Abstract:
|
We show how to analyze annual count surveys of using state-space models where the observation or sampling distribution is a lognormal distribution with an estimated variance, and the process distribution is a simple ARIMA time series model such as a random walk. Even though there is high correlation between the observation and process variances, we are able to obtain annual estimates and forecasts with standard errors. This is an improvement over the usual practice of smoothing with three-term moving average and using the final average as a forecast. We provide model diagnostics to discriminate among different state-space models, and we compare these models with moving averages using examples of Sandhill cranes, Atlantic brant, Cackling geese, and trumpeter swans
|