Modelling Catchment-Scale Responses to Climate Change - Climate Change Impacts on Freshwater Ecosystems - page 243

forecasting (Araújo & New 2007) by applying several different models to the
same problem and evaluating areas of agreement and disagreement. As described
below, Buisson et al . (2008) adopted such an approach to model the future
distribution of 35 fish species in French streams.
Seven different statistical techniques were used. Fish presence-absence data
were extracted from the Office National de l'Eau et des Milieux Aquatiques
(ONEMA) for 1110 sites distributed over the whole country. The sites chosen
were judged to have minimal human disturbance, and the data were obtained by
standardized electrofishing techniques. Data from fish species present in more
than 25 sites were retained for study, amounting to 35 species (see Table 10.3).
These data were then related to climatic and environmental variables. Three
variables were used to describe climatic conditions: mean annual precipitation,
mean annual air temperature and annual air temperature amplitude, derived from
the difference between mean air temperature of the warmest month and mean air
temperature of the coldest month. Future values for each of these three descriptors
were derived for the 2080s from three GCMs: HadCM3 (Hadley Centre for
Climate Prediction and Research, the United Kingdom), CGCM2 (Canadian
Centre for Climate Modelling and Analysis) and CSIRO2 (Commonwealth
Scientific and Industrial Research Organisation, Australia). Predictions of future
climate were made for each of these using four greenhouse gas emission scenarios.
These were scenarios A1, A2, B1 and B2 from the IPCC SRES (Naki´enovi´ et al .
2000). Thus, 84 separate modelling runs were performed for each fish species:
seven species distribution models × three GCMs × four emission scenarios.
The study also recognized that climatic variables are not the only ones to
affect fish distribution: rivers and streams have a variety of habitats, and a large
lowland river will clearly be different from a small headwater stream even if the
climate is the same. Some of this habitat variation is correlated with information
that can be obtained from databases and thus can be taken into account in the
analysis. In this study, three derived variables were used to represent habitat
variation: (i) elevation above sea level; (ii) a parameter representing position on
the upstream-downstream gradient, derived from catchment area and distance
from the source and (iii) stream velocity derived from width, depth and slope at
the sampling site. As these derived variables are likely to be correlated with the
climatic ones (e.g. high elevation with low annual air temperature), the deviations
from the expected values at each site were used in the analysis to give six
independent variables.
The species distribution models were then calibrated for each species using a
randomly selected 777 river sites from the database. The remaining 333 sites
were used to validate the calibration in an iterative process, and when these were
satisfactory, predictions of probability of occurrence of each species at each of
1110 sites were converted into presence-absence values using a threshold
maximizing the sum of two measures: sensitivity (i.e. the percentage of presence
correctly predicted) and specificity (i.e. the percentage of absence correctly
predicted). The calibrated models were then used to predict fish species
distributions for 2080 for each of the 12 scenarios. The future probabilities of
occurrence were transformed into presence-absence values by using the same
threshold values as for current predictions.