High-resolution debris-cover mapping using UAV-derived thermal imagery:limits and opportunities

Debris-covered glaciers are widespread in high mountainranges on earth. However, the dynamic evolution of debris-covered glaciersurfaces is not well understood, in part due to difficulties in mappingdebris-cover thickness in high spatiotemporal resolution. In this study, wepresent land surface temperatures (LSTs) of supraglacial debris cover and theirdiurnal variability measured from an unpiloted aerial vehicle (UAV) at ahigh (15 cm) spatial resolution. We test two common approaches to derivedebris-thickness maps by (1) solving a surface energy balance model (SEBM)in conjunction with meteorological reanalysis data and (2) least squaresregression of a rational curve using debris-thickness field measurements. Inaddition, we take advantage of the measured diurnal temperature cycle andestimate the rate of change of heat storage within the debris cover. Bothapproaches resulted in debris-thickness estimates with an RMSE of 6 to 8 cmbetween observed and modeled debris thicknesses, depending on the time ofthe day. Although the rational curve approach requires in situ fieldmeasurements, the approach is less sensitive to uncertainties in LSTmeasurements compared to the SEBM approach. However, the requirement ofdebris-thickness measurements can be an inhibiting factor that supports theSEB approach. Because LST varies throughout the day, the success of arational function to express the relationship between LST and debristhickness also varies predictably with the time of day.During the period when the debris cover is warming, LST is heavily influenced by the aspect of the terrain. As a result, clear-sky morning flights that do not consider the aspect effects can be problematic.Our sensitivity analysis of variousparameters in the SEBM highlights the relevance of the effective thermalconductivity when LST is high. The residual and variable bias of UAV-derivedLSTs during a flight requires calibration, which we achieve with bare-icesurfaces. The model performance would benefit from more accurate LSTmeasurements, which are challenging to achieve with uncooled sensors in high mountain landscapes.