achieve the desired reliability, thus significantly reducing the cost of data
collection (while at the same time possibly increasing the complexity of the
analysis). For any specified experimental design, the optimum number of
samples is a function of the variability of the measurements, the minimum
effect size to be detected, the alpha level that is considered significant, and
the power of the test (the ability of the test to determine when there are
no significant effects) (Kirk 1995). However, different designs will be better
able to control for effects like external variables, differing target populations, spatial autocorrelation, and treatment effects (Cressie 1993). If the
variability is not available from previous studies, pilot studies, iterative sampling, or test sampling can provide estimates of the variability of the features of interest. Once the number of samples has been decided, the actual
samples must be selected probabilistically from the populations of interest.
The randomization method will already be determined based on the design
selected.
10.3.4.4 Protocols
Following standard protocols for field sampling can enhance the applicability of the data to multiple modeling projects and can allow the estimation of measurement error. By maintaining consistency in field methods,
data from multiple studies can be readily compared and combined in a
model. Although measurement error is inevitable in field data collection,
standard sampling protocols have generally been designed to minimize
measurement error. In many cases, previous studies have been conducted
on standard sampling methods to characterize the quantity and variability
of the measurement error.
Overall, the collection of field data is a necessary method of obtaining
data for modeling when sufficient and appropriate data are not available.
The expense and time can vary significantly depending on the data needed.
The key to successful field data collection is having a clear objective of what
data are necessary for the model. With specific data in mind, and with
the help of an experienced statistician, an experimental or observational
study can be designed that will efficiently obtain the data with minimal
measurement error. In this case fewer assumptions are passed along to the
ecological model.
10.3.5 Data Simulation
Another method of obtaining data for ecological models is through the use
of computer simulation. Data sets with known parameter values make it
possible to test the accuracy of model outputs. Simulated landscapes have
been used to develop generalized ecological models of landscape processes
(e.g., With 1997). Simulated data are also useful for testing the sensitivity
of the model to assumptions related to the model structure or the data used
10. Effective Ecological Modeling: Data Issues
193
collection (while at the same time possibly increasing the complexity of the
analysis). For any specified experimental design, the optimum number of
samples is a function of the variability of the measurements, the minimum
effect size to be detected, the alpha level that is considered significant, and
the power of the test (the ability of the test to determine when there are
no significant effects) (Kirk 1995). However, different designs will be better
able to control for effects like external variables, differing target populations, spatial autocorrelation, and treatment effects (Cressie 1993). If the
variability is not available from previous studies, pilot studies, iterative sampling, or test sampling can provide estimates of the variability of the features of interest. Once the number of samples has been decided, the actual
samples must be selected probabilistically from the populations of interest.
The randomization method will already be determined based on the design
selected.
10.3.4.4 Protocols
Following standard protocols for field sampling can enhance the applicability of the data to multiple modeling projects and can allow the estimation of measurement error. By maintaining consistency in field methods,
data from multiple studies can be readily compared and combined in a
model. Although measurement error is inevitable in field data collection,
standard sampling protocols have generally been designed to minimize
measurement error. In many cases, previous studies have been conducted
on standard sampling methods to characterize the quantity and variability
of the measurement error.
Overall, the collection of field data is a necessary method of obtaining
data for modeling when sufficient and appropriate data are not available.
The expense and time can vary significantly depending on the data needed.
The key to successful field data collection is having a clear objective of what
data are necessary for the model. With specific data in mind, and with
the help of an experienced statistician, an experimental or observational
study can be designed that will efficiently obtain the data with minimal
measurement error. In this case fewer assumptions are passed along to the
ecological model.
10.3.5 Data Simulation
Another method of obtaining data for ecological models is through the use
of computer simulation. Data sets with known parameter values make it
possible to test the accuracy of model outputs. Simulated landscapes have
been used to develop generalized ecological models of landscape processes
(e.g., With 1997). Simulated data are also useful for testing the sensitivity
of the model to assumptions related to the model structure or the data used
10. Effective Ecological Modeling: Data Issues
193