Testing the potential of streamflow data to predict spring migration of ungulate herds.

In mountainous and high latitude regions, migratory animals exploit green waves of emerging vegetation coinciding with rising daily mean temperatures initiating snowmelt across the landscape. Snowmelt also causes rivers and streams draining these regions to swell, a process referred to as to as the 'spring pulse.' Networks of streamgages measuring streamflow in these regions often have long-term and continuous periods of record available in real-time and at the daily time step, and thus produce data with potential to predict temporal migration patterns for species exploiting green waves. We tested the potential of models informed by streamflow data to predict timing of spring migration of mule deer (Odocoileus hemionus) herds in a headwater basin of the Colorado River. Models using streamflow data were compared with those informed by traditional temperature-derived measures of the onset of spring. Non-parametric linear-regression techniques were used to test for temporal stationarity in each variable, and logistic-regression models were used to produce probabilities of migration initiation. Our analysis indicates that models using daily streamflow data can perform as well as those using temperature-derived data to predict past-migration patterns, and nearly as well in potential to forecast future migrations. The best performing model was used to generate probabilities of onset of migration for mule deer herds over the 69-year period-of-record from a streamgage. That model indicated spring migration has been trending toward earlier initiations, with modeled median initiations shifting from a Julian day of 123 in the mid 20th century to Julian day 115 over the most recent two decades. The period of 1960 to 1979 had the latest modeled median initiations with Julian day of 128. The analyses demonstrate promise for merging existing hydrologic and biological data collection platforms in these regions to explore timing of past migration patterns and predict migration onsets in real-time.