PortalForecastsAug516.csv (16.74 kB)
Portal Rodent Trapping Forecasts on 5 August, 2016
dataset
posted on 2016-08-06, 01:21 authored by Weecology GroupWeecology Group, Glenda YenniGlenda YenniThis is the second forecast made for the rodent trapping at the long-term Portal research site in southern Arizona. Three models were used to achieve forecasts with 90% prediction intervals. Forecasts are for the site as a whole (total means across all plots, all treatment types).
Two naive models using the defaults in the forecast package in R were used to create forecasts for the total granivore captures (species="Total") over the next 12 trapping periods, the first being tomorrow.
The "Forecast" model
forecast(abundances$total,h=12,level=0.95,BoxCox.lambda(0),allow.multiplicative.trend=T)
The "AutoArima" model
forecast(auto.arima(abundances$total,lambda = 0),h=12,level=0.95,fan=T)
Species-level predictions, as well as another total captures prediction, were made using the tscount package in R.
The "NegBinom Time Series" model, by species (species codes are the first letter of the genus and species)
model=tsglm(abundances[[s]],model=list(past_obs=1,past_mean=11),distr="nbinom")
predict(model,12,level=0.9)
There are also total captures predictions created from the individual species predictions and their prediction intervals (model="NegBinom Time Series", species="SpeciesSums").
