Analytical Approach to Adjusting Reporting Changes

29 2008 ‐ present. Due to management actions implemented in 2008, which based daily catch limits on harvest history, Maryland experienced significant and systemic over‐reporting from the commercial fishery. This problem has continued through 2010. Instead of relying on catch reports, MDNR has been able to estimate commercial harvest using a combination of fishery dependent and fishery independent surveys. Catch per unit effort data from a sentinel fleet was applied to an estimate of effort number of crab pots generated by Versar Slacum et al. 2010. Details of the reporting issues and harvest estimate calculations can be found in the 2009 CBSAC blue crab advisory report Chesapeake Bay Stock Assessment Committee 2009. 5.1.3. Potomac River Fisheries Commission There have been no changes to how the Potomac River Fisheries Commission records its data. Data are reported by individual watermen on a daily basis. Prior to 1964, landings from the Potomac River were allocated back to either Maryland or Virginia according to which portion of the river from which they were taken.

5.2. Analytical Approach to Adjusting Reporting Changes

As documented above, there have been significant changes to how landings are reported and estimated in both states, although not in the Potomac River. Fogarty and Miller 2004 applied time series analysis to assess the impact of the 1981 reporting change in Maryland. They used a time series model with both an intervention term and a transfer function to represent underlying changes in abundance as measured by fishery ‐independent surveys. These authors concluded that the 1981 reporting change in Maryland had a significant impact on the landings reported. Miller et al. 2005 used a similar approach involving both intervention and transfer functions. These authors identified significant reporting impacts in Maryland in 1981 and in Virginia in 1993. The finding of a significant reporting effect of the 1993 change in Virginia is controversial. For this assessment, we again used time series analyses to quantify the impacts of reporting changes on estimates of landings in the commercial fisheries. However, importantly, we did not use a transfer function in these analyses. We chose to abandon the use of the transfer function because, to an extent, inclusion of the transfer function amounts to a stock assessment in its own right. Thus, we suggest that it is inappropriate to use landings time series that have been adjusted for both survey abundance estimates and reporting changes in a subsequent assessment model to estimate abundance. Such an application appears somewhat circular and likely biases results. Thus for this assessment, all time series approaches to correct for reporting changes in 30 landings time series used only intervention terms in classical Box‐Jenkins time series methodologies. The overall model can be written as t i i t t z B B I B c        Eq. 2 where c is the catch, Θ is a constant, B is the backshift operator, ω is an estimated parameter related to the impact of the intervention I which is a 0,1 variable whose value is 1 for years after the intervention and 0 in prior years, and θ and φ are polynomial parameters related to a moving average and autoregressive time series model that result from the model fitting so that the residuals from the model z are a pure white noise process. The approach to fitting was to first check the raw landings time series for stationarity. Where necessary the time series was differenced or otherwise filtered to achieve stationarity. The appropriate order of the moving average and autoregressive terms was then determined by using the auto.arima function in R v.2.11.1, which uses the Akaike’s Information Criterion AIC to determine the order for the two polynomial parameters that will give the best model fit and have residuals that do not significantly differ from a pure white noise process. The estimated regression parameters ω, θ, ϕ from the model fitting were sequentially tested to determine if the magnitude of the effect was significantly different from zero using a t‐test. If the effect was significant it was included and the ARIMA model rerun. In the case that an intervention was found to be significantly different from zero, the portion of the time series that occurred prior to the intervention was adjusted based on the most recent period of the time series. This assumes that current management strategies for reporting in each state are the most accurate and therefore the landings from this period are the most reliable. Assessment of stationarity and the intervention analyses were conducted in R v. 2.11.1.Appendix II ‐‐ R Core Development Team 2007. Both Virginia and Maryland time series had to be 1 st order differenced in order to achieve stationarity and significant interventions were found for both states. Details of the results of the reconstructed landings are presented in section 5.3.

5.3. Reconstructed Commercial Landings