An Application of Statistics in Climate Change: Detection of Nonlinear Changes in a Streamflow Timing Measure in the Columbia and Missouri Headwaters

Detecting the impact of climate change in natural systems is typically hampered by limitations caused by short time series, especially for instrumental records. Many researchers trying to assess change have assumed that the change is linear over time. The models used in this work incorporate correlations over space and time and compare models that assume linearity to those that estimate trends nonlinearly using a nonparametric regression technique. This chapter discusses a method to assess the evidence for nonlinear trends in the timing of streamflow in the headwaters of the Missouri and Columbia Rivers. Streamflow timing refers to the arrival date of characteristics of the yearly distribution of streamflow, most often the mean or median of the flow. The day of the median or when half the yearly flow has occurred is considered and changes in that timing from 1951 to 2005 are shown. Streamflow timing in gage locations are high elevation and not disturbed by irrigation practices, should primarily record climate change induced trends in snowmelt. The conjecture is that earlier timing measures should be observed over the study period due to warmer temperatures causing earlier snowmelt. The details of this trend and statistical evidence regarding it are the main concerns of this work, which extends into nonparametric spatial-temporal models.