Analysis
We removed any days with missing data, which generally occurred because of equipment malfunctions, such as dead batteries or water damage. When possible, we used the redundancy of the stream sensors or camera data to fill in missing data gaps. Camera photos were used to measure stream flow depth to 0.05 m increments on an hourly time scale. We summarized each measurement on the daily scale, for example mean daily stage height, as an input in our calculations.
We selected five ecologically relevant and informative flow metrics encapsulating all five major components of a flow regime (Poff et al. 1997): (1) skew in daily stage or the mean daily stage divided by the median daily stage (magnitude), (2) Julian date of first no-flow day (timing), (3) number of no-flow days (duration), (4) high pulse count, where a pulse is defined as a flow event greater than 3 times the median stage (frequency), and (5) number of reversals, or when flow rates change from rising to falling or vice versa, (rate of change). For all flow metrics, we substituted stream stage, or the depth of water present as a suitable proxy for discharge (Booth and Konrad 2017). All metrics were calculated with the camera stream stage data and compared for consistency with sensor data where applicable. We considered a no-flow day to be any calendar day in which there was no flow observed (i.e., the riffle was dry, for the entire 24-hour period).
We compared the metrics among the nine sites to characterize variation across intermittent streams. We converted date of first no flow period to a Julian date and centered and scaled each of the flow metrics prior to analysis. To visualize differences among sites and the relative importance of each metric in describing variation across sites, we constructed a Principal Component Analysis (PCA) in the vegan package of R (Oksanen et al. 2022; R Core Team Version 4.2.2, 2022).
To explore the role of physical watershed characteristics on flow classifications (completely dry, isolated pools, or connected), we evaluated how four environmental covariates related with whether streams fully dried or dried to isolated pools. Stream segment slope and depth to restrictive layer were highly correlated (>0.65) with other covariates and were removed from all analyses. We evaluated the group means for this analysis, constructing boxplots to visualize differences among groups. We used a Mann-Whitney U Test to compare medians among groups.
We next explored the relationships between physical predictor variables (watershed area, mean slope, percent forested, and mean depth to water table) and flow metrics. Because this is an exploratory analysis and due to small sample sizes, we evaluated relationships among flow metrics and landscape variables by calculating correlation correlations for all combinations of metrics and covariates. When data was available, we used all ten sites to evaluate correlations, however only 9 sites had complete data for some variables.