Assessing Wind Energy Potential and Wind-Induced Risks at the Princess Elisabeth Antarctica Research Station

The Princess Elisabeth Antarctica Research (PE) station is a “zero emission” research station powered by wind and solar energy. The station’s popularity has become so prominent that expansion of energy capacity for research is planned. For this purpose, the wind energy potential surrounding the current station must be investigated to identify the best wind turbine locations.

Due to extreme weather conditions on the Antarctic plateau, and the typical katabatic wind-induced risks of Antarctica, it's challenging to perform accurate wind measurements using a meteorological mast system. Currently, an old 30m antenna mast is placed East of the station which could potentially be equipped with ultra-sonic anemometers. However, due to the complexity of the wind profile at the station due to the Sør-Rondane mountain range and Utsteinen Nunatak to the South, this single 30m-mast will not be sufficient to provide an accurate and reliable wind assessment.

Instead, wind measurements can be made remotely by a wind-doppler LiDAR, which measures not only over the entire reach of a wind turbine, but up to 300m aloft making an incredibly interesting case (and an Antarctic Plateau’s first) for meteorological validation of numerical weather prediction (NWP) models, as well as the best option for a wind assessment. This remotely sensed wind data will be accompanied by weather stations operated by PE’s team for several seasons already. With this high-resolution wind data reaching up to 300m aloft, NWP models can be used in novel approach of machine learning (ML) to reconstruct a wind profile of the entire area surroundings of the station.

Our goals are:
1) to reconstruct a high-resolution wind profile near the surface and aloft around the PE station for wind energy potential assessment, and 
2) to publish this novel collection of wind profiles, synchronized with existing meteorological datasets in the region to the public, with our findings on ML architecture, the significance of variables required for accurate wind assessment and extreme weather event identification on the Antarctic Plateau.

With CRYOS at the forefront of these wind assessment techniques using ML models in combination with remotely sensed wind data in the Swiss Alps, [1-2] and our extensive collaboration with the PE station during previous field trips studying snow physics. [3-5] We do not only possess a great deal of knowledge of operations at PE station, but are in the position to apply our cutting-edge meteorological tools for weather model validation and wind-induced risk assessment.

The Princess Elisabeth Antarctic Research (PE) station is located North of the Sør-Rondane mountain range, just over 200km inland in the transition from the coast to the Antarctic Plateau. The PE station is designed to be a “zero-emission” research station where energy is generated using wind and solar energy sources around the station itself. [6] The station operates only during Austral summer for a period of approximately four months where an effort is made to use only renewable energy sources in all activities at the station.

Due to the large success of the PE station’s construction in 2007-08, and the continuous expanse of the field work activities in the area of operations since then, the International Polar Foundation (IPF) who manages the station is looking to expand on their energy supply for future seasons and to replace old wind turbines that are no longer adequate or serviceable.

As conditions on the Antarctic Plateau are extreme, it proves to be a challenging endeavour to perform state-of-the-art Wind Energy Assessment (WEA) measurements to understand where new wind turbines can be placed. A WEA usually spans over several months of 10min averaged wind speed measurements in all three spatial dimensions and several heights above ground to get a clear view of wind patterns and shear throughout the lower atmosphere. However, when the topography of the nearby terrain is complex, the flow of the wind gets disturbed, and the measurements done at a certain location are only valid for a small area.

In general, a meteorological mast with ultra-sonic anemometers along its length is constructed in the vicinity to perform such measurements, after which the wind profiles are analysed and an estimate of wind power performance is given, as stated by the IEC 61400 standards: “Wind measurements are to be made over the course of one year up to at least two-thirds of the wind turbine height”. However, it is often not clear how this single location would compare to neighbouring locations, for example on the other side of the ridgeline where wind conditions may be drastically different. At PE station in particular, the proximity of Utsteinen and Nunatak near the PE station and the vast size and high mountain tops and passages of the Sør-Rondane mountain range cause large disturbances in the southernly flows. Such complex terrain has sparked great scientific discussions on which methods should be applied instead of the IEC standard, as it is rendered unapplicable in these terrains. [7-9]

The convex shape of Antarctica’s ice shelf, the extreme cold temperatures and the polar day/night cycle and the Sør-Rondane mountain range lead to interesting wind phenomena that are seldom observed in any other place on Earth. A key wind phenomenon that occurs in an almost routinely fashion is the katabatic wind. [10-14] Antarctic katabatic winds find their origin near the top of the icesheet after which they accelerate due to the gravitational and Coriolis force producing extremely violent and cold wind gusts often exceeding 50 m/s. [15-16] The Sør-Rondane mountain range is also theorized to be a large influence on katabatic winds towards the PE station, which makes a strong case for applying such rigorous analysis of the wind profile to better understand these meteorological phenomena.

Lastly and although quite uncommon, no-wind conditions at PE-station can be devastating for their zero-emission operations as the station will solely rely on solar energy sources. Thus, forecasting of prolonged periods of little to no wind can help in planning energy conservation at PE station to overcome generation deficits. All such wind-induced risks can benefit from more accurate wind measurements to improve the wind assessment around PE station, and it is crucial in better understanding the environmental risks for personnel and infrastructure in field work and permanent installations. [17]

We therefore run into three main challenges, (1) the wind profile measurements in the lower atmosphere in the extreme environments, (2) the analysis of a large area around PE station to find the most suitable wind turbine locations, (3) the wind analysis for wind turbine specification. The latter can also be used in risk assessment and as complementary information for field activities on the Antarctic Plateau.

By applying a novel method of wind energy assessment using Machine Learning (ML) as a post-processing downscaling model that can be applied on Numerical Weather Prediction (NWP) models (more detail provided in “Overarching scientific project towards wchih the field trip will contribute”), wind measurement can be taken remotely from a wind-doppler Light Detection and Ranging (LiDAR) system which uses the back-scattering of laser light from aerosols in the atmosphere to obtain wind components along the laser path. Measurements from different heights above ground can be used to reconstruct a wind profile. Wind measurements are made in the four cardinal wind directions (28° from the vertical) and vertically to reconstruct the entire wind profile up to 300m above ground level. [18]
Using data from the existing network of weather stations operated by IPF around the PE station, several ultra-sonic anemometers placed on key terrain features, and the LiDAR wind profiles, a ML algorithm, such as successfully applied in the Swiss Alps [1-2] can be trained and tested.
It has furthermore been proven that ML can be put to good use in the identification of extreme weather events. As a multi-seasonal database has already been collected by the meteorological stations operated around PE, the algorithm can be trained specifically to identifying dangerous wind situations that may be occurring in the hours ahead. [10,12,14] The selection of wind profiles collected during this field trip will enable a validation of the model as the LiDAR will be located much closer to PE station than the other meteorological stations.

During this field trip, the LiDAR system will be installed on the southern side of the PE station in proximity to an electricity access point, located downwind of the Sør-Rondane mountain range, Utsteinen and Nunatak (see map attached in proposal [copyright Google Maps]), to collect wind profiles at 1-minute intervals for the entire duration of the field trip. Ultra-sonic anemometers will be placed near the LiDAR system on key local topographic features such as rocks, ridgelines and along preferential wind directions at PE station to get a clear picture of winds during a typical Austral summer period. The collected data set will be integrated with the meteorological stations into one large training set for the ML algorithm.

Using the CRYOS professional mapping drone, which has been deployed previously at PE station, a high-resolution digital elevation model can be generated with the most up-to-date topography, snow height, and surface roughness. This data will be used for linking near-surface winds to winds aloft.

After the field expedition, the collected data will be cleaned, quality controlled, and potentially bootstrapped (generating “pseudo” data of a specific wind event to increase its frequency in the dataset to allow the ML model to better identify rare extreme wind events, we refer to [19] for a basic overview of bootstrapping data). Then, adaptions on the ML model architecture will be made to adapt it for the Antarctic weather conditions and their specific NWP models. The resulting model architecture will be trained and tested using the dataset from which a wind assessment of the area around PE station will be made.

This field trip will contribute to the PhD project “Accurate wind measurements and forecasting in the complex terrain for wind energy production” which is co-funded by the EPFLglobaLeaders [20] Doctoral Fellowships of the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement (No. 945363), and the Laboratory of Cryospheric Sciences (CRYOS) at École Polytechnique Fédérale de Lausanne. The PhD lasts for 48 months until October 2026 with the aim of designing and implementing novel terrestrial remote-sensing wind measurement tools for complex terrain and extreme environments.

To address the emissions caused by humanity over the past centuries, efforts are made to introduce more renewable energy sources to replace polluting energy sources all over the globe. In particular, renewable wind energy becomes a remarkable option covering the baseload energy demands and providing energy during night and winter periods when solar energy is not or less available. Besides, nuclear energy is becoming less available due to the lack of societal support and corresponding political implementation (such as in Switzerland [21]), or limitations of construction (such as in Antarctica [22]).

Several efforts are being made to investigate the wind energy potential of various location all over the globe, though the vast complexity and chaotic nature of the weather system makes a universal solution improbable due to the large variety of boundary conditions and intricate processes in the atmosphere. [23-24] Instead, models are designed for a specific spatial and temporal scale while keeping a reasonable computational cost and time such that model predictions are timely delivered. [23]

On the synoptic scale (order of 10-1’000km), Numerical Weather Prediction (NWP) models apply basic physical laws combined with the most accurate data describing the given state of the atmosphere as initial and boundary conditions and for assimilation into the model to predict a (specific) future weather situation and state of the atmosphere.

Based on the output of such large-scale NWPs, Regional Weather Models (RWM) (order of several kilometres) apply the same laws of physics to model more intricate local phenomena at smaller spatial and temporal resolution, including a simplified version of the topography to get a more detailed weather forecast. In Switzerland, the COSMO and ICON models are run down to a resolution of 1 km.

Based on these RWM outputs, often subject to relatively large biases as they do not include some of the most intricate physical effects in their calculation, statistical bias correction can be applied where actual weather measurements can be compared to the prediction. These relations can then be applied in reverse to the model output itself to reduce the model bias. In recent years, the advent of neural networks has worked its way into these bias corrections and on some occasions simple neural networks are successfully used. [1,25-27]

More recently, CRYOS has taken the post-processing of the RWM outputs a step further by applying ML models that use high-resolution topography of the Swiss Alps and combines this with the RWM wind prediction. [1] By training this model on data from Swiss alpine weather stations, this ML model was able to spatially downscale the RWM wind prediction from 1.1km to 53m whilst almost totally removing bias introduced by the RWM.

Such results are excellent for wind profile estimation and wind energy assessment, they rely on local measurements to train the ML model for generating accurate predictions. Most wind measurements are taken around 10 m above the surface, a level still quite easily accessible. Higher elevation measurements are scarce as much bigger infrastructure and logistics are required for such endeavours.

In state-of-the-art of wind assessment, complicated wind profiles resulting from flow disturbances due to complex topography cannot be modelled by standard NWP models and require more advanced and computationally expensive computational fluid dynamics models to provide a realistic and sufficiently accurate wind assessment. [9,28-29] It has been shown by previous work of CRYOS that the use of ML as a post-processing tool to NWP models can remove the model prediction bias altogether and reduce errors significantly in the Swiss Alps. [1]

To provide accurate wind assessments for wind turbine installations, wind predictions must be made around turbine height. However, measurements at the hub height of wind turbines are challenging in extreme environments and complex terrain due to complicated logistics and safety issues. Therefore, new remote-sensing tools need to be deployed in combination with advanced ML models that combine NWP model output with high spatial resolution measurements around the hub height to spatially resolve the wind speed distribution and to reduce errors and biases. With such methods, wind forecasting will become much more accurate and precise at a given location. Furthermore, multi-seasonal wind measurements using tall meteorological masts of several tens of meters used for wind power assessments may be replaced in the future at sites where these high-cost, high-risk masts cannot be operated and/or serviced.

The principal innovative idea is to combine a central wind-doppler LiDAR system with surrounding ultra-sonic anemometers, suitably placed in the complex terrain to complement the LiDAR measurements in areas which are outside the field of view of the LiDAR.

We then combine a ML model with these measurements to, (1) reconstruct the full 3-dimensional wind field, and (2) relate the local flow to either meteorological model output or long-term weather station data observed in the vicinity of the LiDAR site. The aim is the re-construction of the local flow situation (including the wind profile) for selecting an optimal wind turbine site and for predicting high-resolution wind profiles between 0 and 300m above ground level. Furthermore, these results can be used to estimate wind potential over longer time scales based on the ML model and corresponding long-term weather data.

The CRYOS laboratory enjoys an excellent relationship with the International Polar Foundation (IPF) which operates the Princess Elisabeth Antarctic Research Station (see letter of support attached to this application). CRYOS also operates its own field measurement setup for drifting snow near the PE station since several years. The visits over the past seasons, resulted in very detailed knowledge of the station, its staff, and the surrounding environment.

Access to PE (personnel and cargo) is by airplane via Cape Town, South Africa; logistics are handled by the IPF and the Antarctic Logistics Center International Ltd (ALCI). All regulations arising from the Antarctic Treaty will be fully considered and respected. CRYOS will provide its own Wind-doppler LiDAR system including auxiliary meteorological sensors and equipment to construct and maintain the setup throughout the field campaign. Also, a professional mapping drone is available from CRYOS for detailed local terrain measurements which will be required for numerical wind field reconstruction. This drone already has been operated successfully at the PE station in previous field campaigns by our laboratory staff.

The LiDAR system will be located a few hundred meters away from the PE station as to not interrupt other activities around the station and to ensure an undisturbed data acquisition throughout the field campaign. The LiDAR will be connected to the electrical grid of the station ensuring continuous power supply.

The main risks of this field trip are seen in the logistics. The LiDAR system requires careful handling in transport and is paramount for the success of the project. Our laboratory has ample experience with cargo transport to the PE station in Antarctica and logistical plans will be drafted much in advance of the expedition. This will include precautions for logistical delays and incidents. The second uncertainty is the weather conditions, which could lead to delays in the schedule. This can be mitigated by planning for a sufficiently long field period so that delays will not compromise the execution of the intended work.

All expedition participants have to pass a thorough medical check prior to departure. Furthermore, they participate in a mandatory safety training provided by specialists of the IPF; modules include basic GPS navigation and orientation, crevasse rescue, first aid and skidoo training.

Any outdoor activities will be planned considering weather forecasts and conditions and after approval by the responsible station commander. When needed, professional field and mountain guides will accompany scientists in the field. Thus, we are confident that our project can be completed successfully without any harm to personnel, equipment, and the environment.

Since the LiDAR will be installed in the vicinity of the PE station, it can be connected to the electrical grid of the station, which ensures continuous and reliable power supply for the deployed systems.

To the best of our knowledge, the wind profile data that will be collected during this field campaign will be unique and the first of such thorough wind measurements aloft on the Antarctic Plateau. With our previously verified ML methods in the Swiss Alps, we theorise that a similar application on the complex flows around PE station will be of similar success in producing results of very low bias and with reduced error compared to other state-of-the-art wind assessment methods.

A six-week measurement campaign should provide us with sufficient data to teach the ML algorithm the vastly differing wind situations that are observed on the Antarctic ice sheet. As a single such event can be used as to bootstrap the dataset to enhance risk scenario identification.

We aim to publish the data of the experimental campaign and the analysis including ML architecture. The original, cleaned and quality-controlled dataset will be made available by its own DOI, and intermediate datasets may be supplied upon request to interested scientists. Besides this data, the meteorological network operated by IPF around PE station is already publicly available, but will be synchronized and appended to our measurement campaign dataset in publication for a complete meteorological overview of the area.

The wind assessment made around the PE station will serve as an objective overview for policy making at the station from which a decision can be made on the expansion of their renewable energy infrastructure.

The project partners recognize the urgency of the climate crisis and the need for CO2 emission reduction on a global scale. Our field trip has the main objective of enabling low-carbon footprint research on Antarctica and therefore the logistics related to this field trip will be carbon-offset.

The majority of emissions result from transport of personnel towards Antarctica. Therefore, the flights from and to Antarctica will be fully offset.

The candidate did an internship at the University Centre in Svalbard (UNIS) while completing his MSc degree. For three months, the candidate worked on developing a fixed-wing drone with a hyper-spectral camera to scan for Aurora Borealis above the cloud layer. This internship was performed in collaboration with UNIS, the Eindhoven University of Technology (TU/e), and the Netherlands Aerospace Centre (NLR). However, no field trip activities were performed during this internship. All activities were confined in the boundaries of the town of Longyearbyen.

Furthermore, the candidate worked on data analysis of temperature data on Svalbard, all this work was performed fully remotely.

A letter of recommendation from the Station Commander at Princess Elisabeth Antarctica Research Station (2023 PEA Brandon Van Schaik.pdf) is attached separately.

A letter of recommendation from the candidate's supervisor prof. Michael Lehning (LetterRecommendationvanSchaik.pdf), is attached separately.

The map (LiDAR_at_PE_station2023.pdf) is attached separately.

The candidate’s CV (CV_Brandon_van_Schaik.pdf) is attached separately

Regarding the budget: The total budget of the expedition amounts to at least 35’000 CHF. The CRYOS lab will cover all costs beyond the call limit.

[1]

J. Dujardin and M. Lehning, “Wind-Topo: Downscaling near-surface wind fields to high-resolution topography in highly complex terrain with deep learning,” Quarterly Journal of the Royal Meteorological Society, vol. 148, no. 744, pp. 1368-1388, 2022.

[2]

F. Kristianti, J. Dujardin and M. Lehning, “Combining Weather Station Data and Short-Term LiDAR Deployment to Estimate Wind Energy Potential with Machine Learning: A Case Study from The Swiss Alps,” Manuscript submitted for publication, 2023.

[3]

C. G. Sommer, N. Wever, C. Fierz and M. Lehning, Dataset: Expedition to Princess Elisabeth Antarctica Station, 2016/2017, Davos, Switzerland: SLF, 2017.

[4]

N. Wever, E. Keenan, C. Amory, M. Lehning, A. Sigmund, H. Huwald and J. Lenaerts, “Observations and simulations of new snow density in the drifting snow-dominated environment of Antarctica,” Journal of Glaciology, pp. 1-18, 2022.

[5]

C. Sommar, N. Wever, F. C. and M. Lehning, “Investigation of a wind-packing event in Queen Maud Land,,” The Cryosphere, vol. 12, no. 1, pp. 2923-2939, 2018.

[6]

International Polar Foundation, “Antarcticstation.org,” [Online]. Available: http://www.antarcticstation.org/station.

[7]

S. Barber, M. Buehler and H. Nordborg, “The wide range of factors contributing to wind resource assessment accuracy in complex terrain,” Wind Energy Science, vol. 7, no. 1, pp. 1503-1525, 2022.

[8]

R. Barthelmie, H. Wang, P. Doubrawa, G. Giroux and S. Pryor, “Effects of an escarpment on flow parameters of relevance to wind turbines,” Wind Energy, vol. 19, no. 12, pp. 2271-2286, 2016.

[9]

R. DeLeon, M. Sandusky and I. Senocak, “Simulations of Turbulent Flow Over Complex Terrain Using an Immersed-Boundary Method,” Boundary-layer meteorology, vol. 167, no. 3, pp. 399-420, 2018.

[10]

M. Pichault, C. Vincent, G. Skidmore and J. Monty, “Short-Term Wind Power Forecasting at the Wind Farm Scale Using Long-Range Doppler LiDAR,” Energies, vol. 14, no. 2663, pp. 1-21, 2021.

[11]

D. Qi and A. Majda, “Using machine learning to predict extreme events in complex systems,” Proceedings of the National Academy of Sciences - PNAS, vol. 117, no. 1, pp. 52-59, 2020.

[12]

J. Meiyazhagan, S. Sudharsan, A. Venkatesan and M. Senthilvelan, “Prediction of occurrence of extreme events using machine learning,” European physical journal plus, vol. 137, no. 1, pp. 1-20, 2021.

[13]

R. Eskandarpour and A. Khodaei, “Machine Learning Based Power Grid Outage Prediction in Response to Extreme Events,” IEEE transactions on power systems, vol. 32, no. 4, pp. 3315-3316, 2017.

[14]

S. Srinivas Kolukula and P. Murty, “Improving cyclone wind fields using deep convolutional neural networks and their application in extreme events,” Progress in oceanography, vol. 202, no. 1, p. 102763, 2022.

[15]

M. R. van den Broeke, N. van Lipzig and E. van Meijgaard, “Momentum budget of the East Antarctic atmospheric boundary layer: Results of a regional climate model,” Journal of the atmospheric sciences, vol. 59, no. 21, pp. 3117-3129, 2002.

[16]

A. Davis and R. Mcnider, “The development of antarctic katabatic winds and implications for the coastal ocean,” Journal of the atmospheric sciences, vol. 54, no. 9, pp. 1248-1261, 1997.

[17]

E. Fossat, “Some results after 10 years of site testing at Concordia, Antarctica,” Ithaca: Cornell University Library, arXiv.org, vol. 1, no. 1, pp. 1-15, 2011.

[18]

VAISALA, “VAISALA WindCube v2.1,” 1 2023. [Online]. Available: https://www.vaisala.com/en/wind-lidars/wind-energy/windcube. [Accessed 17 1 2022].

[19]

T. Joseph, “Towards datascience,” Towards datascience, 17 6 2020. [Online]. Available: https://towardsdatascience.com/bootstrapping-statistics-what-it-is-and-why-its-used-e2fa29577307. [Accessed January 2023].

[20]

Ecole Polytechnique Fédérale de Lausanne, “EPFLglobaLeaders Doctoral Fellowship,” [Online]. Available: https://www.epfl.ch/education/phd/doctoral-studies-structure/customized-curricula/epflglobaleaders/.

[21]

Swiss Federal Office of Energy, “Nuclear Energy,” 02 07 2020. [Online]. Available: https://www.bfe.admin.ch/bfe/en/home/supply/nuclear-energy.html.

[22]

W. Shafer, “Five Years of Nuclear Power at McMurdo Station,” Antarctic Journal of the United States, vol. 2, no. 2, pp. 38-39, 1967.

[23]

Consortium for Small-scale Modeling, “Consortium for Small-scale Modeling,” [Online]. Available: https://www.cosmo-model.org/.

[24]

S. Emeis, “Winds in Complex terrain,” in Winds in Complex Terrain, Copenhagen, Springer, 2012, pp. 75-93.

[25]

A. Winstral, T. Jonas and N. Helbig, “Statistical Downscaling of Gridded Wind Speed Data Using Local Topography,” Journal of Hydrometeorology, vol. 18, no. 2, pp. 335-348, 2017.

[26]

P. Sheridan, S. Vosper and S. Samantha, “A Physically Based Algorithm for Downscaling Temperature in Complex Terrain,” Journal of applied meteorology and climatology, vol. 57, no. 8, pp. 1907-1929, 2018.

[27]

H. Huang, C. Scott, S. Huang and A. Hall, “Downscaling near-surface wind over complex terrain using a physically-based statistical modeling approach,” Climate dynamics, vol. 44, no. 1-2, pp. 529-542, 2014.

[28]

I. Troen, A. Bechmann, M. Kelly, N. Soerensen, P. Réthoré, D. Cavar and H. Joergensen, “Complex terrain wind resource estimation with the wind-atlas method: Prediction errors using linearized and nonlinear CFD micro-scale models,” European Wind Energy Association Conference and Exhibition, 2014.

[29]

S. Koller, B. Schaffner, M. Hertach, M. Geissmann, T. Schlegel and D. Kroepfli, “Windatlas Schweiz: Jahresmittel der modellierten Windgeschwindigkeit und Windrichtung,” MeteoTest, Bern, Switzerland, 2022.