3.1 Assessment of housing unit performance
We retrieved data from 30 (n = 14 pond loggers, n = 16 control loggers) of the 32 loggers deployed, with two pond loggers underwater at the time of collection. We downloaded temperature data for the entire study period (between 1 July 2018 and 21 June 2019) for 26 loggers. Four pond loggers at sites T1, T2, T8, and T13 failed due to a potentially faulty logger backing design that was addressed by the manufacturer during the time between initial deployment and site visits in June 2019; all pond loggers were replaced by the manufacturer, and replacements were deployed following data retrieval in June 2019. No subsequent issues emerged (0% failures) using loggers with the updated backing for other experiments during which loggers were submerged in water for months at a time (M.C. Mims, unpublished data).
Overall, we found that the logger housing design successfully protected the loggers from physical damage, even when disturbed by cattle, but there were some considerations and limitations. Careful placement of loggers in the deepest point in the pond is imperative for accurate hydroperiod estimation. At site T11, we observed that the pond logger did not appear to be placed at the lowest point within the pond, as was intended. We observed very shallow water pooled in another location near the logger in summer 2019 that dried a few days later. Therefore, the data collected from this logger may not accurately reflect the pond inundation state. Additionally, rock piles placed on top of the rugged housing likely affected absolute temperature readings. Though rock color or density may have had differential effects among loggers, we suspect the variation among loggers was likely low overall. Furthermore, because this method considers tSD rather than absolute temperature, we do not anticipate these differences had substantial effects on results.
Another potential issue with our physical design was the accumulation of sediment or other debris within the rugged housing unit that interfered with temperature readings from the pond loggers at sites A14, T4, T9, and T17 (see Figure S3 for example). Although we have relatively high confidence in inundation timing, drying dates were less precise largely due to the accumulation of sediment in the housing unit. At site A14, mud was discovered in the logger housing on 31 March 2019 and was cleared. In the days preceding 31 March 2019, the temperature standard deviations from the pond logger were considerably lower than those from the control logger at this site, despite a lack of water in the pond, resulting in the paired pond-control model falsely predicting that the pond was in a wet state. After the sediment was cleared, the difference in these tSDs decreased to nearly zero, and the paired pond-control model correctly designated the pond state as dry. Mud and debris found inside the rugged housing of the pond loggers at sites T4 and T9 in late June likely caused the tSDs of these pond loggers to remain low relative to those of the control loggers even after drying, leading to false wet predictions in the paired pond-control models. In addition, we occasionally observed animals inside housing units, including several salamanders inside the housing for the pond logger at site T9 (Figure 2).
To improve drying date precision, the housing unit design would likely need to exclude sediment, which is difficult to do without making other compromises. Solutions for avoiding the issue of sediment in rugged housing units, and the subsequent decoupling of pond and control data, include packing the housing unit with insulation or other material that would not allow sediment to enter. However, this can lead to issues such as a buoyant housing unit and may affect the temperature readings if the material is a good insulator.