Forest inventory data can offer huge numbers of plots that pose challenges not encountered with traditional data. In this first of three units for forest inventories we show that models fitted to large networks having many small plots make noisy predictions, but they can hold as much information as smaller networks of few large plots.
Resources
Forest inventory vignette A
Data collection for structured inventories
For today
- Breakout to the discussion problem from last time:
Exercise. My budget can no longer support the costs of monitoring a full network. I can reduce costs by limiting the number of plots, the sizes of plots, or both. For this study it is important that the network can provide estimates of the effects of predictors on tree abundance. Using the FIA aggregated data as a model, determine the impact of sample size \(n\) versus effort \(E\) (plot area) on estimates of effects from standAge, meanTemp, and annualPrec in xdataCluster for a species of choice.
-
Fit the model with
glmusingfamily = poisson, see code block below. -
For small \(n\), refit with a
fractionof the plots. I could select them randomly using thesample( n, n*fraction), wherenis the number of plots. If you do this, remember that \(x\) and \(y\) must maintain their alignment. -
For small \(E\), refit the model with a sample of trees from each plot, as would occur it plot areas were reduced. I can do this with the
function rbinom( n, y, fraction ), whereyis the count for each plot (see below).
- Concepts for the second Forest inventory vignette B
For next time
- forest inventory vignette B