High Altitude Weather Station Data at USGS Benchmark Glaciers

Level 0 Data Processing: Raw data are compiled from multiple sources, primarily remote telemetry and raw downloaded logger files. Timeslips in logger are adjusted to match nearest 15 minute logger interval. Additional older raw records may be available via digitized scans of analog strip charts prior to installation of digital automated weather stations.

Level 1 Data Processing: Temperature Data All temperature sensors 1. Plausible value check - remove values below -40, above +30 ° C. This is informed by knowledge of the specific sites in our analysis. Precipitation Data 1. Remove maintenance-related noise from the record, resulting from precipitation gage liquid filling and emptying from the record. If the sum of incremental precipitation received in 45 minutes is over 6 cm, the incremental precipitation received during this period is set to null, as well as the timesteps immediately preceding and after to ensure entire maintenance impacts are removed from the record. Wind Data Wind Speed 1 .Plausible value check, setting wind speeds of greater than 75 m/s to NAN. 2. Check for a minimum variability in wind speed of at least 0.5 m/s every hour, indicating that instrument is not rimed, and is recording wind speed. Wind Direction 1. Plausible value check; remove values greater than 360° and less than 0°. 2. Minimum variability check; remove data where wind direction measured changes less than 10° in 3 hours. The World Meteorological Organization (WMO) recommendation is for 10 ° in 1 hour, but for data measured at higher frequency. Radiation Data Radiation data should be used with caution at this time. Little effort has been made to correct for sensor drift, issues with rime accumulation, and other confounding factors. Raw data is provided here with these caveats. 1. Plausible value check; remove values less than 0 and greater than 1600 Watts per meter squared (W/ m2) (Zahumensky, 2004). Values less than 0 are set to 0. Air Pressure Data 1. Plausible value check; remove values less than 50 kilo pascals (kPa), greater than 110 kPa.

Level 2 Data Processing: Temperature Data 1. Remove transient malfunctions with a median-based filter. A moving window of 8 values (2 hours for 15 min data) is used to identify sensor spikes. Where the absolute difference of a single observation from the 8-observation rolling median is greater than 4 ° C, the value is set to window median. Create site-best temperature 1. If multiple temperature sensors exist at the site, the records are compared and combined to create a timeseries of the best site temperature. If an aspirated temperature sensor exists at the site, this is considered the primary sensor due to better performance in calm, sunny conditions; passive temperature sensors are secondary. 2. In primary (aspirated) temperature sensor record, replace values with passive-sensor median where: • Absolute difference between aspirated sensor and mean of passive sensors greater than 2 °C and • Passive sensors agree with each other within 0.5 °C and • More than one passive sensor record is available. These values come from Zahumensky (2004). 3. Fill remaining gaps in primary temperature sensor record with median of all other sensors available, when the standard deviation among the remaining sensors is less than 1.5. Secondary sensors may be composed of a secondary aspirated, 2 passive sensors, or some other combination thereof. 4. Remove outliers due to intermittent sensor noise with a hampel filter. A hampel filter uses a moving window of 7 values, identifying outliers as values greater than 3 median absolute deviations from the window median. This is only applied in locations where the initial timestep of the logged data is sub-hourly. 5. Remove remaining transient outliers in final ‘best-temperature time series via an 8-sample median-based filter, as above. 6. Create time-aggregated Precipitation Data 1. Remove high-amplitude noise related to wind with a median-based hampel filter. • A 6-sample hampel filter identifies measurements greater than 2 median absolute deviations from period median, and fills with median. Window examines 3 measurements before and after each value, for a window period of 45 min on either side of examined value for 15-minute data. 2. Remove transient sensor malfunctions, resulting from electronic noise, using the daily median of incremental precipitation for the original timestep of data (15 minutes). If a single value of incremental precipitation is over 1 cm, that measurement is removed, and set to the median of incremental precipitation that day. 3. Remove low-amplitude noise, including wind and temperature-related diurnal fluctuation noise, with a smoother that retains total precipitation catch, while creating a monotonically increasing timeseries, as defined by Nayak (2010). Preserves timing and quantity of precipitation, while removing ‘negative’ precipitation measurements, a result of temperature and wind-related noise. Wind, Radiation, Air Pressure Data • Same level of QC process as L1; time-aggregated data is provided as a convenience. Data QC following recommendations of Zahumenský, I., 2004, Guidelines on quality control procedures for data from automatic weather stations: World Meteorological Organization, no. 955, p. 2–6.