Correlogram Analysis Revisited

Consideration of the Hurst coefficient h offers new insight into the interpretation of observed correlograms for streamflow sequences. Autoregressive models, for which h = ½, cannot reproduce the structures of those correlograms. The structures are symptomatic of long‐term persistence as indicated by observed values of h being greater than ½. The tendency has been to test a sequence for independence, and if the hypothesis of independence is unacceptable, then (1) the generating process for the sequence is approximated by a short memory process, and (2) variations in the high lag serial correlation coefficients are ascribed to chance. If indeed the sequences are generated by long memory processes, then the powers of most independence tests are low for short records. Consequently, correlograms may be useful indicators of long‐term persistence when more formal tests give contrary results.