Results of using climate-oceanological predictors in a Random Forest model for the forecast of stock abundance dynamics of pink salmon runs in Western Kamchatka

Abstract

An improvement of the forecast of pink salmon runs is one of the most actual problems in management of salmon stocks of Kamchatka. Algorhythm of Random Forest with the use of iterative selection of most important climate-oceanological predictors seem to be promising. Three authenticly influental predictors have been found for the model of Random Forest at the error of the forecast of 28%. Verification of the obtained results used alternative methods of selection of the best predictors. Obtained results have demonstrated possibility of the use of considered climate-oceanological predictors in modeling the forecasts of stock abundance dynamics of pink salmon in Western Kamchatka.