Abstract:In order to investigate the performance of Rank Set Pair Analysis (RSPA) in predicting groundwater dynamics and its response to the abovementioned two parameters

a series of scenarios were carried out by using the long-term groundwater table records in the Jingtai site of the Baiyin county

Gansu Province. First

the size of historical set T was assigned as one of eight different values ranging 3 to 15 respectively. The effect of historical set size on prediction error was studied

in the case of single subsequent value. Then

by keeping historical set size T fixed

the influence of number of subsequent value n on prediction error was investigated. The study revealed that

firstly

when historical set size was less than 8

the error was steady and acceptable and when the size became greater than 8

the error might remarkably increase; secondly

RSPA performed well in predication of groundwater dynamics within a short term

however

poorly in predicting a long-term dynamics of groundwater; thirdly

RSPA had serious disadvantage in error accumulation occurred in every step; and fourthly in comparison with single subsequence value

the multi-subsequence value could make the accumulation of error of RSPA increases slowly

yet

it still could not fulfill the requirement of predicting a long term groundwater dynamics.