We developed a methodology to predict
brook trout (Salvelinus fontinalis) distribution using
summer temperature metrics as predictor variables.
Our analysis used long-term ﬁsh and hourly water
temperature data from the Dog River, Vermont (USA).
Commonly used metrics (e.g., mean, maximum,
maximum 7-day maximum) tend to smooth the data
so information on temperature variation is lost.
Therefore, we developed a new set of metrics (called
event metrics) to capture temperature variation by
describing the frequency, area, duration, and magnitude of events that exceeded a user-deﬁned temperature threshold. We used 16, 18, 20, and 22C. We built
linear discriminant models and tested and compared
the event metrics against the commonly used metrics.
Correct classiﬁcation of the observations was 66%
with event metrics and 87% with commonly used
metrics. However, combined event and commonly
used metrics correctly classiﬁed 92%. Of the four
individual temperature thresholds, it was difﬁcult to
assess which threshold had the ‘‘best’’ accuracy. The
16C threshold had slightly fewer misclassiﬁcations;
however, the 20C threshold had the fewest extreme
misclassiﬁcations. Our method leveraged the volumes
of existing long-term data and provided a simple,
systematic, and adaptable framework for monitoring
changes in ﬁsh distribution, speciﬁcally in the case of
irregular, extreme temperature events.