The Hintereisferner (HEF) is a valley glacier located in the Ötztal Alps, Austria. HEF is recognized as one of the “reference glaciers” by the World Glacier Monitoring Service. Since 1952/53, ongoing observations form part of a comprehensive catchment monitoring program (Strasser et al., 2018). HEF has played a crucial role as a testbed for developing instruments, methods, and models (Kuhn et al., 1999) and exploring glacier and valley winds (Obleitner, 1994).

Building on this legacy, the Hintereisferner Experiment (HEFEX I) micrometeorological measurement campaign unfolded over 3 weeks in August 2018 (Mott et al., 2020). Subsequently, the more extensive HEFEX II campaign took place in August 2023, aiming to delve into the glacier’s microclimate through a combination of an extensive network of meteorological stations and various remote sensing measurements (Nicholson et al., 2025).

Hintereisferner, a classical valley glacier, measured approximately 6.3 km in length in 2018, with an elevation difference of around 1,200 m (https://wgms.ch/, last access: 11 December 2020). The glacier is nestled in a northeast-oriented valley with steep slopes. Positioned in the “inner dry Alpine zone” (Frei and Schär, 1998), one of the driest regions in the European Alps, Hintereisferner has undergone significant shrinkage in recent decades, with a 15% decrease in glacier area recorded between 2001 and 2011 (Klug et al., 2018; Abermann, 2009).

In this study, we used an InfraTec VarioCAM HD thermal infrared camera with a spatial resolution of 1,024 × 768 pixels to monitor air temperature dynamics near the glacier surface. The camera was aimed at two vertically mounted screens made from a thin, stretched polyester mat coated with acrylate (Haugeneder et al., 2023). Due to the material’s ability to rapidly equilibrate with ambient air temperature, we assume that its surface temperature provides a reliable proxy for air temperature. Following Grudzielanek and Cermak (2015), the emissivity of the screen material is taken as 0.94.

The screens were precisely aligned with the direction of the glacier wind (Figure 1c). These screens were strategically configured to facilitate the measurement of flow patterns and the corresponding temperature dynamics along the primary wind direction. The camera was positioned so that the screen is parallel to the glacier surface. In addition, they allowed for the assessment of the vertical profiles of air temperature below and above the height of the katabatic jet. Data on turbulence gathered during a previous measurement campaign in 2018 (HEFEX I) indicates that a jet typically forms between 1.7 and 2.3 m above the ice surface under katabatic-flow conditions. Screen 1, with dimensions of 1.9 m in height and 3 m in length, was utilized to measure along-flow patterns. Screen 2, measuring 3.8 m in height and 1.9 m in width, captured the vertical temperature profile. As this study focuses on temperature dynamics along the vertical profile in conjunction with station data, the analysis was restricted to Screen 2. Given the distance of the IR camera to the screen, the horizontal and vertical resolution is 6.1 mm/pixel and 5.8 mm/pixel respectively. The TIR camera captured infrared frames at a rate of 30 Hz. To account for slight variations in material thickness and the sensitivity of screen emissivity to minor surface contamination, we adopted the method of Grudzielanek and Cermak (2015), averaging TIR values across 10 horizontal pixels along the vertical profile to enhance the reliability of the temperature measurements. Furthermore, the measurements may be influenced by solar heating of the material, depending on the sun’s angle. As a result, absolute temperatures obtained from daytime TIR measurements can exhibit a positive bias. To address this, it is recommended to focus on relative temperature changes over space or time rather than relying on absolute values (Haugeneder et al., 2023).

While the full HEFEX II campaign was running over 3 weeks in August 2023, thermal infrared measurements were only conducted during the Intensive Observation Period (IOP), lasting 3 days (August 22–24) (Nicholson et al., 2025). Due to the substantial volume of data storage required, the measurements were not carried out continuously but were confined to sequences lasting from 1 hour to several hours during the day and night. For this study, we’ve chosen five compelling sequences during August 22 and 23 to demonstrate the efficacy of our method in characterizing the spatial and temporal temperature dynamics of glacier winds. These sequences illustrate its ability to capture temperature fluctuations during purely katabatic diurnal and nocturnal flows and disturbed situations throughout the day and night at an ultra-high resolution. Combining the TIR data with turbulence data from the near-by tower further allows an analysis of the approximate jet height location.

Meteorological conditions and turbulence were measured using a 5-m weather tower (T275) with various sensors installed at different vertical levels (Figures 1a,b) (Nicholson et al., 2025). Low-frequency 2D wind speed and direction (Young JR 04101) was measured at 5.5 m. Unventilated Low-frequency temperature and humidity sensors (HC2A-S3 Rotronic) were placed at the 2.2 and 4.2 m levels. Low-frequency measurements were sampled every minute and averaged to 10 min intervals. A pressure sensor (CS100 Campbell Scientific) was located in the logger box. Furthermore, the station was equipped with one IRGASON and two 3D sonic anemometers (CSAT3) at 1.1, 2.0, and 4.0 m, respectively. The station was positioned close to the IR measurement setup at the glacier centerline (Figure 1).

Turbulence statistics from this tower at 1.1, 2 and 4 m above the surface were processed using 1 min averages (as suggested by the multi-resolution flux decomposition, Vickers and Mahrt, 2003) with prior linear detrending. Double rotation was used to rotate the data into the streamline coordinates, so that the u refers to the longitudinal, v lateral and w surface-normal velocity component (cf. Mott et al., 2020).

We selected five periods during the IOP characterized by two different wind systems (cf. Mott et al., 2020): (i) purely katabatic conditions with a persistent flow direction from the southwest (defined as 200° at meteorological station T275) and wind velocities greater than 3 m/s (periods P1, P5 in Figure 2); (ii) disturbed flow conditions with wind directions deviating by more than 60° but less than 120° from the dominant katabatic-flow direction, indicating across-glacier winds (P3, P4) (Figure 2). While P3 exhibits a southwesterly across-glacier wind with strong disturbances, P4 is characterized by a northeasterly across-glacier wind, interspersed with short phases of glacier winds. During P2, a slight southerly flow with lower wind speeds occasionally disturbed the down-glacier wind. These conditions are typical at Hintereisferner, as discussed by Mott et al. (2020). They could show that these flows across the glacier measured at the glacier were associated with strong heat advection with the larger-scale flow. A significant increase in near-surface air temperature indicated horizontal heat advection.

During down-glacier wind conditions (P1, P2, P5), measurements revealed that wind speeds were highest at the 1.1 m and 2 m levels and consistently lower at the 4 m level, indicating that the jet maximum was located below 4 m. Additionally, temperatures were lower at the lower sensors, with more pronounced temperature gradients near the surface.

Despite generally lower near-surface wind speeds during across-glacier flows (P3, P4), downward turbulent heat fluxes were similar to or even higher than those observed during down-glacier winds caused by increased air temperatures during these periods (as already demonstrated in Mott et al., 2020). In the early morning katabatic flow conditions (P1), the highest downward (most negative) sensible heat fluxes occurred at the lowest measurement level. However, by late morning (P2), the two upper measurement levels exhibited more downward (negative) heat fluxes.

No clear relationship was observed at night between the magnitude of downward sensible heat fluxes and measurement height. Under disturbed flow conditions, air temperature and wind speed became much more uniform across the three measurement levels, suggesting less stable atmospheric stratification.

Daytime TIR measurements can be affected by solar heating of the screen material, depending on the sun’s angle. Similarly, conventional temperature sensors are also prone to solar radiation effects. To assess the consistency between different temperature measurement methods, we compared air temperature readings from the station with those derived from the TIR camera. Specifically, we used low-frequency (10-min interval) data from unventilated ROTRONIC sensors at heights of 2.2 m and 4.2 m, and compared them to 1-min averaged temperatures obtained from the TIR camera at 2.2 m and at the screen’s highest measurement point of 3.4 m (Figure 3). Since thermal stratification is essential for turbulent heat fluxes, temperature gradients between the first and second measurement level (L1) and the second and third measurement level (L2) are compared between the screen measurements, the fast response sensors (L1 and L2, Irgasons and CSAT) and the low response sensors (only available for L2) (Figure 4). The comparison between high-frequency and low-frequency (instantaneous) measurements, indicated by single points, aims to assess whether the different methods capture a similar range of temperatures. Although the techniques are not directly comparable—the upper measurement heights differ by 0.8 m between screen and tower, and the screen and tower are located approximately 15 m apart horizontally—this comparison provides a general sense of agreement between the datasets. Screen temperatures are corrected for the estimated emissivity of the screen material following the methodology suggested by Grudzielanek and Cermak (2015).

A comparison of the two upper-level temperature measurements reveals distinct differences between daytime and nighttime conditions (Figure 3). During the day, low-frequency sensors record slightly lower temperatures—particularly in period P2—while measurements align well in periods P1 and P3. At night, when shortwave radiation errors are minimal, low-frequency sensors register higher temperatures during period P4, coinciding with a disturbance in the glacier wind. In contrast, temperature measurements at both levels show good agreement during the katabatic glacier wind event in period P5. Overall, the bias between screen temperatures and low-response sensor temperatures is 0.67 °C at measurement level S2 (2.2 m) and 0.48 °C at S3 during katabatic conditions (Table 1), and −0.10 °C at S2 and -0.57 °C at S3 (4.2 m) during disturbed wind conditions. When separated into daytime and nighttime conditions, the temperature biases exhibit a clear contrast: nighttime biases are −0.7 °C and −1.2 °C, whereas daytime biases rise to 1.5 °C and 1.4 °C. The positive daytime biases suggest that the screen temperatures were influenced by shortwave solar radiation, with the magnitude of this effect depending on the solar angle of incidence. This influence was particularly evident during period P2, which showed a stronger positive temperature offset, likely due to enhanced direct solar exposure of the screen. Additionally, lower wind speeds during P2 compared to P1 may have amplified the solar heating effect by reducing convective cooling of the screen material. Conversely, the negative nighttime biases likely reflect radiative longwave cooling of the screen surface. This comparison further highlights the importance of collocated high-frequency air temperature measurements—such as those obtained from fine-wire thermocouples—to accurately validate absolute screen temperatures.

Temperature gradient comparisons (Figure 4) indicate that TIR measurements at the screen closely match those from the fast-response sensors—especially at the upper level—where gradient biases range between 0.05 °C and 0.3 °C per meter. During period P4, fast-response sensor readings at L1 indicate unstable atmospheric conditions, whereas all other instruments, including the screen, suggest slightly stable conditions, reflected in mildly positive temperature gradients. The biases in L1 are higher than in L2 (0.44°–0.77 °C/m), which may be partially attributed to the significantly greater temporal variability of the temperature gradients near the glacier surface.

Furthermore, as illustrated in Figure 5, the temperature variances obtained from the turbulence sensors and the screen measurements indicate that the screen measurements exhibit smaller variances compared to the eddy-covariance sensors during certain periods. This discrepancy is most pronounced during katabatic flow conditions in the forenoon at a height of 1.1 m above ground. In contrast, under disturbed conditions and during nighttime katabatic flows, the temperature variances align closely between screen and station measurements, demonstrating consistency in these specific periods. Furthermore, variances are well captured at higher levels, between 2 m and 4 m above ground. This may be partly attributed to the more dynamic temperature environment at 1.1 m, where temperature fluctuations are more intense, whereas higher levels experience less pronounced high-frequency temperature fluctuations.

This comparison underscores that the screen TIR method tends to underestimate particularly high temperature variances, although it reliably captures relative differences between variances across different heights. The variance underestimate is more pronounced during specific situations, while it aligns well during other periods. These findings may reflect slight differences in flow conditions between the screen and station locations, particularly as the jet height varies significantly over time, potentially influencing temperature fluctuations at specific heights. The temporal dynamics of temperature fluctuations over the entire profile is discussed in Section 3.3.

The slight underestimation of high-frequency temperature fluctuations suggests that the screen material’s thermal response may be too slow to accurately resolve ultra-high frequencies, such as those at 30 Hz. To evaluate the frequency response of the TIR-based screen temperature measurements, we computed temperature spectra (seen in Figure 6) for three heights above the surface (1.1 m, 2 m, and 3.4/4 m) during three distinct observational periods: a katabatic day (P1), a mostly katabatic day (P2), and a katabatic night (P5). Spectra derived from screen temperature data (blue curves) were compared to those from co-located sonic temperature measurements (red curves), with a reference −2/3 slope line included to represent the expected behavior in the inertial subrange. At all heights and across all periods, the screen-derived spectra closely followed the sonic temperature spectra very well up to approximately 0.1 Hz, indicating that low-frequency temperature fluctuations are well captured by the TIR method. However, at higher frequencies ( > 0.1 Hz), the screen spectra diverged notably, exhibiting a steeper slope than the canonical −2/3, consistent with the behavior of a low-pass filter. This deviation suggests that the thermal inertia or slow response time of the screen material limits its ability to resolve rapid temperature fluctuations. As a result, high-frequency variability—particularly in the inertial subrange—is attenuated, leading to an underestimation of small-scale turbulent processes. These findings highlight the utility of the TIR method for capturing mesoscale temperature variability while also underscoring its limitations in representing high-frequency fluctuations critical for characterizing atmospheric turbulence.