On Using Historical Comparisons in Evaluating Cloud Seeding Operations

K. Ruben Gabriel Department of Statistics and Division of Biostatistics, University of Rochester, Rochester NY 14627

Search for other papers by K. Ruben Gabriel in
Current site
Google Scholar
PubMed
Close
and
Demetrios Petrondas Department of Statistics and Division of Biostatistics, University of Rochester, Rochester NY 14627

Search for other papers by Demetrios Petrondas in
Current site
Google Scholar
PubMed
Close
Restricted access

Abstract

Cloud seeding operations are often evaluated by comparing precipitation during operations with records of previous “historical” precipitation. Possible biases that can arise from such comparisons have been discussed elsewhere. This paper uses extensive worldwide precipitation records to examine whether the usual statistical techniques may be validly applied to comparisons of precipitation in successive years, as is done in the above evaluations of cloud seeding. It is concluded that the chance of finding a “significant seeding effect” in the absence of seeding is usually well above the nominal significance level used. We therefore recommend that P-values from such operational/historical comparisons be treated very cautiously, possibly by multiplying them by a suitable factor, e.g., a “5% significant” result of such a comparison should really be regarded as more nearly “10% significant.”

Abstract

Cloud seeding operations are often evaluated by comparing precipitation during operations with records of previous “historical” precipitation. Possible biases that can arise from such comparisons have been discussed elsewhere. This paper uses extensive worldwide precipitation records to examine whether the usual statistical techniques may be validly applied to comparisons of precipitation in successive years, as is done in the above evaluations of cloud seeding. It is concluded that the chance of finding a “significant seeding effect” in the absence of seeding is usually well above the nominal significance level used. We therefore recommend that P-values from such operational/historical comparisons be treated very cautiously, possibly by multiplying them by a suitable factor, e.g., a “5% significant” result of such a comparison should really be regarded as more nearly “10% significant.”

Save