METEOROLOGICAL SOLUTIONS TO SUPPORT WIND ENERGY PRODUCTION IN FINLAND
CONTRIBUTIONS
KAROLIINA HÄMÄLÄINEN
177
CONTRIBUTIONS No. 177
METEOROLOGICAL SOLUTIONS TO SUPPORT WIND ENERGY PRODUCTION IN FINLAND
Karoliina Hämäläinen
Institute for Atmospheric and Earth System Research/Physics Faculty of Science
University of Helsinki Helsinki, Finland
ACADEMIC DISSERTATIONin meteorology
To be presented, with the permission of the Faculty of Science of the University of Helsinki, for public criticism in CK112 auditorium at Exactum (Pietari Kalmin katu 5, Helsinki) on the 12th of November at noon.
Finnish Meteorological Institute Helsinki, 2021
Supervisors Dr. Sami Niemelä
Head of Meteorological and Marine Research Programme Finnish Meteorological Institute, Finland
Professor Heikki Järvinen
Institute for Atmospheric and Earth System Research University of Helsinki, Finland
Reviewers Associate Professor Petri Välisuo Department of Automation Technology University of Vaasa, Finland
Associate Professor Heiner Körnich Head of Meteorological research unit
Swedish Meteorological and Hydrological Institute, Sweden
Custos Professor Heikki Järvinen
Institute for Atmospheric and Earth System Research University of Helsinki, Finland
Opponent Professor Stefan Ivanell Department of Earth Sciences Uppsala Universitet, Sweden
ISBN 978-952-336-145-4 (paperback) ISBN 978-952-336-144-7 (pdf)
ISSN 0782-6117 Edita Prima Oy
http://ethesis.helsinki.fi/
Helsinki 2021
(Erik Palménin aukio 1) Finnish Meteorological Institute
PL 503, 00101 Helsinki Contributions 177, FMI-CONT-177
Päiväys Syyskuu 2021
Tekijä ORCID iD
Karoliina Hämäläinen 0000-0002-3449-7559
Nimeke
Meteorologiset sovellutukset tuulivoiman tukemiseksi Suomessa.
Tiivistelmä
Fossiilisten polttoaineiden korvaaminen uusiutuvilla energian lähteillä on keskeisessä asemassa, kun pyritään hillitsemään ilmaston muutosta. Tuuli- ja aurinkoenergia sekä vesivoima ovat kuitenkin hyvin sääriippuvaisia.
Sään vaihtelut on otettava huomioon, jotta uusiutuvien energianlähteiden päivittäistä tuotantoa pystytään ennustamaan tarkasti. Tässä väitöskirjatutkimuksessa on kehittetty menetelmiä ja sovelluksia, jotka tukevat tuu- livoiman tuotannon suunnittelua ja tuulivoiman tuotannon ennustettavuutta, sekä pyritty vastaamaan kysymyksiin:
Kuinka tarkkoja ovat tuuliennusteet tuulivoiman tuotannon kannalta tärkeillä korkeuksilla? Millainen on tuulen nopeuden vuotuinen jakauma? Paljonko tuulesta voidaan saada tehoa? Miten talviset jäätävät olosuhteet vaikuttavat tuulivoiman tuotantoon? Kuinka jäätäviä olosuhteita voidaan ennustaa?
Väitöskirjatyön ensimmäisessä osiossa on kehitetty resurssikarttoja, joita hyödynnetään tuulivoiman suunnittelussa sekä sopivan tuulipuiston sijainnin valinnassa. Nämä nk. Tuuli- ja Jäätämisatlakset on tuotettu käyttämällä hyväksi numeerisia säänennustusmalleja. Simulaatioita varten valittiin viimeisen 30 vuoden ajalta yhteensä 72 kuukautta, jotka edustavat Suomen tuuli-ilmastoa. Näiden kuukausien valintaan käytettiin kriteereinä keskituulta, tuulen suuntajakaumaa sekä niiden yhdistelmää. Jäätämisen osalta ei voitu suorittaa vastaavaa edustavien kuukausien valintaa kuten tuulelle, sillä jäätämishavaintoja ei ole riittävän kattavasti tarjolla. Tuulen mukaan valitut kuukaudet eivät välttämättä edusta jäätäviä olosuhteita, ja siksi valituille kuukausille suoritettiin herkkyyskokeita lämpötilan ja suhteellisen kosteuden osalta. Herkkyyskokeiden tulokset osoittivat, että valitut kuukaudet ovat riittävän edustavia arvioimaan myös jäätämistä. Atlasten tulokset on esitetty karttamuodossa 2,5 km hilassa ja ne kuvaavat keskimääräisiä tuuliolosuhteita, potentiaalista tehontuottoa, aktiivista sekä passiivista jäätämistä, sekä jään aiheuttamia keskimääräisiä tehohäviöitä 50 m, 100 m ja 200 m korkeuksilla.
Päivittäiset tuulen todennäköisyysennusteet tuovat lisäarvoa päätöksentekoon kuvaten lisäksi ennusteen epävarmuutta. Ennustettu todennäköisyysjakauma ei aina ole realistinen, mutta niitä voidaan pyrkiä parantamaan tilastollisilla korjausmenetelmillä. Tässä väitöskirjatyössä onkin tutkittu miten tuulen todennäköisyysennusteita pystyttäisiin parantamaan tilastollisin menetelmin sekä käyttäen apuna Doppler lidar ja Doppler tutka tuulihavaintoja, joita ei ole aiemmin sovellettu tähän tarkoitukseen. Tulokset osoittavat, että kehitetty kalibrointimenetelmä pystyy uusien havaintojen avulla parantamaan ennustetta etenkin heikkojen, kohtalaisten ja navakan tuulen osalta. Kovien ja myrskytuulien osalta menetelmän ei voitu osoittaa parantavan ennustetta osittain siitä syystä, että voimakastuulisia tapauksia ei osunut riittävästi tarkastelujaksolle.
Tuulen nopeuden lisäksi tuulivoiman tuotantoon vaikuttaa kylmässä ilmastossa voimakkaasti jään kertyminen tuuliturbiinin lapoihin. Jäätämisen mittaaminen on kuitenkin hyvin haastavaa. Tämän väitöskirjatyön viimeisessä osiossa tarkasteltiin kuinka ceilometrillä mitatuista ilmakehän pystyprofiileista johdettuja jäätämishavaintoja voitaisiin hyödyntää jäätämisennusteen verifioinnissa. Uusi havaintomenetelmä osoittautui hyvin lupaavaksi
etenkin havaintoverkon kattavuuden osalta (24 asemaa). Havaintovertailun pohjalta voidaan todeta, että jäätämisennusteen suurimmat virheet johtuvat numeerisesta sääennustemallista saadusta alkutiedosta eli jäätävän pilven paikallisesta ja ajallisesta sijainnista. Tämä uusi havaintomenetelmä tukee jatkosuunnitelmia pilviennusteiden parantamisen osalta sekä jäätämisen todennäköisyysennusteiden kehitystyötä.
Julkaisijayksikkö Meteorologinen tutkimus
Luokitus (UDK) Asiasanat
551.55, 551.582, 551.508, 551.509 tuuli, tuulivoima, tuuliennuste,
629.3.054.6 jäätäminen, ceilometri, Doppler
lidar, Doppler tutka
ISSN ja avainnimeke ISBN
0782-6117 978-952-336-145-4 (paperback)
Finnish Meteorological Institute Contributions 978-952-336-144-7 (pdf)
DOI Kieli Sivumäärä
https://doi.org/10.35614/isbn.9789523361447 Englanti 61
FIN-00101 Helsinki, Finland Contributions 177, FMI-CONT-177 Date
September 2021
Author ORCID iD
Karoliina Hämäläinen 0000-0002-3449-7559
Title
Meteorological solutions to support wind energy production in Finland.
Abstract
The renewable energy sources play a big role in mitigating the effects of power production on climate change.
However, many renewable energy sources are weather dependent, and accurate weather forecasts are needed to support energy production estimates. This dissertation work aims to develop meteorological solutions to support wind energy production, and to answer the following questions: How accurate are the wind forecasts at the wind turbine hub height? What is the annual distribution of the wind speed? How much energy can be harvested from the wind? How does the atmospheric icing affect wind energy production and how do we forecast these events?
The first part of this dissertation work concentrates on resource mapping. Wind and Icing Atlases bring valuable information when planning wind parks and where to locate new ones. The Atlases provide climatological information on mean wind speed, potential to generate wind power and atmospheric icing conditions in Finland. Based on mean wind speed and direction, altogether 72 representative months were simulated to represent the wind climatology of the past 30 years. A similar detailed selection could not be made with respect to icing process due to lack of icing observations. However, sensitivity tests were performed with respect to temperature and relative humidity, which have an influence on icing formation. According to these sensitivity tests the selected period was found to represent the icing climatology as well. The results are presented in gridded form with 2.5 km horizontal resolution and for 50 m, 100 m and 200 m heights above the ground, representing typical hub heights of a wind turbine.
Daily probabilistic wind forecasts can bring additional value to decision making to support wind energy production. Probabilistic weather forecasts not only provide wind forecasts but also give estimations related to forecast uncertainty. However, probabilistic wind forecasts are often underdispersive. In this thesis the statistical calibration methods combined with a new type of wind observations were utilized. The aim was to study if Lidar and Radar wind observations at 100 m’s height can be used for ensemble calibration. The results strongly indicate that the calibration enhances the forecast skill by enlarging the ensemble spread and by decreasing RMSE. The most significant improvements are identified with shorter lead times and with weak or moderate wind speeds. For the strongest winds no improvements are seen, as a result of small amount of strong wind speed cases during the calibration training period.
In addition to wind speed, wind power generation is mostly affected by atmospheric icing at Northern latitudes.
However, measuring of icing is difficult due to many reasons and, furthermore, not many observations are available.
Therefore, in this thesis the suitability of a new type of ceilometer-based icing profiles for atmospheric icing model validation have been tested. The results support the usage of this new type of ceilometer icing profiles for model verification. Furthermore, this new extensive observation network provides opportunities for deeper investigation of icing cloud properties and structure.
Publishing unit Meteorological Research
Classification (UDC) Keywords
551.55, 551.582, 551.508, 551.509 wind, wind energy, windpower,
629.3.054.6 atmospheric icing, ceilometer, Doppler
lidar, Doppler radar
ISSN and series title ISBN
0782-6117 978-952-336-145-4 (paperback)
Finnish Meteorological Institute Contributions 978-952-336-144-7 (pdf)
DOI Language Pages
https://doi.org/10.35614/isbn.9789523361447 English 61
Tomi
Matias
Ukko
I want to thank Bengt Tammelin, Timo Vihma, and Carl Fortelius who gave me my first real job as a meteorologist, as part of the Wind Atlas team back in 2008. I re- member my job interview. Bengt was eager to hear about my engineer background.
Timo was interested in my knowledge related to the boundary layer, which I had background only from university courses. Carl was sitting quietly and asked only one question: “Have you ever used the Batch system“. Well, I had not. Neverthe- less, this was the starting point of my research career. Yet, so little I know what this journey would bring along and whom I would get to know during these years.
I have always had the privilege to work with kind people, who have been will- ing to share their knowledge and wisdom. One of these people is my supervisor Sami Niemelä, to whom I want to give special thanks. You have always supported me, spared your time for my questions, and not only given straight answers but raised new questions, giving me space and time to grow as a researcher.
After the first two papers, I was struggling. However, I was lucky to meet Elena Saltikoff. You offered me, not only the radar data which I was looking for but tutorship and kind words which I so desperately needed. While working with you, you made me realize that I’m starting to become a real scientist. And thanks to Elena, I got to know Anne Hirsikko and Ewan O’Connor, with whom I enjoyed working during my fourth paper. I guess we never had a meeting without a burst of laughter. You all made my work feel so much easier with your positive attitude.
During my career, I’ve met so many nice and professional people both inside and outside FMI. I’ve always felt respected among my colleagues. I can only wish that I’ve been able to return even a small part of all that kindness and respect.
Warm thanks to all my former roommates who have shared their life’s with me.
With the same warm feeling, I want to thank all of you who have shared your lunch and coffee breaks with me. These relationships among colleagues have turned into many friendships.
I wish to thank the Vilho, Yrjö and Kalle Väisälä Foundation, from the personal grand awarded to finalize the Paper IV and writing of this thesis summary. With this support, I was able to get through the otherwise so unique pandemic year 2020. In addition, the work has been supported by other projects and funding agencies: MOTIVA, Academy of Finland, Business Finland (former Tekes), and MATINE.
I sincerely thank Stefan Ivanell for acting as my opponent. I thank my other supervisor and custos Heikki Järvinen, and not forgetting Petri Välisuo and Heiner Körnich for pre-examining this thesis.
Then there is my family. I want to thank my parents, Vuokko and Matti Ljung- berg, who have always allowed me to be who I’m, whether I wanted to be a
10
princess, an engineer, or a meteorologist. I also thank my siblings and friends for being there for me during rough times and rejoiced in my success. Then there are my children, Matias and Ukko, who havefilled my life with joy and given me a purpose to do what I do. Last but not least I want to thank my husband Tomi.
The man who has been supporting me, lift me up when I doubted and pushed me forward during this windy journey to doctorhood.
Karoliina Hämäläinen Espoo, September 2021
Preface 9
List of acronyms 12
List of original publications 15
1 Introduction 17
2 Background 21
2.1 Wind power resources . . . 21
2.2 Harvesting wind power . . . 21
2.3 Wind energy meteorology . . . 22
2.3.1 Wind forecasting . . . 23
2.3.2 Atmospheric icing on structures . . . 24
3 Data and methods 27 3.1 Observational data . . . 27
3.1.1 Ground-based remote sensing wind observations . . . 27
3.1.2 In-situ icing observations . . . 29
3.1.3 Ceilometer icing profiles . . . 30
3.2 Weather models and output data . . . 31
3.2.1 The model chain: from global re-analysis to local scale wind energy applications . . . 31
3.2.2 Global ensemble prediction: IFS-ENS . . . 34
3.2.3 High resolution limited area model: Harmonie-Arome . . . . 34
3.3 Downstream methods . . . 36
3.3.1 Atmospheric icing model . . . 36
3.3.2 Statistical calibration of wind speed forecasts . . . 37
4 Summary of the results 39 4.1 Resource maps . . . 39
4.2 Forecasting of wind energy-related weather parameters . . . 45
4.2.1 Statistical calibration of probabilistic 100 m wind forecasts . 45 4.2.2 Verification of icing profile forecasts by using ceilometer ob- servations. . . 48
5 Conclusions and future perspectives 51
References 55
12
LIST OF ACRONYMS
ABL Atmospheric Boundary Layer
AROME Application of Research to Operations at Mesoscale BCT Box-Cox T-distribution
BS Brier Score BSS Brier Skill Score
CORINE Coordination of Information on the Environment ECMWF European Center for Medium-Range Weather Forecasts EPS Ensemble Prediction System
FMI Finnish Meteorological Institute
GAMLSS General Additive Model for Location, Scale and Shape GDAS Global Data Assimilation System
GHG Greenhouse gas
GLAMEPS Grand Limited Area Ensemble Prediction System ICAO International Civil Aviation Organization IFS-ENS Integrated Forecast System - Ensemble System
Harmonie HIRLAM–ALADIN Research on Mesoscale Operational NWP in Euromed HARATU Harmonie with RACMO Turbulence
HIRLAM High Resolution Limited Area Model LWC Liquid Water Content
MEPS Meterological Cooperation on Operational Numeric Weather Prediction System
MetCoOp Meterological Cooperation on Operational Numeric Weather Prediction (NWP) between Finnish Meteorological Institute (FMI), MET Norway, Swedish Meteorological and Hydrological Institute (SMHI) and Estonian Environment Agency (ESTEA).
MSL Mean Sea Level
MVD Medium Volume Diameter
Nd Cloud Droplet Particle Concentration NWP Numerical Weather Prediction PC Proportion Correct
p.d.f. Probability Density Function POD Probability of Detection RMSE Root Mean Square Error SNR Signal-to-Noise Ratio SR Success Ratio TS Threat Score
VAD Velocity Azimuth Display VVP Velocity Volume Processing
WAsP Wind Atlas Analysis and Application Program WMO World Meteorological Organization
LIST OF ORIGINAL PUBLICATIONS AND AUTHOR’S CONTRIBUTION
I Tammelin, B., Vihma, T., Atlaskin, E., Badger, J., Fortelius, C., Gregow, H., Horttanainen, M., Hyvönen, R., Kilpinen, J., Latikka, J., Ljungberg, K., Mortensen, N. G., Niemelä, S., Ruosteenoja, K., Salonen, K., Suomi, I., Venäläinen, A., 2013: Production of the Finnish wind atlas. Wind Energy, 16(1), 19–35. doi: 10.1002/we.517
II Hämäläinen, K. and Niemelä, S., 2017: Production of a numerical icing atlas for Finland. Wind Energy,20(1), 171–189. doi:10.1002/we.1998
III Hämäläinen, K., Saltikoff, E., Hyvärinen, O., Vakkari, V., Niemelä, S., 2020:
Assessment of probabilistic wind forecasts at 100 m above ground level using Doppler Lidar and Weather Radar Wind Profiles. Monthly Weather Review, 148(3), 1321–1334. doi: 10.1175/MWR-D-19-0184.1
IV Hämäläinen, K., Hirsikko, A., Leskinen, A., Komppula, M., O’Connor, E.J., Niemelä, S., 2020: Evaluation of Atmospheric Icing Forecast with Ground- based Ceilometer Profile Observations. Meteorological Applications, 27(6), 1–12. doi: 10.1002/MET.1964
In PaperIKaroliina Hämäläinen (nee Ljungberg) was part of the research team responsible for planning and performing the model simulations. Later during the project, the author took part in planning and performing literature background study for the roughness length used in down-scaling and took part in writing Sec- tion 2.3. PaperIIwas a continuation to PaperI. In PaperIIthe author participated in all study phases from planning to publication, including developing and testing of the atmospheric icing model. In Papers IIIandIV, the author was responsible for acquiring data, performing model simulations, data analysis and visualization, and writing most of the text, excluding observation descriptions.
1 INTRODUCTION
Modern society is built on technology and transportation. Heating, cooling, light- ing and communication are considered to be basic human needs. The electricity consumption is rapidly growing, like in the form of electric cars, even though the technology has improved the energy efficiency in daily life. However, the climate change is driving governments to take steps towards cleaner and more sustainable societies (Paris Agreement 2015). Cleaner energy production is one of the most important parts of reducing the Greenhouse gas (GHG) emissions (IPCC 2011).
The renewable energy sources like hydropower, wind and solar energy are to re- place fossil fuels. At the moment wind energy is the fastest growing source of renewable energy in Europe (IRENA 2015), and the amount of installed wind tur- bine capacity is globally increasing (Joyce Lee and Feng Zhao 2019). In 2008 the Finnish government stated in their Energy strategy, that the amount of wind en- ergy should be raised from 0.26 TWh to 6 TWh before the year 2020 (Ministry of Employment and the Economy 2008). This goal was achieved during 2019, due to many actions taken by the government to support this aim.
The wind farm planning requires knowledge on prevailing mean wind condi- tions. These conditions determine the optimal wind farm location, turbine type and how the turbines should be positioned in respect to each other. To help in this decision making many wind atlases have been generated (Global Wind At- las 2019; NEWA 2018; Tammelin et al. 2013) during the past ten years. These atlases contain maps of wind conditions, which provide a good first estimate of where to locate new wind farms. However, in cold climate regions wind speed is not the only factor determining the best wind farm locations. In addition, the atmospheric icing conditions need to be taken into account. Ice accumulated on wind turbine blades reduces the efficiency of wind energy production (Davis et al.
2016). Therefore, maps of atmospheric icing have been generated too (Hämäläi- nen and Niemelä 2017; Kjeller Vindteknikk 2020).
Finland’s electricity network is part of Nordpool energy markets, where the electricity producers and consumers can sell and buy electricity in day-ahead mar- kets (Nordpool 2020a). The bids are placed 24 hours ahead in closed auction. If the producers’ production goals are not reached the energy must be bought some- where else during intraday markets (Nordpool 2020b). The difference between the promised and the actual price is set as penalty to the producer (Jónsson et al.
2010). These market rules set the foundation for the predictability of electricity production and increases the reliability of the total electricity network (Holttinen et al. 2011; Miettinen and Holttinen 2018). For the weather-dependent renewable energy sources, such as wind energy, these market rules cause an extra uncertainty.
Accurate weather forecasts can help in providing support to the daily and hourly
18
wind energy production to increase the reliability of the total power network.
Traditional weather forecasts, so-called deterministic forecasts, provide a sin- gle realization of the upcoming weather situation. These forecasts are based on numerical weather prediction (NWP) model runs initiated from the best available estimate of the current atmospheric conditions. To create the initial conditions ob- servations are blended in together with the most recent forecastfields. However, initial conditions can be estimated with limited accuracy, leading to initial con- dition error. The uncertainty within the initial conditions can be described with a probability density function (p.d.f.) (Leutbecher and Palmer 2008). In this ap- proach multiple initial conditions are evolved as functions of forecast lead time, leading to a wider set of outcomes. This is called an ensemble forecast. The in- dividual elements of the ensemble are called ensemble members. In addition to initial errors the model dynamics and physics introduces other sources of forecast errors. Thesemodel errorsarise mostly from the computational limitations, like: i) truncation error,ii)how well sub-grid-scale physical processes are parametrized, andiii)how well the interaction with the atmosphere and land/ocean surface are being described. Even if initial errors and model errors are treated separately it is important to remember that in reality they are not separable. The initial con- ditions always include the forecast model, and the weather forecast model carries out the initial uncertainties throughout the whole model simulation (Fig. 1a).
Figure 1a) Schematic of the ensemble forecast. The gray curve indicates deter- ministic model simulation. The ensemble members are presented with light pink curves. The initial states will evolve as functions of forecast time and the out- comes are affected by the model physics and dynamics and their uncertainties. b) The probabilities can be calculated for each forecast lead time from the ensemble bloom. ©Karoliina Hämäläinen.
The skill and usefulness of the ensemble prediction system is described with reliabilityandresolution. The ensemble forecast is considered reliable when most of the ensemble members populate towards the observed value, leading to smaller ensemble mean RMSE. The resolution covers the width of the predicted distribu-
tion and is called ensemble spread. The spread contains the information related to forecast uncertainty (Buizza 2018). A narrow spread is considered more useful because it indicates more clearly whether or not the event of interest is likely to happen. Hereby, the ensemble forecasts can be used to estimate the probability of a certain weather situation, which for some users provide added value (Mylne 2002; Pinson et al. 2006) beyond a single model realization. For instance, most of the modern wind turbines can operate with wind speeds between 4 - 25 m s−1. A wind energy producer would benefit from the information what the probability is of exceeding a certain threshold value like cut-off wind speed (25 m s−1), when the power production drops suddenly from full production to zero production causing possiblefinancial loss (Fig. 1b).
The reliability and resolution of the ensemble forecast can be improved with statistical calibration. Statistical calibration methods have been developed to min- imize the bias in the ensemble mean and enhance the reliability of the ensemble forecast. Their use requires historical data from both the forecast and from the observations. From the wind energy point of view this is problematic. There are very few observations available from the levels where the wind turbines operate;
i.e. wind turbine hub height. Ground-based remote sensing observations (Doppler Radar or Lidar) may become useful for this purpose. They have been proven to correlate well with mast-based wind measurements (Newsom et al. 2017; Päschke et al. 2015).
Ice formation on wind turbine blades reduces the efficiency of power produc- tion. Therefore, atmospheric icing poses another demand for wind energy produc- tion forecasts. One of the limitations in developing forecasts is the poor availability of proper icing observations. Therefore, a new type of vertical profiling ceilome- ter observations are expected to bring long-awaited relief to this problem. The ceilometer observations provide wider geographical coverage and finer vertical resolution at observing sites compared to traditional mast measurements.
The focus of this thesis is to provide new meteorological solutions to support the wind energyfield. In this thesis the research questions are drawn to three main topics:
1. Resource mapping is important information for the wind power devel- opers. Which areas in Finland are the most promising for producing wind energy?
InPapers IandIIthe aim is to derive the climatology of mean wind speed, wind power generation and atmospheric icing conditions for Finland. In these studies the Numerical Weather Prediction (NWP) models are used to downscale the re-analysis data from the European Center for Medium-Range
20
Weather Forecasts (ECMWF). Wind climatology representing a 30-year pe- riod is reproduced by carefully selecting 6 representative years. Paper I presents the methodology of how the mean wind speed and wind power es- timates are calculated. The same selection of years were later used to create a climatology for atmospheric icing, inPaper II. All the results are presented in gridded form with 2.5 km horizontal resolution and for 50 m, 100 m and 200 m elevation above the ground, representing typical hub heights of a wind turbine.
2. The probabilistic weather forecasts provide important information for the wind energy producers, and also give an estimation about the fore- cast uncertainty. How much could these ensemble wind forecasts be fur- ther improved by using statistical calibration methods combined with a new type of ground-based remote wind observations?
In Paper III the statistical post-processing model together with the new type of ground-based remote wind observations are used to further im- prove the wind forecasts of the Integrated Forecast System - Ensemble Sys- tem (IFS-ENS), produced by ECMWF. The aim is to study if Lidar and Radar wind observations can be used for the calibration of ensemble model wind forecasts. The results highly support the actions taken towards the usage of these new types of ground-based remote sensing wind observations.
3. Verifying forecasts of atmospheric icing is difficult due to a lack of in-situ icing observations. Can the new type of profiling ceilometer observa- tions bring a long needed help to solve this problem?
InPaper IVthe new type of ceilometer-based icing profiles are used to val- idate icing forecasts provided by the atmospheric icing model. The results support the operational usage of the new type of icing observations in veri- fication, and open new possibilities to further icing-model development.
This thesis is structured as follows: the background of this thesis is described in more detail in Section 2. Section 3 briefly describes the observations, models and downstream methods that have been used in this thesis. Section 4 is ded- icated to the main results of this thesis. Finally, the discussion and conclusions are summarized in Section 5. The original papers are re-printed at the end of this thesis.
2 BACKGROUND
2.1 WIND POWER RESOURCES
Finland is located at the norther part of mid-latitudes where the westerly winds are influential. The daily low pressure activity affects prevailing wind conditions. The mean wind direction is south-west, but summertime high-pressure systems over the Russian continent can change the wind direction to south-east. In addition, surface type, topography and land-sea interactions affect local windfields.
The Finnish landscape is horizontally very heterogeneous. The land is fairlyflat with the exception of the northern fjeld areas. Forested and urban areas, fields, thousands of lakes and small islands generate: i) barriers, ii) turbulence, and iii)cause wind channeling in the lower atmosphere. Turbulence is caused by the friction between air flow and surface. This friction depends on surface charac- teristics, such as vegetation or town type and height, which are described with surface roughness length (Schmid and Bünzli 1995; Vihma and Savijärvi 1991).
The surface roughness varies between the seasons, when swayingfields turn into flat snow-covered terrain, and open water areas become ice-covered. The lower part of the atmosphere that senses the surface roughness is called Atmospheric Boundary Layer (ABL) (Garratt 1994). The height of the ABL varies from a few tens of meters to kilometers, depending on the stability of the atmosphere. Above ABL the wind conditions are more laminar and less turbulent, as the surface no longer affects the windflow. The current wind turbine hub heights vary between 80 m and 160 m, with blade length from 40 m to 80 m (IEA 2019b). This means that the rotor is still within the ABL. Hence, the wind direction can play a big role on wind. The wind conditions and turbulence can change a lot if for example the wind direction changes from sea to land. Therefore, despite the higher investment costs, the off-shore wind parks have grown their popularity during the past years due to steadier wind conditions (IEA 2019a).
Altering meteorological conditions can increase the wind energy production in Finland, like: i)frontal systems approaching from a favourable direction (de- pends on wind park geometry), ii) winter-time clear sky situations with no risk of atmospheric icing,iii)weak winds due to stable stratification during cold win- ter periods or night time ,iv)land or sea breeze, andv)wind channeling due to topography and wind direction.
2.2 HARVESTING WIND POWER
The modern wind turbines can operate when the wind speed is 4 - 25 m s−1. The weaker winds are too weak to rotate the blades effectively. However, winds
22
higher than 25 m s−1 are harmful for the structures of the wind turbine due to increasing vibration. The probability of unwanted wind gusts also increases with higher wind speeds. It is rarely too calm for wind power generation, but to avoid structural damage the turbine blades are locked and power production shut down during stormy weather.
The amount of wind powerPretrieved from the wind can be calculated with Equation 2.1 (Burton et al. 2001):
P= 1
2cpρai rARV3 (2.1)
where cp is a wind turbine-dependent dimensionless power coefficient, which depends on blade aerodynamic efficiencyηb, mechanical efficiencyηm and elec- trical efficiencyηe, as presented in Equation 2.2.ρai ris the air density. The colder air is more dense resulting in better power generation. ARdescribes the rotor area swept by the horizontal windV seen by the blades. The rotor areaARdepends on the length of the bladesR(Eq. 2.3).
cp=ηbηmηe (2.2)
AR=πR2 (2.3)
Even though wind power is proportional to the cube of the horizontal wind, more wind power can be generatedi)with longer blades, andii)having the turbine hub at a higher level from the ground, experiencing stronger winds.
2.3 WIND ENERGY METEOROLOGY
As wind energy production is highly dependent on prevailing weather conditions, accurate weather forecasts are needed. Firstly, the meteorological knowledge is essential in the planning phase. The wind park investments are expensive and many questions need answers, like: i)what are the average annual wind condi- tions,ii)what is the dominant wind direction, and iii)is it necessary to invest in icing detection or anti-icing equipment? Hence, during the past ten years many wind (Global Wind Atlas 2019; NEWA 2018) and icing (Kjeller Vindteknikk 2009) atlases have been generated to support the decision making. These atlases do not only support the actors in the wind energy sector, but also act as guidance for land use and community planning.
Thefirst wind atlases were based on wind observations (Troen and Lundtang Petersen 1989; Watson and Landberg 1999). Many times most of the observa- tions were located inland and only a few in coastal areas and mountains. This was
also the situation during the previous wind atlas for Finland (Tammelin 1991).
This type of limitations on observations in areas most suitable for wind energy production generates real issues with the representativeness of the whole atlas.
Nowadays the modern wind atlases withfine resolution output of a few kilome- ters are based on numerical weather prediction models combined with wind ob- servations (Global Wind Atlas 2019; Bergström and Söderberg 2008; Kjeller Vin- dteknikk 2009; NEWA 2018). Numerical atlases provide a good first estimate on long-term (yearly/monthly) variation of the wind conditions during the wind park planning phase. However, local detailed wind conditions depending on the local landscape, can not be modelled accurately with NWP models. Such models have their own limitations with horizontal resolution and surface parametrization.
Hence, once the park location has been chosen, a testing campaign with meteo- rological measurements should take place. The measurement campaign usually takes from one to two years. During this time the wind (and possibly icing) con- ditions are investigated. The testing is often performed utilizing high resolution lidar measurement. The limitation of these detailed measurement campaigns is the interannual variability in wind conditions. Some years are more windy than others. That is why the campaign measurements are compared to long-term wind observations from the nearest mast. The comparison is needed to achieve more ac- curate statistics of average wind conditions on site. Depending on the results, the final decision on investment is made. The results help to choose the right turbine type and additional equipment like icing detectors.
Once the wind park is up and operating the need for short-term (daily/hourly) wind and icing forecasts rises. The following subsections will focus on wind fore- casting and why atmospheric icing needs to be taken into account.
2.3.1 WIND FORECASTING
Wind forecasts support the wind power industry in many time horizons. For build- ing and maintenance, wind forecasts for time horizons from one day to two weeks are needed. Daily wind forecasts are used in the energy markets, when making estimations of power production for the next day in the so-called day-ahead mar- ket (Jónsson et al. 2010). Finally, due to the very variable nature of the wind, the hourly and sub-hourly forecasts are needed to support the balancing of the energy markets and electricity network, and for the effective control of the wind park. If the forecast fails in any of these situations, it can lead tofinancial loss or even risks for human health.
In the numerical weather prediction (NWP) models the windfield is split in two horizontal components. The eastward component is often described with U and the northward component withV. These components are part of the motion
24
vectors. In many forecasting models, like in global IFS-ENS or in limited area Harmonie-AROME, the motion vectors are described by using hydrostatic primi- tive equations (ECMWF 2017) or nonhydrostatic Euler-equations (Bengtsson et al.
2017; Bubnová et al. 1995), respectively. These equations are solved in time and space with semi-implicit two-time-level time integration and semi-Lagrangian ad- vection scheme (Bénard et al. 2010; ECMWF 2017). Eventually, wind speed and direction are derived from these two wind components, for all the grid points and model levels. In the models subgrid scale features are parametrized, as: i)turbu- lence, ii)radiation,iii) cloud microphysics,iv)surface interactions, and v)con- vection (in IFS-ENS) or shallow convection (in Harmonie-AROME). The accuracy of the wind forecast depends on the model’s horizontal and vertical resolution, and how well the interactions between the atmosphere and surface is parametrized.
Since the 1990s the Ensemble Prediction Systems (EPSs) have become more commonly used in weather prediction. The advantage of probabilistic forecasts is that they provide the information related to forecast uncertainty, by generating a range of possible future events (Buizza 2018). The spread of the ensemble forecast tells us how well the coming weather event can be forecasted. The narrow spread in the forecast distribution is an indication that the predictability of the weather event is good. On the other hand, a large spread in forecast distribution indicates larger uncertainty related to the forecast. In an ideal situation the ensemble spread should be as wide as the Root Mean Square Error (RMSE) of the ensemble forecast, to cover the realistic uncertainty range. Thus, the probabilistic wind forecasts can bring additional value for the daily wind energy production estimates (Pinson et al. 2006). However, like traditional deterministic forecasts also probabilistic wind forecasts can be biased, leading to under- or overdispersive probabilistic forecasts.
The verification of the wind forecasts is needed to follow up the model per- formance. However, the standardized SYNOP observations by the World Mete- orological Organization (WMO) are measured at 10 m’s height. From the wind energy point of view these observations do not support wind verification at the height of interest (∼100 m). Therefore, new types of observations are needed to perform the wind energy-related wind verification. In section 3.1.1, the new type of wind observations at 100 m’s height are presented. Furthermore, Section 3.3.2 describes how these new observations can be used to improve the probabilistic wind forecasts by utilizing statistical calibration methods.
2.3.2 ATMOSPHERIC ICING ON STRUCTURES
Forecasting wind energy production in cold environments requires information on potential icing conditions. Atmospheric icing can occur in many ways. Super- cooled liquid rain drops can cause very hard ice cover on structures (Fig. 2a). How-
ever, such icing events are rare in Finland (Kämäräinen et al. 2017). In this study we focus on a more common event, rime ice formation from tiny cloud droplets (Fig. 2b). Rime ice causesfinancial loss to wind energy production (Davis et al.
2016), risk for aviation (Bernabò et al. 2015) and extra weight on power lines, which can exceed the structural limits of the transmission towers (Nygaard et al.
2016). In cold climate regions like Finland the weather conditions favourable for rime ice formation extend from late September until the beginning of May (Hämäläinen and Niemelä 2017). Furthermore, during this same period the en- ergy consumption is largest mainly due to heating, which forms a particular un- certainty for energy systems with a large share of wind energy production.
Figure 2Atmospheric icing. a) Ice formed by freezing rain. b) Rime ice formation by cloud droplets. ©Pixabay.
The rime ice is formed when liquid cloud droplets are in contact with a cold surface. Compared to rain droplets the cloud droplets are tiny. A typical rain droplet has a radius of 1000 µm as cloud droplets have a radius of 10-50 µm (Aguado and Burt 2010). Furthermore, the cloud droplets are light enough to float in the air as fog for several hours. Due to the relatively small size of the cloud droplets the rime ice resembles layered snowflakes more than a hard ice surface (Fig. 2).
The probability of icing increases during weather conditions with fog or low clouds when the temperature is close to zero celsius. However, Westbrook and Illingworth (2011) have proven in their study that liquid cloud droplets can occur in temperatures as low as -37 °C, even the amount of liquid droplets rapidly de- creases at temperatures below -20 °C. The wind drives the cloud droplets against the objects and structures. Thus, the rime ice is accumulated on the wind side of the object. For wind turbines this means the leading edge of the turbine blade. The smallest droplets are light enough to go around the object following the stream-
26
line of the wind. However, due to inertia the larger cloud droplets collide with the object (Fig. 3).
Figure 3 Wind driving larger cloud droplets to collide and form ice over cylin- der shaped object. Smaller particles follow the streamlines of the wind avoiding impact. ©Karoliina Hämäläinen.
Makkonen (1984) was one of the first ones to study the rime ice accretion over structures. These studies concentrated on power line ice accretion and many modern icing models are based on these studies (Makkonen 1998; Makkonen and Lozowski 2008). The same cylinder shape from wires can be considered as an ap- proximation of a leading edge of the wind turbine blade, where most of the ice is accumulated. The accumulated ice over wind turbine blade decreases the aerody- namic properties of the blade, causing a decrease in power production (Davis et al.
2016). For effective aerodynamics, the blades should be as smooth as possible and clean from ice and dirt and retain its optimized shape.
A threshold value of 10 g h−1 for the modelled icing intensity is often used (Hämäläinen and Niemelä 2017; Kjeller Vindteknikk 2020) to distinguish between icing and non-icing conditions. This accumulation speed can fast generate a light ice cover. Furthermore, a very light ice cover (equivalent to sandpaper rough- ness) can already decrease the wind power production with 15 - 20% (Turkia et al. 2013). In moderate icing conditions the shape of the blade starts to transform due to ice cover and the power loss can be 24 - 29%. The accumulated ice load increases the vibration of the turbine structures and in harsh icing conditions the wind turbine must be stopped to prevent structural damage.
3 DATA AND METHODS
In this thesis, a wide range of observational and model data were used to develop downstream methods to improve forecasts of 100 m wind speed and atmospheric icing. A key to developing new methods is to have reliable observations. However, many conventional meteorological observations are located close to ground and do not support development of wind or atmospheric icing forecasts higher up. A new type of ground-based profiling ceilometers, Doppler radar and Doppler lidar measurements were used for wind forecast calibration and to identify atmospheric icing layers. These specific observation types, numerical weather prediction mod- els and downstream methods will be described in this chapter in more detail.
3.1 OBSERVATIONAL DATA
3.1.1 GROUND-BASED REMOTE SENSING WIND OBSERVATIONS
A new type of ground-based wind observations were used to overcome the lack of high masts and hence the lack of wind observations from higher up from the ground. In Paper III, the horizontal 100 m wind fields obtained from Doppler Radar and Doppler Lidar were utilized.
The Wind component parallel to the Doppler radar beam is calledradial wind.
The radar measurement resolution is temporally and spatially high. The Finnish Meteorological Institute (FMI) has 10 C-band Doppler weather radars. These radars measure vertical profiles of the wind in two ways. The Velocity Volume Pro- cessing (VVP) method (Waldteufel and Corbin 1979) is considered more robust and stable compared to the Velocity Azimuth Display (VAD) method (Browning and Wexler 1968). Hence, the VVP profiles were used. The VVP profiles include several (8) elevation scans in given volume, with nominal distance of 2 - 40 km from the radar. It is important to remember that radar measurement is not a point measurement. The radar backscatter is collected from a cylinder-shaped slice in the atmosphere (Fig. 4b). Radars measure a lot of data and cover entire Finland, but for 100m wind calculations we can use only a fraction of the measured data, due to geometrical limitations. Far from radar, one degree wide radar beam gets so wide that there is wind shear inside of a pixel. Also farther from the radar round surface of the earth affects on measurement height.
In this study the three lowest elevations (marked with pink shading, Fig. 4b) were used with a radius smaller than 20 km around the radar, in slices of 200 m above Mean Sea Level (MSL). Thus, the measured 100 m wind data actually in- cludes information from a 200 m thick layer, 360◦ and 20 km around the radar.
This leads to over 80 000 pixels within this radar volume. To save computing time,
28
each radar wind observation is based on limited dataset from 3500 - 5000 pixels.
The pixel data was processed to mimic point observations of mean wind speed and direction, by using the Fourier analysis. The wind speed was identified as the amplitude of thefirst Fourier component byfitting a sine curve to dataset collected around the radar. Respectively, the wind direction was identified from the phase of thefirst Fourier component.
Originally the Doppler radar data used in this study covered a two-year period (2016 - 2017). However, after quality control, part of the data was disregarded.
The following aspects were taken into account during quality control: if i)the number of valid data pixels is less than 100, andii)the standard deviation within data pixels is larger than 0.9 times the wind speed value, then the data pixels were neglected. The purpose for this was to exclude the outliers from the data, such as backscatter from birds, airplanes, wind turbines, or trees sawing in the wind.
Furthermore, C-band weather radars cannot measure air movement directly, but they detect some particles moving in the air. Therefore, no observations were available in dry or clean air situations. Hence, month-to-month differences in the amount of data available were large between station points. Eventually June and July 2016 were chosen for the verification and calibration periods and all 10 radars were used.
b)
Figure 4 Principles of (a) VAD scan with constant elevation angle [a], and (b) VVP scan with 8 elevation angles where blue lines indicate center of the beam and shading nominal 1◦beamwidth, from which the backscatter data is collected.
Left figure ©Karoliina Hämäläinen and right figure from Paper III ©American Meteorological Society.
Lidar windswere based on Halo Photonics Streamline scanning Doppler lidar measurements available from 4 stations. The VAD scans were performed to obtain horizontal 100 m wind speed and direction. During the VAD scan (Browning and Wexler 1968) the laser beam is emitted with certain elevation angle defined in advance (Fig. 4a). The beam "draws" a circle at given level (100 m) and the mean radial velocity of particles in the air is obtained as a function of azimuth. By using the Fourier analysis the horizontal wind speed and direction, at lidar location, are derived from the amplitude and the phase of the sine function of azimuth.
The Doppler lidar measurements were available from 1 Jan. - 10 Oct. 2016.
The lidar measurements were post-processed to minimize the Signal-to-Noise Ra- tio (SNR) using an algorithm by Vakkari et al. 2019. The SNR threshold value of 0.005 was applied to the data. Unfortunately, a large part of the data was missing in July and in September leading to too few observations for calibration and ver- ification purposes. Therefore, eventually, February and March 2016 were chosen as calibration and verification periods for 100 m wind speed forecasts.
3.1.2 IN-SITU ICING OBSERVATIONS
Detecting atmospheric icing is difficult. The commercial instruments available have three types of outputs from these instruments: i) ON-OFF type of signal, ii)icing rate identified from changes in measured signal, and/oriii)ice mass es- timates. Nowadays FMI has icing detectors at every airport in Finland. However, these instruments are installed at 2 m high and used for runway monitoring in case of icing. The only icing instruments installed higher up from the ground are located in Savilahti and Vehmasmäki.
The ON-OFF or icing rate measurements are based on vibration technique.
Like in ice detector by Labkotec (Labkotec 2020), there is a thin vibrating wire around the instrument body, and the vibration frequency changes when the ice is accumulated over the wire. In a similar way, the Rosemount/Goodrich (Camp- bell Scientific 2020) detector has a small vibrating pin, and the signal frequency changes when ice is accumulated. Many times these types of vibrating instruments are equipped with an automated heating cycle. When the signal frequency has changed to below a certain threshold the instrument heats itself to get rid of the ice. After this, the new measurement cycle can start. However, problems occur if the instrument gets dirty or the supporting structures (which are not heated) accumulate too much ice eventually covering the whole instrument with ice.
The ON-OFF type of measurements can also be obtained by combining visibility measurements with temperature measurements. By the definition of International Civil Aviation Organization (ICAO 2013), the visibility inside a fog is less than 1000 m. Moreover, if the temperature is under 0 °C and the visibility is below 1000 m
30
icing conditions can occur.
The IceMonitor by Combitech (Combitech 2020) is based on weighing of the ice load. In addition to ice mass measurement, the icing rate can be identified from the ice mass time series by differentiating the two subsequent ice mass mea- surements. This instrument is by far the only one to give information also from melting. IceMonitor is not so sensitive to prevailing harsh weather conditions, but it can also get over-iced. Therefore, this instrument requires occasional manual ice removal.
For many of the instruments and measuring techniques mentioned above, the video monitoring is recommended for quality control purposes.
InPaper II, the Vaisala FD12P visibility sensor was used to detect icing cases at Puijo station. The data was available from January to April 2010. For the same time period,i)Labkotec LID-3210D data was available from Puijo, andii) Labkotec LID-3310IP data was available from Riutunkari. From Luosto station short individual time series measured with Rosemount and IceMonitor instruments were available during wintertimes from 2005 to 2008. These time series were used (Paper II) to validate the atmospheric icing model described in section 3.3.1.
3.1.3 CEILOMETER ICING PROFILES
A new approach by using ground-based ceilometer attenuated backscatter profiles was taken to detect atmospheric icing. The ceilometer measurements are based on laser technique. The ceilometer emits a laser beam into the atmosphere and receives a backscattered pulse. The measurement is possible only if aerosols, cloud droplets or precipitation exists, enabling backscatter. The laser pulse attenuates in clouds or due to cloud droplets or precipitation, and no signal is available above thick clouds. However, from the shape and magnitude of the backscatter profiles structures of cloud can be identified, such as: liquid clouds, ice clouds and pre- cipitation (Tuononen et al. 2019). The data is available with 10-30 m vertical resolution and with 30s temporal resolution, depending on the instrument.
In Paper IV, the two-step method detecting atmospheric icing based on ceilometer backscatter profiles is described shortly. Firstly, the cloud layers con- taining liquid water droplets are identified. Secondly, based on the wet bulb tem- perature at the cloud base, the layer can be classified as warm or supercooled.
In this study the independent NWP dataset from Global Data Assimilation Sys- tem (GDAS) was used to calculate the wet bulb temperature profiles. The method to identify icing layers has been validated against in-situ icing measurements from two mast locations at Savilahti-Puijo and Vehmasmäki as described with more de- tails in Hirsikko et al. (2019). Also inPaper IV, a series of sensitivity tests were performed to determine the best way to conduct the model to observation compar-
ison. As a result, for each height the ceilometer observation was marked to detect icing, if 30% of time during one-hour time window observations indicated icing conditions. Finally, these ceilometer-based icing profiles were used to validate icing model forecasts during the test period (winter 2016 - 2017). The profiles were measured by using Vaisala CL51, CL31 and CT25K ceilometer instruments, from six stations altogether: Utö, Kumpula, Hyytiälä, Vehmasmäki, Savilahti and Sodankylä. Since the study period, the FMI’s profiling ceilometer network has grown to 24 stations. These profiling ceilometer icing observations provide better spatial resolution, both horizontally and vertically, compared to available in-situ measurements (Hirsikko et al. 2019).
3.2 WEATHER MODELS AND OUTPUT DATA
This section describes all the different models and re-analysis data used in this thesis. The reason for such a wide range of data sources arises from the needs of targeted applications and their end users. Re-analysis data is a good starting point when making climatological estimations of wind and icing conditions. To support the daily operational activities in wind energy production we need nu- merical weather forecasts. Furthermore, the availability of such forecasts and the computing capacity defines what kind of forecasts can be used. For instance, run- ning and using of probabilistic forecasts require a lot computing power and data storage.
3.2.1 THE MODEL CHAIN: FROM GLOBAL RE-ANALYSIS TO LOCAL SCALE WIND ENERGY APPLICATIONS
The model chain for creating useful products suitable for wind energy resource estimations includes several components. The backbone of the chain is the global re-analysis products, which will provide a long-term consistent timeseries of mul- tiple weather parameters. High resolution NWP models will provide better spatial and temporal resolution, but they require computationally costly simulations. Fi- nally, local scale models and dedicated process models provide the most detailed information depending on application.
REPRESENTATIVE YEARS FOR WIND ATLAS
The Finnish Wind Atlas was produced and based on (Paper I) wind climatology, which represents long-term wind conditions during the past 50 years. However, global re-analysis provides only coarse horizontal resolution, which is not suitable
32
Table 1Representative months selected based on ERA-Interim reanalysis.
Month Representative months Most Most for 1989-2007 windy calm January 1991 1993 2000 2007 1989 2004 February 1989 1992 1998 2006 1989 1994
March 1991 1994 2002 2006 1997 2006
April 2000 2003 2005 2005 2007 2004
May 1991 1996 2000 2005 2000 1994
June 1989 1991 1992 1994 2000 1997
July 1992 2000 2002 2006 1999 1997
August 1994 1997 2001 2007 2005 2006
September 1991 1996 2003 2006 2005 1993 October 1995 1997 1998 1999 2005 1992 November 1992 1997 2004 2005 1999 2002 December 1989 1990 2000 2002 1992 2000
for wind energy resource estimations. Furthermore, simulating weather by us- ing high-resolution NWP models for many decades is computationally very costly.
Hence a set of representative years was chosen to cover the wind climatology of past decades in order to reduce the computational burden and to make simulations more feasible. The ERA-Interim reanalysis data by ECMWF was available for 1989 - 2007 (Berrisford et al. 2011). A subset of individual months was selected in such a way that they would represent the large scale wind climatology of the whole ERA-Interim period. Four years altogether were selected. The selection was based on winds at 850 hPa pressure level, with the following criteria: i)Weibull distribu- tion of wind speed,ii)wind direction distribution, andiii)Weibull distribution of wind speed, to represent all 12 wind direction sectors as well as possible, for the period 1989-2007. In addition,i)the 12 most windy months andii)the 12 most calm months, were chosen to describe the climatological variability between the years. In this way, the high-resolution computational burden was reduced from 19 years down to 6 years (Table 1).
REPRESENTATIVE YEARS FOR ICING ATLAS
Due to a lack of long-term icing observations, a similar selection was not possible for icing. The idea behind the new Icing Atlas was to complement the Wind Atlas results, and therefore, the same set of representative years was used for the Finnish
Icing Atlas, as described inPaper II. However, the period selected based on wind climatology is not necessarily representative in respect to icing climatology. There- fore, the representativeness was investigated by performing a sensitivity test, for temperature and relative humidity at 850 hPa height. These two parameters give an indication of possible icing conditions. Mean temperature and relative humid- ity fields were calculatedfirst for representative months separately, and then for the whole period (1989-2007) during which ERA-Interim was available. Finally, mean fields between the selected months and long-term period were compared (see section 4.1).
GENERATING ATLASES
Wind and icing atlases were produced by downscaling global re-analysis data in three steps, as described inPaper I. In thefirst stage, the ERA-Interim reanalysis data was utilized as boundary information for regional High Resolution Limited Area Model (HIRLAM) (Undén et al. 2002). During the time of the study, the HIRLAM was the main operational synoptic-scale NWP model used at FMI. The ERA-Interim reanalysis has a horizontal grid-size of 80 km, whereas HIRLAM has a 7.5 km grid-size. The actual initial state for each HIRLAM run was generated with a variational data assimilation scheme for upper air and with optimum inter- polation for surface. The data assimilation process used a variety of conventional surface and upper-air observations such as: 2 m temperature and humidity, sur- face pressure, sea surface temperature, snow-depth, TEMP/PILOT soundings and AMDAR aircraft observations as input.
The second stage was to downscale the regional 7.5 km model products to the scales more feasible for wind energy applications. The meso-scale model Harmonie-AROME, with 2.5 km horizontal resolution (see more detailed descrip- tion in section 3.2.3), was nested inside the HIRLAM-model. The HIRLAM output was used as initial and boundary conditions for the Harmonie-AROME. With the Harmonie-AROME, short forecasts with +6 h forecast length were run with 3- hourly output. The 3-hourly output was considered adequate enough to describe the diurnal cycle of the wind. From the Harmonie-AROME output, the relevant pa- rameters were linearly interpolated in vertical on 8 height levels from the ground:
50 m, 75 m, 100 m, 125 m, 150 m, 200 m, 300 m and 400 m, which were directly used in the Wind Atlas.
In thefinal stage, very high resolution products were produced for the most rel- evant locations with respect to wind energy production by using a diagnostic down- scaling method WAsP (Wind Atlas Analysis and Application Program, Mortensen et al. 2011) to better describe the local wind fields. Firstly, the generalization process to remove model-dependent effects of surface roughness length and orog-
34
raphy from Harmonie-AROME output was performed. Secondly, the generalized wind climate was generated in the form of LIB-files (Mortensen et al. 2011). Fi- nally, WAsP with CORINE data (Coordination of Information on the Environment:
Büttner et al. 2004; CLC2000 2005) were applied on these cleaned wind fields in order to take into account the local scale influence on the wind. The CORINE data files cover more detailed (25 m resolution) information of land use types and roughness length, compared to NWP models. Thefinal wind speed results for these selected areas are presented with 250 m horizontal resolution for two levels above the ground (50 m and 100 m).
To create the Icing Atlas, wind speed, temperature and Liquid Water Content (LWC) were taken from Harmonie-AROME simulations and interpolated to three height levels (50 m, 100 m, 200 m) and further used as input for the atmospheric icing model. The icing model is described in more detail in section 3.3.1.
3.2.2 GLOBAL ENSEMBLE PREDICTION: IFS-ENS
The IFS-ENS is the operational ensemble prediction system of the ECMWF (ECMWF 2020). The ensemble includes 51 members and the forecasts are pro- duced twice per day with hourly output, up to 15 days. The horizontal resolution of IFS-ENS members is ∼15 km in Northern-Europe, and the ensemble has 91 vertical levels. The full documentation of IFS-ENS is available at ECMWF (2020).
InPaper III, the IFS-ENS data was extracted for Northern Europe. Only wind fields at 100 m’s height were used representing the wind conditions at wind turbine hub heights. The 100 m wind is one of the output parameters available from IFS-ENS. For the model calibration purpose, 15 days long forecasts were retrieved with 3-hourly output for the study period 2016 - 2017.
3.2.3 HIGH RESOLUTION LIMITED AREA MODEL: HARMONIE-AROME The Harmonie-AROME (Harmonie = HIRLAM–ALADIN Research on Mesoscale Operational NWP in Euromed, using AROME physics) is the current operational meso-scale weather forecasting model at FMI. It is jointly developed by 26 coun- tries from Europe and North-Africa and used operationally in 10 countries across Europe (Termonia et al. 2018). Harmonie-AROME has non-hydrostatic dynam- ics (Bénard et al. 2010; Bubnová et al. 1995) with fully compressible Eulerian equations (Laprise 1992; Simmons and Burridge 1981), which are suitable to be used for convective permitting scales. The equations are solved in time and space by using the semi-Lagrangian advection scheme and semi-Implicit two-time-level scheme, originating from the ECMWF global model (ECMWF 2020). The current operational Harmonie-AROME (cy40h1.1, in 2019) has the horizontal grid spac-
ing of 2.5km and 65 levels in vertical (Bengtsson et al. 2017). The lowest model level is at 10 m’s height and the top is at 10 hPa’s height. The model calculation time step is 75s. The model data assimilation configuration used in this study used radiosoundings, aircraft and surface weather station observations. The observa- tions were assimilated by using 3D-Var for upper air and Optimal Interpolation for surface (Gustafsson et al. 2018). The detailed description of physical parametriza- tion schemes can be found in Bengtsson et al. 2017 and Seity et al. 2011. Here, the most essential parameterization schemes from a wind energy point of view (turbulence and cloud microphysics) are briefly discussed.
The turbulence scheme called HARATU (Meijgaard et al. 2012), affects cloud formation and wind speed through the following processes. The turbulent mix- ing length is described by Lenderink and Holtslag 2004, and it includes two parts:
a stable conditions and a near-neutral convective condition. The stability coef- ficients take into account moist processes, like condensation and evaporation of cloud droplets, and their effect on latent heat. In addition, the prognostic equa- tion for turbulent kinetic energy (TKE) includes source and sink terms for: wind shear (+), buoyancy (+/-), transport (+/-) and dissipation of TKE (-).
The microphysics scheme in Harmonie-AROME is called ICE3 (Lascaux et al.
2006; Pinty and Jabouille 1998). ICE3 is a bulk scheme which handles water vapour and the water content in both iced and liquid forms. The package treats prognostic hydrometeors in five phases. Solid hydrometeors are separated into cloud ice, dry snow and a combination of graupel and hail. Liquid water is sepa- rated into cloud water and rain. The ICE3 takes into account evaporation, subli- mation and other interactions between the water phases. In addition, horizontal advection of the hydrometeors and vertical sedimentation are taken into account (Bouteloup et al. 2011).
The Harmonie-AROME model was used inPapers I,IIandIV. InPaper I, the early version of the model was used to create a Wind Atlas (described in section 3.2.1) to represent the current wind climate. The same Harmonie-AROME dataset that was created forPaper Iwas also used inPaper IIin creating the Icing Atlas.
In Paper IV, the Harmonie-AROME was used in forecast mode together with a separate icing model (see section 3.3.1) to predict icing events during winter 2016 - 2017.
36
3.3 DOWNSTREAM METHODS
3.3.1 ATMOSPHERIC ICING MODEL
The atmospheric icing model used in this study is based on an icing algorithm developed by Makkonen (Makkonen 2000). The international ISO STANDARD Atmospheric Icing on Structuresis based on this same algorithm (ISO, E 2017). In this study, the FMI icing model was further developed in Paper IIand validated later inPaper IV.
The icing model calculates the amount of ice accumulated over a 1-m high vertically oriented, freely rotating cylinder with a diameter of 3 cm. The diameter of the cylinder increases as the ice is accumulated. The equation 3.1 describes the rime ice rate[g h−1] over the standard cylinder, taking into account also melting Qm(eq. 3.2):
d M
d t =α1·α2·α3·LW C·A·V−Qm (3.1) and
Qm= (Qh+Qe+Ql)·S f, (3.2) In these equations wind speedV[m s−1]and liquid water contentLWC[g m−3] are taken as input values from the NWP model Harmonie-AROME. The icing model also uses air temperature Tai r[°C], pressure Pai r[Pa]and relative humidity RHai r [g m−3]as input from the NWP model. Wind drives the water droplets towards the cylinder, which has surface areaA[m2]seen by the wind.
The collision (α1), sticking (α2) and accretion (α3) are unitless coefficients and they describe the interactions between the cylinder and cloud water droplets.
The collision coefficient takes into account the Medium Volume Diameter (MVD) of water particles. The MVD is defined as Gaussian distribution of LWC as a func- tion of Cloud Droplet Particle Concentration (Nd). The Nd was set to 100 cm−3 (Hämäläinen and Niemelä 2017). The larger particles tend to hit the object due to inertia, and the collision coefficient gets a value 1. The smaller particles in- stead follow the streamlines of the wind, going around the object. Thus, the value decreases with particle size and is 0 for the smallest particles.
The sticking coefficient was set to 1, indicating that all the liquid droplets are supercooled (when the temperature is below 0°C) and freeze immediately when hitting a cold object. For dry particles like snow, graupel or hail the coefficient would get a value of 0, indicating that the particles would bounce back when hitting an object. Therefore, only the liquid part of the NWP model’s total water content was used as input to the icing model.
