Estimation and Correction of bias of long-term simulated climate data from Global Circulation Models (GCMs)

Global circulation models are often used in simulating long-term climate data for use in hydrologic studies. However, some bias (difference between simulated values and observed data) has been observed especially while simulating precipitation events. The bias is especially evident with respect to simulating dry and wet days. This is because GCMs tend to underestimate large precipitation events with the associated precipitation amounts being distributed to some dry days, thus, leading to a larger number of wet days each with some amount of rainfall. The accuracy of precipitation simulations impacts the accuracy of other simulated components such as flow and water quality. It is, thus, very important to correct the bias associated with precipitation before it is used for any modeling applications. This study aims to correct the bias specifically associated with precipitation events with a focus on the Western Lake Erie Basin (WLEB). Analytical, statistical, and extreme event analyses for three different stations (Adrian, MI; Norwalk, OH; and Fort Wayne, IN) in the WLEB were carried out to quantify the bias. Findings indicated that GCMs overestimated the wet sequences and underestimated dry day probabilities. The number of wet sequences simulated by nine GCMs each from two different open sources were 310-678 (Fort Wayne, IN); 318-600 (Adrian, MI); and 346-638 (Norwalk, OH) compared with 166, 150, and 180, respectively. Predicted conditional probabilities of a dry day followed by wet day (P (D|W)) ranged between 0.16-0.42 (Fort Wayne, IN); 0.29-0.41(Adrian, MI); and 0.13-0.40 (Norwalk, OH) from the different GCMs compared to 0.52 (Fort Wayne, IN and Norwalk, OH); and 0.54 (Adrian, MI) from the observed climate data. There was a difference of 0-8.5% between the distribution of simulated climate values and observed climate data for precipitation and temperature for all three stations (Cohen's d effective size < 0.2). Further work involves the use of Stochastic Weather Generators to correct the conditional probabilities and better capture the dry and wet events for use in the hydrologic and water resources modeling.