el vehículo.
1. Relatividad 1 Antecedentes
1.2 Postulados de la teoría de la relatividad
Predictive modelling using regression trees revealed that data collected using a set of protocols was more accurate than data collected without using a set of protocols, demonstrating the importance of using consistent and tested sampling methods. The Waikato Regional Council dataset was collected using a set of protocols which are effectively the New Zealand Freshwater Fish Sampling Protocols (Joy et al., 2013). These protocols were put in place to provide guidelines for freshwater ecologists when undertaking sampling or monitoring of fish species within New Zealand. The guidelines were intended to produce reliable datasets where actual differences in species populations can be examined while minimising the methodological sampling noise. As described, the National Institute of Water and Atmospheric Research (NIWA) holds a national dataset (the NZFFDB) consisting of thousands of entries of fish samples throughout the country with a time range of close to a century. This database is a great resource for examining historical fish presence, however there is comparatively little spatial or temporal consistency in sampling within any of the records in the dataset, and often there is no record of how a site may have been sampled. Further inconsistencies include records where no fish were caught during
Species NZFFDB correlation NZFFDB deviance WRC correlation WRC deviance Anguilla dieffenbachii 0.267 11.31 0.679 0.019 Anguilla australis 0.384 52.733 0.474 0.039 Gobiomorphus huttoni 0.382 50.019 0.694 0.051
sampling not having been entered, or conversely, records where fish were caught also not being entered ultimately making the understanding of populations less clear. This high degree of variability begins to affect the level of accuracy of analyses and interpretations of fish populations.
Consistency of sampling is of course vitally important and a central tenant of any sampling design and this particularly comes in to play when making relative comparisons between years at a site for a given species. The level of consistency in sampling that can be achieved through implementing standardised sampling methods is high (Figures 2.2 and 2.3). The type of consistency typical in NZFFDB data (Figure 2.4), from which comparisons and analyses of New Zealand fish species have frequently been made both temporally and spatially, and in relation to environmental variables is comparatively low (e.g. Jowett & Richardson (2003)). Many species, such as eels, are already difficult to monitor in terms of total population as different cohorts within the population inhabit vastly different habitats (McDowall, 1990). For example, the large majority of small eels will remain in the sediment of a waterway to avoid predation (Glova, 2001). Netting methods will typically underestimate the numbers of these smaller individuals. Conversely, large eels will typically inhabit deep slow moving water, and areas with in-stream debris, for which electric fishing is often ineffective, underestimating large eels in the population (Baillie, Hicks, van den Heuvel, Kimberley, & Hogg, 2013; Chisnall & Hicks, 1993; Jellyman, Chisnall, Sykes, & Bonnett, 2002). The type of waterway and sampling methodology then can be strongly influential on population estimates. Multiple method use and varied fishing effort, as is generally the rule in the NZFFDB, begins to introduce additional types of unnecessary errors. Consequently, conclusions about fish species from the NZFFDB dataset may be inaccurate with little opportunity to determine how inaccurate.
Consistency in sampling effort and methods has been achieved with the above WRC dataset through strict governance of fishing area and time. A national dataset that was collected in the same manner as the WRC dataset would provide substantially improved predictions for species across the wide range of ecosystems present in New Zealand. Such a dataset would cover a far greater range of REC classes and waterway types than the WRC dataset has done. The reduction in noise around model building and predictions gained from consistent sample collection would allow for a far greater understanding of native fish populations at a national level.
Statistical methods will continue to improve accuracy and versatility of making predictions in species occurrence and abundance however models are only as accurate as the datasets they are based on (Clark, 2004; Cohen, 1988). Further, the opportunities for a greater detail of analyses becomes apparent when sampling error is minimised. For example, Indices of Biotic
Integrity (IBI) which are increasingly used to assess the health of waterways can offer a detailed picture of stream communities and health (Joy, 2005; Joy & Death, 2004; Stark & Maxted, 2007). Being able to use abundance IBIs produces far more accurate picture of the stream community, however this is predicated on reliable abundance data. This has been shown previously in New Zealand freshwater fish data where reduction in variance was observed when using abundance IBI calculations versus presence/absence IBI calculations for the same dataset (Joy, 2013).The NZFFDB provides an immense data resource with thousands of entries for fish samples which represent the sampling effort of many individuals, councils and/companies. It is a shame that the usefulness of such a resource is limited somewhat by variable sample collection and that even presence/absence analyses may be subject to sampling noise. Differences between councils and anthropogenic regions in regards to objectives for freshwater monitoring introduce further noise and to accurately monitor aspects such as relative abundance and reach scale diversity on a national or regional scale, protocols such as those used by Waikato Regional Council in collecting this dataset should be followed. These protocols have not only created a more accurate dataset, but have provided a wider range of metrics able to be analysed, that are important in understanding fish populations.
Models built using relative abundance data also proved to be more accurate than those built using presence/absence data from the WRC dataset and were better able to predict whether a site was an impact or reference site. This improvement again highlighted the benefits of collecting consistent abundance data. Model improvements were small although this may be related to a relatively small dataset in particular the small number of available reference sites. Assessing the performance of relative abundance versus presence/absence models in years to come following the collection of a larger dataset, may illustrate greater improvements in model performance with similar analyses when using relative abundance data.
In conclusion, consistency in data collection provides several benefits to understanding native fish populations. Freshwater ecologists can be confident that relative differences observed in populations between years are actual rather than due to sampling methodology. The improvements in model performance when using data from the WRC dataset versus those from the NZFFDB further demonstrate the types of gains that may be achieved by employing a consistent set of protocols. It is recommended, with demonstrations throughout this thesis, that protocols similar to these be adopted nationwide in all sampling efforts. This approach will provide robust and accurate datasets through which populations of native fish can be more effectively managed by utilising a more accurate understanding of population dynamics, seasonal changes, habitat preferences, and ecosystem interactions between and within species.