Distributed summer air temperatures across mountain glaciers: climatic sensitivity and glacier size

Abstract. Near-surface air temperature (Ta) is highly important for modelling glacier ablation, though its spatio-temporal variability over melting glaciers still remains largely unknown. We present a new dataset of distributed Ta for three glaciers of different size in the south-east Tibetan Plateau during two monsoon-dominated summer seasons. We compare on-glacier Ta to ambient Ta extrapolated from several, local off-glacier stations. We parameterise the along-flowline climatic sensitivity of Ta on these glaciers to changes in off-glacier temperatures and present the results in the context of several available distributed on-glacier datasets around the world. Climatic sensitivity decreases rapidly up to 2000–3000 m along the down-glacier flowline distance. Beyond this distance, both the Ta of the Tibetan glaciers and global glacier datasets show a slower decrease of climatic sensitivity. In general, observations on small glaciers (with 



Introduction 42
Near-surface air temperature (Ta) is one of the dominant controls on glacier energy and mass 43 balance during the ablation season (Petersen et  Ta has favoured the use of simple, space-time invariant relationships of Ta with elevation, typically 47 that of the free-air environmental lapse rate (ELR). The physical processes of the free-air (that 48 which is independent of the surface boundary layer), however, are not appropriate to describe the 49 variability of Ta for local glacier boundary layers (Figure 1a), especially when the above-ice 50 temperature gradient (within ~10 m of the ice surface) heightens under warm 'ambient' (off-51 glacier) conditions (van den Broeke, 1997; Greuell and Böhm, 1998;Oerlemans, 2001;52 Oerlemans and Grisogono, 2002; Ayala et al., 2015). As a result, any extrapolation of Ta  53 observations from an off-glacier location, particularly those at lower elevations, are likely to lead 54 to an overestimation of snow and ice ablation in melt simulations (Petersen and Pellicciotti, 2011;55 Pellicciotti et al., 2014; Shaw et al., 2017). Whilst this problem has been long understood (Greuell 56 et al., 1997; Greuell and Böhm, 1998) To date, two main, simplified model approaches have been developed and tested to represent air 71 temperature over glaciers (Figure 1a). The first is the statistical model by Shea and Moore (2010) 72 developed to reconstruct Ta across glaciers of varying size in the Canadian Rockies from ambient 73 temperature records. This approach considered the ratio of observed on-glacier temperature and 74 estimated ambient temperature for the elevation of a given point (hereafter 'TaAmb') above and 75 below a critical threshold temperature for the onset of the glacier katabatic boundary layer (KBL). 76 The parameterisations that operate as a function of the along-flowline distance have since been 77 explored by Carturan et al. (2015) and Shaw et al. (2017) on smaller glaciers in different parts of 78 the Italian Alps. Carturan et al. (2015) found that the original published parameterisations were 79 sufficient to explain Ta on small, fragmenting glaciers up to distances of ~2000m. However, 80 investigation by Shaw et al. (2017) on a small alpine glacier found a pattern of along-flowline Ta 81 that was better described by an alternative, thermodynamic model approach. This second, 82 physically-oriented approach was developed by Ayala et al. (2015) based upon modifications of 83 the original model by Greuell and Böhm (1998) to account for a relative 'warming effect' evident 84 on the termini of some mountain glaciers compared to upper elevations that were fully dominated 85 by katabatic winds. The modified model (termed 'ModGB' in the literature) accounts for the 86 down-glacier cooling of Ta at increasing flowline distances due to sensible heat exchange and 87 adiabatic heating (Greuell and Böhm, 1998). It adds, however, an additional warming factor based 88 upon on-glacier observations in the lower sections of the glacier (e.g. at the greatest flowline Thus the ModGB method operates on the physical principles of the glacier boundary layer 95 (Greuell and Böhm, 1998) though it corrects for relative warming on the lower portion of glacier 96 (Ayala et al., 2015). To establish the magnitude of this warming, however, along-flowline data in 97 the lower portion of the glacier are essential. Because the available distribution of on-glacier 98 observations is often limited and rarely extends for the entire length of the glacier boundary layer, 99 this additional correction for warming and the number of physical unknowns of ModGB can lead 100 to high variability in Ta estimates on the glacier terminus (Troxler et al., 2020) ( Figure 1a). In 101 contrast to this, the statistical method of Shea and Moore (2010) provides a more simplified 102 estimation that has fewer assumptions and parameters, though it does not explicitly account for 103 the physical processes on the glacier, especially those that are thought to be the cause of relative 104 warming for the glacier terminus. It also provides a parameter that more specifically represents 105 the glacier 'climatic sensitivity' of the on-glacier Ta (defined here as the ratio of changes in 106 observed Ta on-glacier to changes in TaAmb). Despite its more conceptual nature, because of its 107 greater generalisability typical of a more simplistic statistical approach, we adopt the Shea and 108 Moore (2010) method to further investigate along-flowline Ta in this study.

110
To the author's knowledge, no study has investigated the variability of on-glacier Ta at different 111 sites around the world (with the exception of three glaciers considered by Ayala et al., (2015)). 112 As such the transferability or generalisability of models and/or model parameters remain mostly 113 unknown, and analysis of individual glacier sites, while beneficial to process understanding, may 114 not advance the science on how to treat the on-glacier Ta in models. In this study, we make a step 115 toward this by utilising new datasets of on-glacier temperature observations on three glaciers of 116 varying size in the south-east Tibetan Plateau. We analyse the main controls on along-flowline Ta  117 and its climatic sensitivity and present these new findings in the context of 11 other distributed 118 on-glacier observations around the world made to date. 119 Specifically we aim to i) understand the variability of Ta with the along-flowline distances at three 120 glaciers in the south-east Tibetan Plateau, ii) identify and quantify the climatic sensitivity of on-121 glacier Ta for different meteorological conditions and glacier sizes and iii) parameterise the along-122 flowline Ta using the Shea  We present the observations of Ta from a total of 20 air temperature logger locations (Table 1), 150 13 of which are situated on-glacier (4680 -5369 m a.s.l.) and seven off-glacier (4648 -5168 m 151 a.s.l.). These stations (hereafter referred to as 'T-loggers') observed Ta at a 2 m height using 152 HOBO U23-001 temperature-relative humidity sensors (accuracy +0.21°C) within double-153 louvered, naturally-ventilated radiation shields mounted on free-standing tripods. The T-loggers 154 recorded data in 10 minute intervals that are averaged to hourly data for analysis. We identify a 155 common observation period over the summers of 2018 and 2019 that range from 12 th July -18 th 156 September. For these date ranges, we observe only small data gaps for some T-loggers (Table 1). 157 We apply the nomenclature of TXG, whereby X refers to the T-logger number on each glacier and 158 G refers to the glacier number. 159 160 We additionally present Ta observations at two automatic weather stations (AWS) at elevations 161 ~4600 m a.s.l. (off-glacier, henceforth 'AWS_Off') and ~4650 m a.s.l. (on Parlung4, henceforth 162 'AWS_On') for the same time period (Figure 2). For distributing off-glacier air temperature, we 163 consider AWS_Off as our reference station. The AWS Ta observations are provided by Vaisala 164 HMP60 temperature-relative humidity sensors (accuracy +0.5°C) also housed in naturally-165 ventilated radiation shields. 166 167

3.2.
Uncertainty of air temperature observations 168 To provide an estimate of observation uncertainty, we compared the hourly divergence of two find that for these hours (when the KBL development is theoretically at its strongest (e.g. van den 175 Broeke, 1997; Oerlemans and Grisogono, 2002)), that 95% of hourly differences were < 1°C 176 ( Figure S1).

3.3.
Meteorological information 185 We obtained information regarding Ta, incoming shortwave radiation and relative humidity 186 (AWS_Off), on-glacier wind speed (AWS_On) and 'free-air' wind speed and direction (ERA5 -187 C3S, 2017). We used these data to explore the relationships of hourly on-and off-glacier 188 temperatures (section 4.2) for different prevailing conditions. 189 190

Elevation information 191
We used the 12.5 m Alos Palsar (ASF DAAC, 2020) digital elevation model (DEM) to provide 192 elevation information for the catchment (Figure 2b). We utilised this DEM in order to calculate 193 flowline distances (m) for each glacier from the TopoToolbox functions in Matlab (Schwanghart 194 and glaciers. In addition, the generated flowlines may also be dependent upon the quality and 199 resolution of the DEM available between the aforementioned studies. However, we do not 200 analyse lateral Ta variations in this study and consider that the impact of varying methods for 201 flowline generation to be negligible when assessing observations at a few select points on the 202 glacier. 203 204

Methods 205
For this study we use local, off-glacier Ta data from AWS_Off for aggregation of on-glacier sub-206 groups or for distribution of Ta in space. Sub-grouping allows one to interpret general causal 207 factors that dictate on-glacier behaviour, whereas the distribution in space allows a direct 208 comparison of on-and off-glacier temperatures and the effect of the glacier boundary layer. The 209 following subsections outline the sub-grouping (4.1) and distribution (4.2) methodologies. The  also report the equivalent on-glacier lapse rate that would be calculated for the above conditions. 222 223

Comparison of on-and off-glacier air temperature 224
We extrapolate AWS_Off Ta records to the elevation of each on-glacier T-logger (Table 1) to 225 quantify the Ta differences within the glacier boundary layer (Figure 1a). We derive an hourly 226 variable lapse rate between AWS_Off and off-glacier T-loggers deemed to be independent of the 227 glacier wind layer, thus excluding those T-loggers in the immediate pro-glacial zones. 228 Specifically, we use AWS_Off and T-loggers T194, T294 and T1390 to construct a 'catchment lapse 229 rate' where the origin of the calculated regression must pass through the elevation of AWS_Off 230 that acts as the forcing station in this study (see supplementary information, Figure S2). We 231 consider this as the best available approach to estimate the ambient lapse rate for the catchment. 232 We compare the hourly estimates of extrapolated off-glacier Ta  The Shea and Moore (2010) approach (hereafter 'SM10') estimates on-glacier Ta using TaAmb at 239 a given elevation by: where T* (°C) represents the threshold ambient temperature for the onset of katabatic flow and 245 T1 is the corresponding threshold Ta on the glacier. Parameters k1 and k2 are the climatic 246 sensitivities of on-glacier Ta to TaAmb below and above the threshold T* (Figure 1b

Global datasets of on-glacier temperatures 275
To explore the generalisability of the SM10 approach and provide context to the findings of the 276 Parlung catchment, we explore the calculated k1 and k2 parameters for several of the available 277 distributed on-glacier datasets published to date ( Figure S3, Table 2). We subset summer periods 278 to when all available on-glacier observations are available at a given site. authors. For all other sites, we derive TaAmb from the most locally available off-glacier AWS and 282 the published lapse rate from the relevant studies (Table 2). In the absence of lapse rate 283 information, we apply the ELR (-6.5°C km -1 ) to extrapolate Ta to the elevation of the on-glacier 284 observations. 285 For each glacier site, data are limited to those hours when all stations for that glacier are available 286 and the k1 and k2 parameters (equation 1) are only calculated when; i), >10% of the total hourly 287 data at a given station is above or below T* (to have enough data to calculate k2 and k1, 288 respectively) and, ii) the linear regression to derive each parameter is significant to the 0.95 level. 289 For those on-glacier stations that do not satisfy the above requirements, we do not calculate the 290 k1 and k2 parameters.

292
Finally, we group the derived k2 sensitivities of the SM10 approach against the climatology that 293 describes the given glacier(s) location. For this, we consider the mean summer (JJAS or DJFM in 294 the southern hemisphere) air temperature (MSAT) and the total annual precipitation for the year(s) 295 of study at each location (Table 2). MSAT is derived from the ERA5 product for the glacier 296 centroid location and corrected to the mean glacier elevation by the ELR. However, total 297 precipitation from ERA5 has been shown to have considerable bias when tested against in-situ 298 observations (e.g. Betts et al., 2019), and so we provide the best available value from the relevant 299 literature (Table 2). We note that a full analysis of the local climate is beyond the scope of this 300 work, though we attempted a generalised analysis in order to link any clear differences in the 301 global datasets to climatological influences. 302 303 suggest a potential non-linear behaviour of lapse rates between AWS_Off and the top of the 330 flowline for Parlung390, though we lack the off-glacier observations above the flowline origin to 331 test this (Figure 4b). We therefore utilise a piecewise lapse rate at the point of the highest off-332 glacier lapse rate station (T1390 -red line in Figure 4) to account for the discrepancy between the 333 estimated and observed Ta at T6390, which is assumed to be near to the flowline origin where 334 climatic sensitivity is theoretically equal to 1 (i.e. that on-glacier observations = Taamb). 335 336 Figure 5 presents the hourly differences between TaAmb and observed Ta at each site. The 337

Results
deviation of estimated and observed Ta theoretically begins at a critical temperature threshold, T* 338 (Shea and Moore, 2010) and this effect can be observed at T-logger sites on Parlung94 and 339 Parlung4, particularly those at greater flowline distances. Coloured by the hourly wind speeds 340 recorded at AWS_On, the beginning of the temperature deviations (T*) aligns well with the onset 341 of katabatic winds on Parlung4 (and only assumed for the other glaciers due to lack of on-glacier 342 wind observations - Figure 5). Despite being pro-glacial stations, T14 and T24 reveal a similar, 343 albeit weaker effect of the glacier boundary layer, possibly due to larger glacier flowline and 344 sustained effect of the katabatic wind into the pro-glacial area.

346
The mean bias offset of along-flowline Ta using the catchment lapse rate is shown in Figure 6. 347 For the coolest 10% of hours at AWS_Off (P10), there is generally minimal offset between TaAmb  348 and observed Ta for the entire dataset. This clearly does not hold true for P90 conditions ( Figure  349 6a), as already established (Figure 4), and offsets of Ta (TaAmbobserved Ta) are up to 5.8°C at 350 flowline distances of > 7000 m on Parlung4. These effects appear to heighten beyond 2000 m 351 along the flowline (Parlung94), though slight offsets can be witnessed for all glaciers. This is 352 generally associated with drier conditions, and for hours of greater relative humidity (AWS_Off), 353 offsets are small (Figure 6b). Considering 'free-air' wind variability provided by ERA5 354 reanalysis, Ta offsets are largest for the dominant south-westerly wind direction (85% of hours) 355 and when free-air wind speeds are smallest (Figure 6c and 6d). However, un-corrected, gridded 356 wind speeds do not appropriately represent the local 'free-air' boundary conditions and thus the 357 interaction of off-glacier wind speeds and the glacier boundary layer development remain unclear 358 for these glaciers. For all but the coolest ambient temperatures (Figure 6a), observations at the 359 greatest flowline distances deviate the most from the estimated values.

361
This offset is highly variable in time, however, and related to the prevailing conditions of a given 362 year (Figure 7). Considering the maximum daily Ta offsets at the on-glacier T-Logger closest to 363 the terminus on each glacier (Table 1), we find that Parlung94 and Parlung4 T-loggers have 364 similar magnitudes of Ta offsets during the mid-summer months, particularly for 2018 (Figure 7). 365 These maximum offsets are in clear relation to the incoming shortwave radiation record at 366 AWS_Off ( daily Ta offsets on Parlung390 steadily increase during July and August then fall close to zero in 370 September. The bias offsets for Parlung4 and Parlung94, however, remain sizeable (Figure 7). 371 Because our study period focuses on the core monsoon period ( to the observations at Parlung glaciers and the original study are notably distinct (Figure 8, Table  382 3). This behaviour is further highlighted when observing other published or revised datasets for 383 the context of this work (Figure 8b) Glacier, Canada (Shea and Moore, 2010). 396 397 Figure 9 shows the k2 parameters plotted against flowline distance, coloured by rankings of 398 MSAT and precipitation totals ( Table 2). The warmest of the investigation sites (during the 399 measurement years) appear to lie closer along the original SM10 parameterisation until ~4000 m, 400 whereas deviation of the k2 parameters from this line appears larger for observations at relatively No clear patterns are visible with relation to mean annual precipitation totals, although the 405 observations at Juncal Norte are noted as the driest of the study sites (Figure 9b).

406
A clear difference between observations of CMBC and Parlung at large flowline distances is the 407 distance from the glacier terminus, which suggests a possible difference in processes being 408 compared between sites. Accordingly, we plot the k2 parameters as a function of the normalised 409 flowline, adjusted by the total length of glacier under the year(s) of observation (Table 2) al., 2015) and become more sensitive to ambient air temperatures due to a lack of katabatic 447 boundary layer development (Figures 6 and 7). the sensitivity of the glacier to external temperature changes (shown by Ta bias offsets) has a 453 sizeable temporal variability that can be controlled by the monsoon weather conditions (such as 454 ambient air temperature, humidity and incoming radiation) and can sometimes be independent of 455 the glacier size (Figure 7). Whilst we cannot determine the impact of monsoon timing and 456 intensity upon the climatic sensitivity of these glaciers with the current dataset, we are able to 457 determine that the observed relationship to flowline distance is consistent to that of other regions 458 of the world (Figure 8). Future work on Tibetan glaciers should attempt to extend monitoring to 459 the pre-monsoon period to identify if a seasonal onset for the changing glacier climatic sensitivity 460 can be defined, and how the monsoon may affect it. Particular focus should be given to understand 461 the local meteorological conditions for each glacier, as this may explain some of the variability 462 in Ta offset values, and why they may sometimes be independent of the along-flowline distance 463 (Figure 7). 464 465 6.2. Parameterising glacier climatic sensitivity 466 In this study, we discuss the climatic sensitivity of on-glacier Ta based upon observations above 467 a threshold ambient temperature for the onset of katabatic conditions (T*). This sensitivity to 468 ambient temperature during relatively warm conditions, indicated by the k2 parameter of Shea 469 and Moore (2010) (Figure 1), demonstrates a generally consistent behaviour between the T-logger with a slightly greater sensitivity to the ambient temperature (i.e. larger k2 values - Figure 8b). 473 Whilst the newly presented dataset for the Parlung catchment provides an important confirmation 474 of the climatic sensitivity for some Tibetan glaciers, further studies of individual glaciers can 475 provide only local parameterisations for climatic sensitivity that may not be applicable to other 476 sites. Accordingly, we have made here a first attempt at combining many of the published datasets 477 regarding distributed Ta on mountain glaciers around the world (Table 2) to examine the potential 478 for generalisability of a model accounting for climatic sensitivity (Figure 8).

480
We found a sizeable spread in the climatic sensitivities of Ta for the on-glacier datasets considered, 481 though a consistently rapid decrease of sensitivity along glacier flowlines is found for most sites 482 up until ~2000-3000 m of distance (Figure 8b). While localised meteorological and topographic 483 factors likely interact to explain the spread of sensitivities at small flowline distances (Figure 8b), 484 the results suggest that small glaciers with flow lengths < 1000 m would reflect a 0.7-0.8 485 sensitivity to changes in TaAmb. Beyond this distance, the climatic sensitivities notably follow 486 one of two patterns; a continued, albeit less rapid decrease in sensitivity (more generally following 487 the model proposed by Shea and Moore (2010)), or a tendency toward increasing sensitivity at 488 the largest flowline distances (more related to those findings of the 'ModGB' model - Figure 1a). 489 With reference to the relative Ta differences among only on-glacier observations, these have been 490 termed as down-glacier 'cooling' or 'warming', respectively for many past studies (Ayala et al., to account for, though we did not explicitly test this within our study due to a requirement for 515 more data and a greater number of parameters and assumptions (Shaw et al., 2017). The strength 516 of this so called along-glacier 'warming effect' could therefore be governed by local topography 517 (adjusting the boundary layer convergence or divergence) or the total glacier flowline distance 518 and the large fetch of a cool air parcel overcoming the competing effect of warm, up-valley winds 519 (Figure 1d -as seen at T24 in Figure 5). 520 521 By subjectively grouping glaciers by the presence of the observed increase in climatic sensitivity 522 and normalising the flowline distance of the observations by the total flowline for each glacier, 523 we identify that the relative increases in climatic sensitivity begin at ~ 70% of the total flowline 524 distance (Figure 10). A smaller climatic sensitivity can be observed for larger glaciers (Figure  525 10a), which is consistent with the development of the KBL over a large fetch (Greuell and Böhm,526 1998; Shea and Moore, 2010), though the length itself indicates nothing clear about why greater 527 climatic sensitivity exists for some glacier termini (Figure 10b). 528 529 The clear outlier of these datasets is Juncal Norte Glacier in Chile (Figure 8b). It is interesting to observations would be required to test this. 540 541 6.3.

Future directions for researching air temperatures on glaciers 542
A limitation of our work is the dependency of the derived 'global' climatic sensitivities ( Figure  543 8b) to the available off-glacier data and the published lapse rates to extrapolate them to the 544 relevant elevations on-glacier. In our case, we are able to identify a potentially non-linear lapse 545 rate of TaAmb for the highest elevations over Parlung94 and Parlung390 (Figure 4). Although we 546 cannot confirm this without off-glacier observations above the top of the flowline (Carturan et al., 547 2015), we are able to well constrain ambient air temperature distribution using hourly 548 observations at several off-glacier locations to derive the best possible 'catchment lapse rate'. For 549 other datasets (Table 2), we rely upon the available off-glacier data and lapse rates that are not  hard to clearly define due the variable instrumentation (sensors and radiation shielding), on-557 glacier location and local topographic and micro-meteorological effects of each study site (Table  558 2). Further work on a unified model of estimating Ta should need to address these issues, perhaps 559 with further, dedicated analyses. 560 561 In our study, we apply the parameterisation of Carturan et al. (2015) to derive along-flowline 562 values of the theoretical onset of the KBL (T*). While these values appear appropriate for our case 563 studies (based upon manual inspection), they were derived for a more limited number of total 564 observations. We experimented with a static T* value of 5°C in order to test the sensitivity of our 565 analysis to the assumptions of T*, though found a minimal change in our derived k2 sensitivities 566 (not shown). 567 568 Finally, in this study we assess climatic sensitivity based upon ambient air temperatures above 569 this T* threshold. We identify, however, that this is partly different to the climatic sensitivity 570 presented by earlier works (Greuell et al., 1997;Greuell and Böhm, 1998;Oerlemans, 2001;571 2010), which considered all hours of the on-glacier observations when comparing to extrapolated 572 off-glacier Ta. In some instances, over estimation of on-glacier Ta also for cooler conditions may 573 produce a consistent 'all-hour' climatic sensitivity value (i.e. where k1 and k2 sensitivities are 574 similar - Figure 1b). However, ignoring separate effects (k1 and k2) due the rise of the KBL 575 (Figure 1c, Figure 5) arguably over-simplifies the glacier's climatic sensitivity and therefore does 576 not aptly describe the two behaviours separated by an onset event (Shea and Moore, 2010; Jiskoot 577 and Mueller, 2012). Accordingly, we caution somewhat the direct comparison of the climatic 578 sensitivity presented here and that of previous works, though consider the use of k2 to be an 579 appropriate indicator of climatic sensitivity for this work going forward. As previously 580 mentioned, we have considered the approach of Shea and Moore (2010) to be a more 581 generalizable method for calculating glacier climatic sensitivity and thus estimating on-glacier Ta. 582 However, the competing effects of glacier katabatic and up-valley winds need to be incorporated 583 to address the challenges that less simplistic methods (i.e. ModGB) were designed for. 584 585 Based upon the findings of this work, we recommend that future research i) attempt to standardise, 586 where possible, the measurement and comparison of off-and on-glacier air temperature, 587 potentially exploring more the use of artificially-ventilated radiation shields that are less prone to 588 heating errors (Georges and Kaser, 2002), ii) instrument glaciers of varying size in the same 589 catchment to explore the relative importance of glacier size and local meteorological conditions 590 (Figure 7), and iii) model the detailed interactions of air flows on the glacier termini using, for We presented a new dataset of distributed on-glacier air temperatures for three glaciers of 597 different size in the south-east Tibetan Plateau during two summers (July -September). We 598 analysed the along-flowline air temperature distribution for all three glaciers and compared them 599 to the estimated ambient temperatures derived from several, local off-glacier stations. Using this 600 information, we parameterised the along-flowline climatic sensitivities of these glaciers using the 601 method proposed by Shea  that is closer to 1). We therefore confirm earlier evidence regarding the high sensitivity 607 of small glaciers (flowline distances < 1000 m) to external climate, and thus future 608 warming. 609 2. The largest offsets between observed on-glacier and estimated off-glacier air 610 temperatures are found for the warmest off-glacier hours, during drier, clear sky 611 conditions of the summer monsoon period. 612 3. Above the established onset of the katabatic boundary layer, climatic sensitivity to 613 ambient temperature decreases rapidly up to ~2000-3000 m along the glacier flowline. 614 Beyond this distance, both the Tibetan glaciers and other datasets of the literature show 615 a slower decrease of climatic sensitivity. 5. The terminus of some glaciers remain associated with other warm air processes, possibly 623 due to boundary layer divergence, warm up-valley winds or debris cover heating. We find 624 that these effects are evident only beyond ~70% of the total glacier flowline distance, 625 although further work is required to explain this behaviour. A better understanding of 626 temperature variability for this lower 30% is highly important as most of the summer 627 melting will occur for this sector of the glacier.

629
In summarising the findings from all available distributed on-glacier datasets to date, we identify 630 some key directions for future work on this subject. This includes comparing local influences of 631 glacier size and micro-meteorology and standardising measurement practices, where possible, to 632 aid the conclusions for a generalised model of on-glacier air temperature estimation. 633 634