Some manufacturing processes could benefit from modelling; a typical objective would be to optimize a function of the output such as the cost per unit production. Distribution and logistics are other areas where optimization could deliver greater profits. The list really is endless. If there is a process involved, then it can be modelled; if there are significant costs associated with the process then a key question for the business is how can the costs be minimized. It may be that the best way to answer that question is to model the process and experiment with the model. The alternative of experimenting with the actual process may be a very costly error.
At ISL we have extensive modelling experience and we are interested
in understanding all manner of processes. If you have a problem where you
think modelling could help, please get in touch. We would be happy to discuss
your problem and let you know how we might help.
One of the most important advantages of having a mathematical
model of a process is that one can experiment with the model rather than the process. Invariably the
experiments with the model are best done on a computer. The power of modern
computers allows extensive simulations with even very complex models to
be done very quickly once the model has been programmed. Of course, for the
experimental results of a computer simulation model to be of any practical use,
it is crucial that the model is "ground-truthed" with real (rather than simulated)
data. At ISL we have considerable experience in developing and ground-truthing
models from existing data sets and in survey design for the collection of additional
data. Computer models can often be used to indicate what additional data would be beneficial
for refining an existing model to make it more realistic and more useful.
The first step in an "ideal" statistical data analysis is an experimental or survey design, where the methods used to collect and analyze the data are determined. If this is done properly then the data analysis is usually straightforward; although problems can arise when the assumptions of the survey design are found to be invalid. The more common situation (unfortunately!) is where there is an existing data set which needs to be analyzed in an attempt to answer some questions, or to estimate some parameters of interest. The data have often been collected in a non-random fashion which can make the analysis somewhat challenging (as "standard' statistical assumptions are usually inappropriate). However, whether you have some data already, or want to collect some for a statistical analysis, we would be happy to discuss your particular problem.
At ISL we have written programs in numerous languages and have been involved with computers
for most of our working lives. If you have existing code which needs to be modified or you
have some specific programming requirements we can probably help. We have particular expertise with C
and C++.
All ISL directors are familiar with the "scientific method" and reading and writing scientific
documents. Our particular speciality is in fisheries science, but with training in the generic sciences of mathematics, statistics, and information technology we have skills relevant to many scientific disciplines.
We have a great deal of experience in fisheries science, particularly with regard to quantitative areas of research.
Underlying every stock assessment there is a population model. Even if the model is not explicitly used in an assessment, the assessment method will be justified on the basis that under certain assumptions with a particular population model, the method will produce sensible results. We are familiar with many different types of models, with a particular speciality in age-structured deterministic observation error models.
The main motivation for creating a population model for a commercial fish stock is to enable a quantitative stock assessment for the stock. There are three components in a stock assessment: the population model, the data, and the estimation procedure. In New Zealand, age structured deterministic observation error models are almost exclusively used (with stochastic components introduced primarily in forward projections for "risk analysis"). Data available for stock assessments range from basic biological information on growth and stock structure to abundance indices from trawl and acoustic surveys. Estimation methods vary depending on the preferences of the scientists performing the stock assessment. The single most common method used in the late 1990s was MIAEL estimation (developed by ISL director Patrick Cordue) which was used to assess hoki and several other important commercial species. Since 2001, (when the last MIAEL stock assessment of hoki was done) Bayesian methods are now almost exclusively used.
ISL directors have experience with acoustic and trawl survey design, the collection of fisheries statistics, and the design of catch sampling programmes.
ISL director Roger Coombs has experience in all aspects of underwater fisheries acoustics having been involved in the development and use of acoustic systems for quantitative stock assessment for about the last thirty years.