl(3ctecl as
an increased mean flo,., for the seeded period as compared with the mean flo'-1 for the
unseeded period of record. If the seeding has produced a time or space redistribu-
tion of annual flow, a change in the variance v,ou.ld indicate that effect. The data
available and the change in river flow expected provided the different sets of
conditions on which design of a particular test method ''laS based. The detailed
statistical discussion Hill not be repeated here, but the object of, and the under-
lying assumption for, each test are listed below. The basic time unit of seeding
operation is the sto=' s duration. The period over ",hich the streamflov' reflects
a given storm may vary from a fevr days to several months. The tests supplied by
Markovic are best suited to comparisons of aru1ual flows.

33. Target sample u-test: The object of this test is to compare the slaeded
period mean river flo"\., of a target vratcrshed vii th the population mean of nonseeded
flows from the same watershed. The basic assumptions underlying application of the
u-test are:

(a) the annual river flows of the target watershed are normally
distributed; .

(b) the annual observations are stochastically independent;

(c) the population parameters are known.

The first two assumptions are not always satisfied (ThOro, 1957), but Markovic found
that 72 per cent of 446 samples of river flol" were normally d1stributed with a 0.95
confidence probability, and that the assumption of stochastic independence is
usually justified ~lere serial correlation analysis shoHs weak dependence. The
third assmnption is rarely met and requires a very long period of observation
(100 years or more) to be completely accepted.

34, !.arget d.oub1e s~le t-test: The object of this test is to compare the
mean river flo"\'lS of a target i'la tershed for non-seeded and seeded periods. The basic
assumptions Ul1derlying this test are:

(a) the annual river flowa of the target '-Ja tershed are normally
distributed;
(b) the annual flows are stocha,stically indepemlent;