Methodology:

A full description of this project can be found in Pritchard et al. (in review).

* Airborne survey*

We used a wide band mono-pulse dipole radar with a centre frequency of 7 MHz (Pritchard, King et al. 2020) for this survey. Between 27th October and 6th November 2019, we deployed our radar platform to survey the glaciers of Nepal's Khumbu Himal (Everest area) in the upper Dudh Koshi river basin. Our survey flights covered >200 line-km spanning altitudes of 3700 m to 6700 m (Pritchard et al., in review).

Based on the findings of Pritchard, King et al. (2020), we designed our survey patterns to include multiple glacier cross profiles because these are less prone to ambiguity between radar returns from the glacier bed and valley side walls. Our flightlines typically followed a continuous zig-zag path with crossings spaced at ~800 m over each glacier trunk for the up-glacier survey limb, with these crossings subsequently linked by a central glacier long-profile on descent. To minimise radar spreading losses, surveys were flown with the radar platform as close as safely possible to the glacier surface, typically a few metres to tens of metres given the considerable roughness of the glacier surfaces in this area. Reaching the highest section of the survey (Everest's Western Cwm), however, required a spiralling rather than direct ascent over the Khumbu Icefall and so achieved multiple glacier crossings at a wider range of ground clearances. To ensure dense radar sampling, we flew these glacier profiles as slowly as was practicable (typically ~10 m s-1 (36 km h-1)). Transit flights to and from the glaciers were at higher speeds of up to ~40 m s-1 (140 km h-1).

*Raw radar data *

We collected a raw radar data file for each survey flight. The filename format gives the date and time of file creation as yyyymmddhhmmss.hdf5, e.g., 20191103021438.hdf5.

File contents:

Each file contains two channels of radar data as recorded by the digitiser in the radar receiver system. Channel A is amplified, Channel B unamplified.

For each channel, there are multiple Traces, numbered sequentially (e.g., Trace00001). Each trace records a series of radar receiver signed voltages as a measure of returning signal strength.

Channel metadata:

The total duration of each trace (and therefore its range, in number of samples) is given by the NoOfSamples parameter in the channel metadata. Each sample has a duration of 2 nanoseconds, hence a trace with 10152 samples is 20304 ns long (a range of around 1700 m in ice, or 3000 m in air).

The 'SampleRate' parameter gives the timeout time (in seconds) for the receiver to wait to detect a returning pulse before assuming that triggering has failed and moving on to the next set.

Each trace represents the averaged voltages of the number of independent radar pulses given by the 'Stacks' parameter.

'Repeats' is an internal system parameter that sets the number of samples to store before writing to disk.

The GPS time is recorded along with the corresponding file start and end 'system' times for the onboard computer, to allow the system timestamps to be synchronised to the GPS data in post processing.

Trace metadata:

In each file, Trace00001 metadata shows the number format (Type) and the precise system date and time (yyyy,mm,dd,hh,mm,ss.ssssss). The 'GGA' string reports the GPS time at completion of this trace as hhmmss.s (e.g., 082520.8) and the GPS position at that time (ddmm.mmmmm, N/S, dddmm.mmmmm, E/W) (e.g., 2748.34269,N,08641.95129,E means 27deg 48.34269' N, 086deg 41.95129' E).

A new system time and GPS time and position is then stored for every 1000th subsequent trace, hence for trace 1001, 2001 etc. This is because GPS data could not be written to disk at the same rate as the trace creation. High frequency, high precision GPS times and positions were stored internally on the GPS, however, and ma...(64)

Data collection:

Instrumentation:

-The wide band mono-pulse dipole radar (centre frequency of 7 MHz) that we used in this survey was built by the British Antarctic Survey and is known as 'DELORES'.

-We processed the radar data in Reflex-Win Version 10.1.

-The clutter model was written in python and is described and archived here:Recinos, B., Goldberg, D. N., Boyle, D., & Pritchard, H. (2025). bearecinos/radar-declutter: First radar-declutter release (v1.0.0). Zenodo. https://doi.org/10.5281/zenodo.15488954

Data quality:

With the combination of our 3 kHz radar pulse repetition frequency, the average flying speed while surveying of ~10 m s-1 (36 km h-1) and the trace stacking and horizontal interpolation of our data processing, the horizontal sampling in our ice thickness survey averages 1.15 m (max 3.0 m, SD 0.27 m) along flightlines. However, the horizontal resolution of the bed target is limited by the Fresnel zone (effective radar footprint) of our transmitted pulses. At a frequency of 7 MHz and with the typical range of the glacier bed from the radar (~100-600 m, mean of ~160 m), the Fresnel zone has a radius of approximately 30-80 m (mean ~40 m).

In the vertical, the range resolution of the picked surface and bed is nominally a quarter wavelength (~7 m), but the 'optimal vertical resolution' (the resolvable range to a single discrete, prominent reflector (King 2020)) is ~1 m, and the 'practical vertical precision' for such horizons is ~2 m at this frequency (Pritchard, King et al. 2020). This implies that the practical vertical precision in thickness is the combination in quadrature of these two precisions, i.e., ~2.8 m.

The absolute accuracy of the thickness is subject to the accuracy of our assumed radar velocity in ice (0.168m/ns) with which we convert two-way radar travel times to ice thicknesses, and this is somewhat dependent on the unknown and potentially variable depth-averaged glacier water content. A range of velocities from 0.165 to 0.172 m/ns has, for example, been employed for temperate and cold ice above and below the equilibrium line of an alpine glacier (Macheret, Moskalevsky et al. 1993). This range of velocities implies a difference in ice thickness of approximately +/-2% for our survey (equivalent to a change in mean thickness from 139 m when using a velocity of 0.168 m ns-1 to between 137 m and 142 m for the reasonable range of velocities). Given the dependence on water content, this could be manifest as a thickness bias that varies broadly with altitude, with our results potentially too thick by up to 2% at lower (warmer) altitudes, too thin by up to 2% at higher (colder) altitudes. Potentially more significant bias (e.g., tens of metres) could result from mistaking a non-bed reflection horizon for the bed, which we sought to avoid with our clutter modelling.

To assess the consistency of our picked thickness measurements, we quantified the difference in thickness at 79 flightline crossovers. The mean absolute crossover difference was 9 m and the mean relative difference was 7% of thickness (median 6 m and 5%) (Table 1). We also compared our airborne survey results to previous ground-based radar surveys on Khumbu (Gades, Conway et al. 2000) and Ngozumpa glaciers (Pritchard, King et al. 2020). On Khumbu Glacier, seven radar cross profiles were surveyed in 1999 over the glacier tongue below the Khumbu Icefall, totalling 3.3 km in length (Gades, Conway et al. 2000). Of these, two lines crossed within ~200 m horizontally of two of our successfully surveyed cross profiles. While the earlier ground-based survey achieved profile lengths of ~500 m each, spanning most of the glacier width, we were only able to pick the bed over around 70 m of our profiles at each location. Although these surveys differ in date, method and exact location, the thicknesses reported by both surveys are similar: we measured maximum thicknesses of 445 m close to line P and 440 m close to line BC, compared to maxima of ~370 +/- 20 m (line P) and 440 +/- 20 m (line BC) in the previous study (Pritchard, King et al. 2020). A thickness of inferior or egual to 450 +/- 70 m close to these lines was also observed by a terrestrial gravity survey in 1976 (Moribayashi 1978).

____________________________________________

Crossover summary | Mean | SD | Median | Max |

____________________________________________

Relative difference (%) | 7% | 7% | 5% | 30% |

_...(30)