PARENT SESSION
Oral Session 16: Statistics and Biometrics.
Presiding: E Garton and T Simons
Monday, August 2, 8:00 AM to 11:30 AM, Meeting Room D 139.
Estimating variation in landscape analyses: To err is human, to report it is divine.
Hummel, Susan*,1, Cunningham, Patrick2, 1 PNW Research Station, Portland, OR, USA2 PNW Research Station, Corvallis, OR, USA
ABSTRACT- The variation we discover while exploring inhabited landscapes, whether in form or function, is a legacy of the past and a basis for future conditions. Computer models are used increasingly to describe landscape dynamics, test scale-related hypotheses, and evaluate potential treatment effects over space and time. In combination with mapping software, it is now easy to simulate and to display spatial change for many metrics. However, because a variance estimator is usually not available for the results of simulation, it is uncommon for spatially explicit landscape analyses to include information on the precision of their predictions. Yet it is vital for scientists to report a range of outcomes and the probabilities associated with them, especially when results are considered in land use decisions. Our objective was to predict the mean area of late seral forest structure over 30 years in a 6070 ha landscape in Washington State and to estimate associated confidence intervals for each decade. We stratified and sampled the area to measure forest structure, dynamically simulated forest growth, and then empirically approximated the distribution of late seral forest in each stratum in each decade using 200 bootstrapped samples. These bootstrapped distributions enabled us to select and assign the probability of transition to late-seral forest structure randomly for each patch in each stratum using 5000 iterations of Monte Carlo simulation. By the third decade, the mean area of late seral forest structure was 3673 ha, with a standard error of 1.54 ha; over half of this structure was predicted to develop from one of the original stratum.
Key words: forest structure, bootstrap sampling