Use of generalized extreme value covariates to improve estimation of trends and return frequencies for lake levels

摘要

One of the most important tools in water management is the accurate forecast of long-termnand short-term extreme values for flood and drought conditions. Traditional methods of trendndetection are not suited for hydrologic systems while traditional methods of predicting extremenfrequencies may be highly inaccurate in lakes. Traditional frequency estimates assumenindependence from trend or initial stage. However, due to autocorrelation of lake levels, initialnstage can greatly influence the severity of an event. This research utilizes the generalizednextreme value (GEV) distribution with time and starting stage covariates to more accuratelynidentify trend direction and magnitude and provide improved predictions of flood and droughtnstages. Traditional methods of predicting flood or drought stages significantly overpredict ornunderpredict stages depending on the initial stage. Prediction differences can exceed one meter,na substantial amount in regions with flat topography; these differences could result in significantnalterations in evacuation plans or other management decisions such as how much lake water tonrelease in preparation for an approaching hurricane, appropriate lake levels to maintain, minimumnstructure floor elevations and more accurate forecasting of future water supply or impacts tontourism. The methods utilized in this research can be applied globally.