A multiannual ground temperature dataset covering sixteen high elevation sites (3493–4377 m a.s.l.) in the Bale Mountains, Ethiopia

Tropical mountains and highlands in Africa are under pressure because of anthropogenic climate and land-use change. To determine the impacts of global climate change on the afro-alpine environment and to assess the potential socio-economic consequences, the monitoring of essential climate and environmental variables at high elevation is fundamental. However, long-term climate observations on the continent above 3,000 m are very rare. Here we present a consistent multinannual ground 5 temperature dataset for the Bale Mountains in the southern Ethiopian Highlands, which comprise Africa’s largest tropical alpine area. 29 ground temperature data loggers have been installed at 16 sites since 2017 to characterise and continuously monitor the mountain climate and ecosystem of the Bale Mountains along an elevation gradient from 3493 to 4377 m. At five sites above ∼3900 m, the monitoring will be continued to trace long-term changes. The generated time series provide insights in the spatiotemporal ground temperature variations at high elevation, the energy exchange between the ground surface and atmosphere, as 10 well as the impact of vegetation and slope orientation on the thermal dynamics of the ground. To promote the further use of the ground temperature dataset by the wider research community dealing with the climate and geo-ecology of tropical mountains in Eastern Africa, it is made freely available via the open-access repository Zenodo: https://doi.org/10.5281/zenodo.5172002 (Groos et al., 2021b).


Data loggers
To monitor ground temperature in the Bale Mountains and to establish a modern reference for the paleoclimatic interpretation of periglacial landforms such as the large relict stone stripes on the Sanetti Plateau (Groos et al., 2021c), we have installed two 5 different types of data loggers (see Fig. 2): high-quality UTL-3 Scientific Dataloggers (hereafter abbreviated as GT) and lowcost tempmate.®-B2 ground temperature data loggers (hereafter abbreviated as TM). The GT data loggers are developed by GEOTEST Ltd. in collaboration with the Swiss Institute for Snow and Avalanche Research (WSL). They are mainly deployed to monitor ground temperature and permafrost in high mountain environments (e.g. Hoelzle et al., 1999;Imhof et al., 2000;Schrott et al., 2012;Frauenfelder et al., 2018;Rist et al., 2020). The GT data loggers consist of a waterproof housing, a YSI 10 44005 thermistor for measuring temperature, a memory for up to 65.000 readings, a replaceable 3.6 V lithium battery for the power supply, and a USB 2.0 interface for the data transfer. The measuring accuracy at 0°C is better than ± 0.1°C and the thermometric drift at 0°C is less than ± 0.01°C per 100 months (Geotest, 2021).
The TM data loggers are developed by Tempmate Ltd. and have the size of a button cell. They consist of a splashproof housing, an unspecified thermistor, a memory for up to 8192 readings, and an irreplaceable 3.0 V battery. We wrapped the TM 15 data loggers in thin tape for better protection ( Fig. 2e and f). A logger-pan-to-USB cable is needed for connecting the loggers to a computer and retrieving the data. The measuring accuracy is ± 0.5°C in the temperature range from -10 to 65°C (Tempmate, 2021).

Ground temperature monitoring
To monitor ground temperature and frost occurrence in the Bale Mountains at different depths and elevations, we buried high-20 quality GT data loggers at 2, 10, and 50 cm depth on Tullu Dimtu, on the northern and southern slopes of Mount Wasama, and at two stone stripe locations on the Sanetti Plateau ( Fig. 1, Table 1). The five selected sites are located between 3877 and 4377 m and are all sparsely vegetated. On Mount Wasama, GT data loggers were installed on the northern as well as on the southern slope to analyse the impact of slope orientation on seasonal ground temperature variations. Standard loggers without an external cable were used for the ground temperature measurements near the surface (at about 2 cm depth) and loggers with 25 an external thermistor cable for the measurements at 10 and 50 cm depth ( Fig. 2c and d). Each pit that was excavated for the data logger installation was backfilled in the same order to ensure as little disturbance as possible.
In addition to the fifteen GT data loggers, fourteen low-cost TM data loggers were distributed on the Sanetti Plateau and in two northern valleys (Wasama and Web) to increase the elevation range and spatial coverage of near-surface ground temperature measurements ( Fig. 1, Table 1). Due to the much lower accuracy of the TM data loggers compared to the GT data loggers, 30 we performed a comparative measurement indoor over several hours with logger GT04 as reference. Since the root-meansquare deviation of each TM data logger from the reference measurement was smaller than the stated accuracy of ± 0.5°C, a calibration was unnecessary. For a direct cross-comparison in the field, data logger TM08 was installed next to GT13 in 2 cm depth on Tullu Dimtu. Two TM data loggers (no. 16 and 17) were buried below small Erica trees in 10 cm depth for 5 https://doi.org /10.5194/essd-2021-268  Most data loggers have been monitoring ground temperature at an hourly resolution from January 2017 onward. Others were installed one year later (Table 1). While the low-cost TM data loggers were collected in January 2020, the high-quality GT data loggers continued measuring. The data download needs to be performed on site as a reliable mobile radio network for 5 data transfer is not available in this remote mountain area. During the measurement period from January 2017 to January 2020 (last read-out), several issues occurred and caused data gaps in many of the ground temperature time series (Fig. 3). The position of the data loggers in the field was originally marked with small coloured plastic poles. However, the markers were too conspicuous. Vandalism led to the loss of several items (GT01, GT04, GT18, TM01 to TM03, TM08, and TM11). Dwarf shrubs and stones were used subsequently to mark the measurement sites.

Data post-processing
All ground temperature data stored by the GT and TM loggers were checked manually and automatically using a simple filter to identify erroneous values as temporarily recorded by a few low-cost loggers. We only removed hourly measurements from the time series that were either implausible (e.g. values in the order of -20 to -40°C) or that deviated from previous or subsequent 5 measurements by more than ±10°C. No other loggers than TM05, TM06, and TM07 were affected by this correction. Data logger GT07 was unintentionally installed at about 6 cm depth in January 2017 and not at 2 cm as planned. Because of its relocation towards the surface after the first read-out in December 2017, an increase in the daily temperature amplitude was noticed. To correct for this, we calculated hourly ground temperature gradients between 6 and 10 cm depth from the GT07 and GT08 data by applying a simple linear regression model. We used the obtained gradients to extrapolate the GT07 measurements 10 from 6 to 2 cm in the period 21 January to 10 December 2017. All measurements of any data logger that were not recorded on the full hour were adjusted to the full hour by linear interpolation. See Section 6 ("Data availability") for access to the original logfiles and for further information regarding any corrections made to each time series.
To obtain a complete and consistent data set of hourly ground temperatures in the Bale Mountains for the period 1 February 2017 (first measurement) to 20 January 2020 (last read-out), we applied a statistical gap-filling approach. Most of the ground 15 temperature measurements from different locations or depths overlap for a certain period in time (see Fig. 3) and allow a statistical correlation to be established. We applied a simple linear regression model to interpolate missing data points in the time series of a logger using existing data from a nearby logger. If multiple loggers with a similar distance were considered for the interpolation, we chose the one that yielded the best fit (i.e. the highest coefficient of determination R²) and lowest root-mean-square error (RMSE). The overlapping measurement period between the predicting logger and dependent logger 20 was split into a calibration and validation part. For the interpolation of incomplete time series in 10 or 50 cm depth, we drew on existing data from 2 cm depth of the same location. We used a moving average of the data from 2 cm depth to account for the time-lag response in greater depths to meteorological changes. The number of preceding hours considered for the calculation of the moving average that yielded the best prediction (high R² and low RMSE) of the ground temperatures in 10 or 50 cm depth was chosen.

25
The time series of the data loggers TM08 and TM15 to TM17 were not interpolated as the data served only for comparative experiments (low-cost vs. high-quality loggers and vegetated vs. barren locations) and were dispensable for the temporal and elevational analysis. See Section 6 ("Data availability") for access to the final ground temperature dataset and a detailed information sheet regarding the gap-filling of the individual time series.

Data analysis 30
As the main focus of this contribution is the presentation and publication of the generated ground temperature dataset, we conducted a basic statistical analysis to quantify frost occurrence and spatio-temporal ground temperature variations in the Bale Mountains. We also included in the analysis meteorological data from the Tuluka automatic weather station (AWS) on the southern Sanetti Plateau to better understand the factors controlling ground temperature variations. The meteorolgoical data are accessible through an on-demand processing database system within the framework of the joint Ethio-European Research Unit 2358 "The Mountain Exile Hypothesis" (Wöllauer et al., 2020), but they have not yet been made publicly available. Fourteen data loggers (excluding TM08) from 2 cm depth and five loggers (excluding TM14 to TM17) from 10 and 50 cm depth were considered for the calculation of mean annual ground temperatures, daily ground temperature cycles, thermal gradients, 5 number of frost days, frost penetration depth, and elevational gradients. To emphasise seasonal ground temperature variations related to changes in insolation, cloudiness, and humidity, we conducted the calculations separately for the entire study period, the dry season (Bega: November to February), and the two rainy seasons (Belg: March to June; Kiremt: July to October).
Moreover, time series from different sites were compared to investigate differences in ground temperature between northfacing and south-facing slopes (GT16, GT02, and GT03 vs. GT17, GT05 and GT06), to study differences between vegetated 10 and sparsely vegetated areas (TM16 and TM17 vs. TM14 and TM15), and to assess differences in the performance of low-cost and high-quality data loggers (TM08 vs. GT13).

Data quality
Both the high-quality and low-cost data loggers have reliably and accurately recorded ground temperature at an hourly resolu- 15 tion as long as the power supply was ensured. We noticed a relatively short battery life of two years for some of the GT and TM data loggers, leading to a substantial data loss between consecutive read-out dates. Two years are shorter than the battery life stated by both manufacturers for the hourly sampling interval (GT: ca. 3-5 years; TM: ca. 5 years). The temporary power loss caused longer data gaps in some ground temperature time series. Implausible ground temperature measurements that were caused by a drop in battery voltage were only noticed in the time series of three low-cost loggers (TM05, TM06, and TM07). 20 Another reason for shorter data gaps was the limited memory capacity of the low-cost loggers, which was insufficient if the period of hourly measurements between two consecutive read-out dates was longer than 341 days.
Because of the high number of thermistors installed in the Bale Mountains, data gaps in the affected ground temperature time series could be interpolated using hourly data from other measuring sites. The validation of the simple linear regression models applied for the interpolation of the time series revealed a strong correlation between the measured and predicted ground 25 temperatures with an R² of 0.85 ± 0.13 and a RMSE of 1.9 ± 1.4°C (the provided uncertainty is the standard deviation of all model performances). The relatively high RMSE is the result of the great diurnal ground temperature amplitude close to the surface (see Section 4.2).
The cross-comparison between the low-cost data logger (TM08) and high-quality data logger (GT13) at the summit of Tullu Dimtu revealed a strong correlation (R² = 0.98) between the measured ground temperatures. Both loggers measured almost the 30 same mean ground temperature (8.44 vs. 8.48°C). Only the standard deviation of the TM08 measurements was slightly larger than that of the GT13 measurements (9.0 vs. 7.3°C) as GT13 was installed at a slightly greater depth than TM08. This shows that the tested low-cost loggers, which have not been explicitly designed for scientific applications, are suitable for short-term and mid-term (days to months) ground temperature measurements and experiments in (tropical) mountains.

Ground temperature variations
The ground temperatures observed in the Bale Mountains from January 2017 until January 2020 show characteristic short-term and long-term variations (Fig. 4). On the highest peak Tullu Dimtu (4377 m), daily mean ground temperatures fluctuate around 7.6°C and range between minimum 3°C and maximum 12°C. The mean multiannual air temperature at the same site is  (Fig. 4).
The analysis of meteorological data from the Tuluka AWS (3848 m) on the southern Sanetti Plateau reveals that incoming shortwave radiation in the Bale Mountains follows a clear seasonal cycle (Fig. 5c). The incoming shortwave radiation reaches its maximum during the dry season and is generally reduced during the two rainy seasons from March to October because of the frequent presence of clouds. In contrast to that, daily air temperatures are highest during the rainy seasons and lowest during 15 the dry season (Fig. 5b). The asynchronicity between the seasonal maxima of daily air temperature and incoming shortwave radiation can be explained by variations in the net longwave radiation flux, which is not directly measured by the AWSs in the Bale Mountains. The increased fraction of water vapour in the atmospheric boundary layer above the Sanetti Plateau during the rainy season as indicated by the relative humidity in Fig. 5d leads to a greater nocturnal absorption of the outgoing longwave radiation and, thus, to a greater warming of the atmosphere than during the dry season. Ground temperature variations on the 20 Sanetti Plateau in turn do not simply reflect changes in the net shortwave radiation flux, air temperature or air humidity (Fig.   5a). To a certain degree, they are also controlled by ground moisture (indicated by the precipitation sum in Fig. 5e), which affects the surface energy balance through evaporative cooling and heat absorption.
The impact of clouds as well as air and ground moisture on the surface energy balance is also reflected in the diurnal ground temperature cycle. During the dry season, the diurnal ground temperature amplitude in the Bale Mountains is in the order of 25 15-25°C and, thus, more pronounced than the daily amplitude (ca. 10-15°C) during both rainy seasons (Fig. 6a) Sanetti Plateau up to 100 d per year. However, temperatures below 0°C were measured exclusively near the surface as the freezing front penetrates only the uppermost centimetres of the ground. At 10 cm depth and below, frost was not detected at any of the logger locations during the entire study period.
The diurnal ground temperature amplitude decreases considerably with depth. Ground temperatures at 50 cm depth and below vary little throughout the day (Fig. 6a). The difference between the mean daily ground temperature near the surface and at 50 cm depth is relatively constant and rarely larger than 2°C (Fig. 6b). Although the effect is not very pronounced, the thermal gradient from the surface to 50 cm depth tend to be negative during the dry season and constant or positive during the rainy seasons. Annual ground temperatures increase from the highest peak Tullu Dimtu (4377 m

Influence of slope orientation and vegetation on ground temperature
The comparative experiment on Mount Wasama in the northern part of the Bale Mountains ( Fig. 1) shows clear differences between the thermal regime of the southern and northern slopes (Fig. 7a). The southern slope is on average more than 2°C 10 warmer and reveals a more pronounced seasonality and larger diurnal amplitude, which favours freezing and thawing and might explain the exclusive presence of solifluction lobes on the southern slope. While the mean daily temperature on the southern slope peaks towards the end of the dry season (January to February) when the sun is in its zenith, it reaches its maximum on the northern slope a few month later when the sun approaches its northernmost position.
The ground temperature differences between vegetated and unvegetated areas on the Sanetti Plateau are less obvious (Fig. 7b). 15 Small Erica trees and bushes reduce the diurnal temperature amplitudes of the ground they are shading, but the vegetation itself has only little impact on the seasonal ground temperature variations. Like on Mount Wasama, both south-exposed logger locations on Tullu Dimtu (vegetated and unvegetated; TM16 and TM15) have their temperature maxima at the end of the dry season. The vegetated and unvegetated monitoring sites in the flat part of the plateau (TM17 and TM14) heat up rather during May to July. This means that slope orientation has a larger impact on long-term ground temperature variations, whereas 20 vegetation mainly affects short-term variability and the diurnal amplitude.

Discussion
The from an afro-alpine study site. Many of the installed data loggers were collected in January 2020 after three years of operation, but the hourly ground temperature monitoring will be continued at five sites between 3877 and 4377 m on the Sanetti Plateau and on Mount Wasama (see Fig. 1) to study the long-term climate and environmental change at high elevation. The data that will be obtained in the future will also be made publicly available via the repository stated in Section 6 ("Data availability"). However, the advantage of a denser monitoring network is twofold: firstly, a greater variety of locations can be monitored; sec-5 ondly, data gaps in one particular time series can be interpolated using data from a nearby monitoring site. The large number of monitoring sites in the Bale Mountains enabled us to generate a gapless three-year hourly ground temperature dataset. Since other deterministic or stochastic methods such as machine learning might further improve the interpolation of data gaps in the original time series (e.g. Lepot et al., 2017), we publish the original logfiles along with the final ground temperature dataset.
Although an in-depth evaluation of the presented dataset is beyond the scope of this contribution, the conducted statistical anal-  The comparison of the observed ground temperatures with meteorological data from the Sanetti Plateau shows that the mean annual air temperature is several degrees lower than the ground temperature at the same elevation. The offset can be explained by the strong insolation and relatively low air density of the highest tropical mountains. A similar offset between the mean annual air temperature and ground temperature has also been observed on Kilimanjaro (Yoshikawa et al., 2021). Moreover, the comparison on the Sanetti Plateau reveals that temporal ground temperature variations are predominantly controlled by 25 fluctuations in the net radiation as well as changes in the ground water content, which regulates the thermal balance through heat absorption and evaporative cooling (Fig. 5). Vegetation dampens the diurnal ground temperature amplitude, whereas slope orientation determines the seasonal timing of the ground temperature maxima (Fig. 7). To calculate long-term ground temperature trends and assess ongoing climate and environmental change in the afro-alpine belt, the established time series need to be further extended.  hans et al., 2016). Due to the large number of monitoring sites in the Bale Mountains, the ground temperature data could be analysed in a similar way to generate maps of spatial ground temperature variations. Moreover, correlations between monitored air temperature and ground temperature can principally be used to generate air temperature maps from remotely-sensed land surface temperatures (e.g. Pepin et al., 2016). In the Ethiopian Highlands, distributed meteorological, ecological, and ground 5 temperature data are of particular interest to better understand the relationship between spatial ground temperature variations and the scattered distribution of Erica trees (Miehe and Miehe, 1994;Lemma et al., 2019;Mekonnen et al., 2019) as well as the activity patterns of the endemic giant root rat ("Tachyoryctes macrocephalus") on the Sanetti Plateau (Vlasatá et al., 2017).
Besides the aforementioned ecological topics, in-situ ground temperature data are also required to study freeze-thaw cycles in the afro-alpine belt with the aim to elucidate the implications for the formation of contemporary periglacial landforms (e.g. 10 Grab et al., 2004). Moreover, the current measurements are required as a modern reference to estimate the Late Pleistocene cooling that probably provided the preconditions for the formation of the relict sorted patterned ground on the Sanetti Plateau (Groos et al., 2021c). Given that the monitoring is continued successfully over the next years, the extended ground temperature dataset may be evaluated in terms of the elevation-depended warming observed in other mountain ranges worlwide (e.g. Pepin et al., 2015). Eventually, the ground temperature dataset may also be used to validate satellite-based or drone-based thermal 15 imagery (e.g. Kraaijenbrink et al., 2018) and, in combination with the meteorological data, to evaluate the performance of regional climate models in the mountains and highlands of Eastern Africa (e.g. Collier et al., 2019).
The folder "Ground_Temperature_Data" contains the followings files in ods and csv formats (the date format is DD.MM.YYYY hh:mm East Africa Time): -"Hourly_Ground_Temperatures_Corrected": Compilation of corrected hourly ground temperature data from all GT and TM data loggers installed in the Bale Mountains (see Table 1). The time series of each logger begins with the start of 25 measurement and ends with the last readout. Some time series contain data gaps (see Fig. 2).
-"Information_Sheet_Data_Correction": Overview table regarding all modifications applied to any of the original ground temperature data.
-"Hourly_Ground_Temperatures_Interpolated": Compilation of complete (i.e. interpolated) hourly ground temperature time series from all GT and TM data loggers (except TM08 and TM15 to TM17; see Section 3.3 "Data post-processing") 30 for the period 1 February 2017 (first measurement) to 20 January 2020 (last read-out).
-"Information_Sheet_Data_Interpolation": Overview The folder "GT_Logfiles" contains the original logfiles of all GT data loggers (see Table 1) in text format (the date format is YYYY.MM.DD hh:mm:ss East Africa Time). The folder "TM_Logfiles" contains the original logfiles of all TM data loggers (see Table 1) in text format (the date format is DD.MM.YYYY hh:mm:ss East Africa Time).     ground temperature variations at 10 cm depth at sites with an Erica cover and sites without. Note that the data loggers TM14 and TM17 are located on west-exposed slopes while TM15 and TM16 are located on south-exposed slopes. A local regression with a smoothing span of 0.32 was applied to derive long-term ground temperature variations from hourly measurements.