Salmonid thermal habitat contraction in a hydrogeologically complex setting

Broadening our understanding of river thermal variability is of paramount importance considering the role temperature plays in aquatic ecosystem health. At the catchment scale, spatial statistical river network models (SSN) are popular for analyses of river temperature, as these are less "data hungry" than other modeling methods, and have offered invaluable insights into how thermal habitats of salmonids may change with climate warming. However, recent work has demonstrated that hydrogeological complexity can disrupt river temperature spatial autocorrelation. We test the prediction that the non-linearity of hydrological processes inherent in a hydrogeologically complex setting, such as the Miramichi River, invalidates the SSN approach, and a Random Forest (RF) model can overcome these complexities. In all instances, RFs outperformed SSNs when predicting average (T-wA) and maximum (T-wM) August river temperature during 2017, and were quite robust (T-wA and T-wM: R-2 = 0.93; RMSE = 0.6 degrees C; R-2 = 0.91; RMSE = 1.0 degrees C, respectively). We conclude that RF models can capture the inherent non-linearity of hydrological processes in complex hydrogeologic settings. We examined thermal habitat change for adult and 1+/2+ Atlantic salmon-AS-(Salmo salar), and all age classes of brook trout-BKT-(Salvelinus fontinalis), during August 2017, with thresholds of behavioral thermoregulation specific to the catchment. We assumed a baseline = T-wA and investigated river network contraction (km) for T-wM. During T-wA,T- all habitat was suggested to be thermally suitable for 1+/2+ AS (<23 degrees C), but 4.2% was unsuitable for adult AS and BKT of all ages (>20 degrees C). For T-wM,T- similar to 80% of the catchment was predicted to be unsuitable for adult AS and BKT. We examined two boundaries for behavorial thermoregulation in 1+/2+ AS: >23 degrees C and >27 degrees C. For the >23 degrees C boundary, similar to 27.7% of the catchment is thermally unsuitable during T-wM, and 4.9% is thermally unsuitable for the >27 degrees C boundary. T-wA in August 2017 was identical to long-term (1970-1999) July-August T-wA, as such these thermal maps will be useful for resource managers.