ArticlePDF AvailableLiterature Review

Global maps of soil temperature

Authors:
Jonas J. Lembrechts at University of Antwerp
Jonas J. Lembrechts
Johan van den Hoogen
Johan van den Hoogen
Johan van den Hoogen
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Juha Aalto at University of Helsinki
Juha Aalto
Michael B Ashcroft at University of Wollongong
Michael B Ashcroft
Show all 406 authorsHide
Download full-text PDFDownload full-text PDF
Read full-text
Download citation

Abstract and Figures

Research in global change ecology relies heavily on global climatic grids derived from estimates of air temperature in open areas at around 2 m above the ground. These climatic grids do not reflect conditions below vegetation canopies and near the ground surface, where critical ecosystem functions occur and most terrestrial species reside. Here, we provide global maps of soil temperature and bioclimatic variables at a 1-km² resolution for 0–5 and 5–15 cm soil depth. These maps were created by calculating the difference (i.e., offset) between in-situ soil temperature measurements, based on time series from over 1200 1-km² pixels (summarized from 8500 unique temperature sensors) across all the world’s major terrestrial biomes, and coarse-grained air temperature estimates from ERA5-Land (an atmospheric reanalysis by the European Centre for Medium-Range Weather Forecasts). We show that mean annual soil temperature differs markedly from the corresponding gridded air temperature, by up to 10°C (mean = 3.0 ± 2.1°C), with substantial variation across biomes and seasons. Over the year, soils in cold and/or dry biomes are substantially warmer (+3.6 ± 2.3°C) than gridded air temperature, whereas soils in warm and humid environments are on average slightly cooler (-0.7 ± 2.3°C). The observed substantial and biome-specific offsets emphasize that the projected impacts of climate and climate change on near-surface biodiversity and ecosystem functioning are inaccurately assessed when air rather than soil temperature is used, especially in cold environments. The global soil-related bioclimatic variables provided here are an important step forward for any application in ecology and related disciplines. Nevertheless, we highlight the need to fill remaining geographic gaps by collecting more in-situ measurements of microclimate conditions to further enhance the spatiotemporal resolution of global soil temperature products for ecological applications.
Mean annual soil temperature shows significantly lower spatial variability than air temperature. (a) Global map of mean annual topsoil temperature (SBIO1, 0-5 cm depth, in °C), created by adding the monthly offset between soil and air temperature for the period 2000-2020 (Figure 2) to the monthly air temperature from CHELSA. A black mask is used to exclude regions where our models are extrapolating (i.e. interpolation values in Figure 5 are <0.9, 18% of pixels). Dark grey represents regions outside the modelling area. (b-c) Density plots of mean annual soil temperature across the globe (b) and for each Whittaker biome separately (c) for SBIO1 (dark grey, soil temperature), compared with BIO1 from CHELSA (light grey, air temperature), created by extracting 1,000,000 random points from the 1-km² gridded bioclimatic products. The numbers in (c) represent the standard deviations of air temperature (light grey) and soil temperature (dark grey). Biomes are ordered according to the median annual soil temperature values (vertical black line) from the highest temperature (subtropical desert) to the lowest (tundra)
Mean annual soil temperature shows significantly lower spatial variability than air temperature. (a) Global map of mean annual topsoil temperature (SBIO1, 0-5 cm depth, in °C), created by adding the monthly offset between soil and air temperature for the period 2000-2020 (Figure 2) to the monthly air temperature from CHELSA. A black mask is used to exclude regions where our models are extrapolating (i.e. interpolation values in Figure 5 are <0.9, 18% of pixels). Dark grey represents regions outside the modelling area. (b-c) Density plots of mean annual soil temperature across the globe (b) and for each Whittaker biome separately (c) for SBIO1 (dark grey, soil temperature), compared with BIO1 from CHELSA (light grey, air temperature), created by extracting 1,000,000 random points from the 1-km² gridded bioclimatic products. The numbers in (c) represent the standard deviations of air temperature (light grey) and soil temperature (dark grey). Biomes are ordered according to the median annual soil temperature values (vertical black line) from the highest temperature (subtropical desert) to the lowest (tundra)
… 
Models of the temperature offset between soil and air temperature have low standard deviations and good global coverage. Analyses for the temperature offset between in situ measured topsoil (0-5 cm depth) temperature and gridded air temperature. (a) Standard deviation (in °C) over the predictions from a cross-validation analysis that iteratively varied the set of covariates (explanatory data layers) and model hyperparameters across 100 models and evaluated model strength using 10-fold cross-validation, for January (left) and July (right), as examples of the two most contrasting months. (b) The fraction of axes in the multidimensional environmental space for which the pixel lies inside the range of data covered by the sensors in the database. Low values indicate increased extrapolation
Models of the temperature offset between soil and air temperature have low standard deviations and good global coverage. Analyses for the temperature offset between in situ measured topsoil (0-5 cm depth) temperature and gridded air temperature. (a) Standard deviation (in °C) over the predictions from a cross-validation analysis that iteratively varied the set of covariates (explanatory data layers) and model hyperparameters across 100 models and evaluated model strength using 10-fold cross-validation, for January (left) and July (right), as examples of the two most contrasting months. (b) The fraction of axes in the multidimensional environmental space for which the pixel lies inside the range of data covered by the sensors in the database. Low values indicate increased extrapolation
… 
The mean annual soil temperature (SBIO1, 1 x 1 km resolution) modelled here is consistently cooler than ERA5L (9 x 9 km) soil temperature in forested areas. (a) Spatial representation of the difference between SBIO1 based on our model and based on ERA5L soil temperature data. Negative values (blue colours) indicate areas where our model predicts cooler soil temperature. Dark grey areas (Greenland and Antarctica) are excluded from our models. Asterisk in Scandinavia indicates the highlighted area in panels d to f (see below). (b) Distribution of the difference between SBIO1 and ERA5L along the macroclimatic gradient (represented by SBIO1 itself) based on a random subsample of 50,000 points from the map in a). Red line from a Generalized Additive Model (GAM) with k = 4. (c-e) High-resolution zoomed panels of an area of high elevational contrast in Norway (from 66.0-66.4°N, 15.0-16.0°E) visualizing SBIO1 (c), ERA5L (d) and their difference (e), to highlight the higher spatial resolution as obtained with SBIO1
The mean annual soil temperature (SBIO1, 1 x 1 km resolution) modelled here is consistently cooler than ERA5L (9 x 9 km) soil temperature in forested areas. (a) Spatial representation of the difference between SBIO1 based on our model and based on ERA5L soil temperature data. Negative values (blue colours) indicate areas where our model predicts cooler soil temperature. Dark grey areas (Greenland and Antarctica) are excluded from our models. Asterisk in Scandinavia indicates the highlighted area in panels d to f (see below). (b) Distribution of the difference between SBIO1 and ERA5L along the macroclimatic gradient (represented by SBIO1 itself) based on a random subsample of 50,000 points from the map in a). Red line from a Generalized Additive Model (GAM) with k = 4. (c-e) High-resolution zoomed panels of an area of high elevational contrast in Norway (from 66.0-66.4°N, 15.0-16.0°E) visualizing SBIO1 (c), ERA5L (d) and their difference (e), to highlight the higher spatial resolution as obtained with SBIO1
… 
Figures - available via license: Creative Commons Attribution-NonCommercial 4.0 International
Content may be subject to copyright.
Global Change Biology - 2022 - Lembrechts - Global maps of soil temperatu
re.pdf
Content uploaded by Matteo Petit Bon
Author content
All content in this area was uploaded by Matteo Petit Bon on Feb 15, 2022
Content may be subject to copyright.
Global maps of soil temperatu
re.pdf
Available via license: CC BY-NC
Content may be subject to copyright.
Glob Change Biol. 2022;00:1–35.  
|
 1wileyonlinelibrary.com/journal/gcb
Received: 25 August 2021 
|
 Accepted: 27 October 2021
DOI: 10.1111/gcb.16060
INVITED REVIEW
Global maps of soil temperature
Jonas J. Lembrechts1 | Johan van den Hoogen2 | Juha Aalto3,4  |
Michael B. Ashcroft5,6 | Pieter De Frenne7 | Julia Kemppinen8 |
Martin Kopecký9,10  | Miska Luoto4 | Ilya M. D. Maclean11  |
Thomas W. Crowther2 | Joseph J. Bailey12  | Stef Haesen13  | David H. Klinges14,15  |
Pekka Niittynen4 | Brett R. Scheffers16  | Koenraad Van Meerbeek13  |
Peter Aartsma17  | Otar Abdalaze18  | Mehdi Abedi19  | Rien Aerts20  |
Negar Ahmadian19  | Antje Ahrends21  | Juha M. Alatalo22  |
Jake M. Alexander23  | Camille Nina Allonsius24  | Jan Altman9,10  |
Christof Ammann25  | Christian Andres26  | Christopher Andrews27  |
Jonas Ardö28  | Nicola Arriga29  | Alberto Arzac30  | Valeria Aschero31,32  |
Rafael L. Assis33  | Jakob Johann Assmann34,35  | Maaike Y. Bader36  |
Khadijeh Bahalkeh19  | Peter Barančok37  | Isabel C. Barrio38  |
Agustina Barros32  | Matti Barthel26 | Edmund W. Basham14  | Marijn Bauters39  |
Manuele Bazzichetto40  | Luca Belelli Marchesini41  | Michael C. Bell42  |
Juan C. Benavides43  | José Luis Benito Alonso44  | Bernd J. Berauer45,46  |
Jarle W. Bjerke47  | Robert G. Björk48,49  | Mats P. Björkman48,49  |
Katrin Björnsdóttir50  | Benjamin Blonder51  | Pascal Boeckx39  | Julia Boike52,53  |
Stef Bokhorst20  | Bárbara N. S. Brum54  | Josef Brůna9 | Nina Buchmann26  |
Pauline Buysse55 | José Luís Camargo56  | Otávio C. Campoe57  | Onur Candan58  |
Rafaella Canessa36, 59  | Nicoletta Cannone60  | Michele Carbognani61  |
Jofre Carnicer62,63  | Angélica Casanova- Katny64  | Simone Cesarz65,66  |
Bogdan Chojnicki67  | Philippe Choler68,69  | Steven L. Chown70  |
Edgar F. Cifuentes71  | Marek Čiliak72  | Tamara Contador73,74  | Peter Convey75  |
Elisabeth J. Cooper76  | Edoardo Cremonese77  | Salvatore R. Curasi78  |
Robin Curtis11 | Maurizio Cutini79  | C. Johan Dahlberg80,81  | Gergana N. Daskalova82  |
Miguel Angel de Pablo83  | Stefano Della Chiesa84  | Jürgen Dengler65,85,86  |
Bart Deronde87 | Patrice Descombes88  | Valter Di Cecco89  | Michele Di Musciano90  |
Jan Dick27  | Romina D. Dimarco91,92 | Jiri Dolezal9,93  | Ellen Dorrepaal94  |
This is an open access article under the terms of the Creat ive Commo ns Attri bution-NonCo mmercial License, which permits use, distribution and reproduction
in any medium, provided the original work is properly cited and is not used for commercial purposes.
© 2022 The Author s. Global Change Biology published by John Wiley & Sons Ltd.
Jonas J. L embrechts an d Johan van den Hoogen sho uld be considered joint f irst author.
Ivan Nijs and Jona than Lenoir should be co nsidered joint s enior author.
2 
|
    LE MBRECHTS E T aL.
Jiří Dušek95  | Nico Eisenhauer65,66  | Lars Eklundh28  | Todd E. Erickson96,97  |
Brigitta Erschbamer98  | Werner Eugster26  | Robert M. Ewers99 |
Dan A. Exton100  | Nicolas Fanin101  | Fatih Fazlioglu58  | Iris Feigenwinter26  |
Giuseppe Fenu102  | Olga Ferlian65,66  | M. Rosa Fernández Calzado103 |
Eduardo Fernández- Pascual10 4  | Manfred Finckh105  | Rebecca Finger Higgens106  |
T'ai G. W. Forte61  | Erika C. Freeman107  | Esther R. Frei108,109,110  |
Eduardo Fuentes- Lillo111,1 ,112  | Rafael A. García111,113  | María B. García114  |
Charly Géron1,115  | Mana Gharun26  | Dany Ghosn116  | Khatuna Gigauri117  |
Anne Gobin118,119  | Ignacio Goded29  | Mathias Goeckede120  |
Felix Gottschall65,66  | Keith Goulding121  | Sanne Govaert7 |
Bente Jessen Graae122  | Sarah Greenwood123  | Caroline Greiser80  |
Achim Grelle124  | Benoit Guénard125  | Mauro Guglielmin126 |
Joannès Guillemot127,128  | Peter Haase129,130  | Sylvia Haider131,65  |
Aud H. Halbritter132  | Maroof Hamid133  | Albin Hammerle134  |
Arndt Hampe135  | Siri V. Haugum132,136  | Lucia Hederová9 |
Bernard Heinesch137  | Carole Helfter138  | Daniel Hepenstrick139  |
Maximiliane Herberich140  | Mathias Herbst141 | Luise Hermanutz142  |
David S. Hik143  | Raúl Hoffrén144  | Jürgen Homeier145,146  | Lukas Hörtnagl26  |
Toke T. Høye147  | Filip Hrbacek148  | Kristoffer Hylander80  | Hiroki Iwata149  |
Marcin Antoni Jackowicz- Korczynski150,28  | Hervé Jactel151  | Järvi Järveoja152  |
Szymon Jastrzębowski153  | Anke Jentsch46,154  | Juan J. Jiménez155  | Ingibjörg
S. Jónsdóttir156  | Tommaso Jucker157  | Alistair S. Jump158  |
Radoslaw Juszczak67  | Róbert Kanka37  | Vít Kašpar9,159  | George Kazakis116 |
Julia Kelly160  | Anzar A. Khuroo133  | Leif Klemedtsson48  | Marcin Klisz153  |
Natascha Kljun160  | Alexander Knohl161  | Johannes Kobler162  | Jozef Kollár37  |
Martyna M. Kotowska146  | Bence Kovács163  | Juergen Kreyling164  |
Andrea Lamprecht165  | Simone I. Lang166  | Christian Larson167  |
Keith Larson168  | Kamil Laska148,169  | Guerric le Maire127,128  |
Rachel I. Leihy170  | Luc Lens171  | Bengt Liljebladh48  | Annalea Lohila172,173  |
Juan Lorite103,174  | Benjamin Loubet55  | Joshua Lynn132  | Martin Macek9 |
Roy Mackenzie73  | Enzo Magliulo175  | Regine Maier26  | Francesco Malfasi60  |
František Máliš176  | Matěj Man9 | Giovanni Manca29  | Antonio Manco175  |
Tanguy Manise137 | Paraskevi Manolaki177,178,179  | Felipe Marciniak54 |
Radim Matula10,180  | Ana Clara Mazzolari32 | Sergiy Medinets181,182,183  |
Volodymyr Medinets181  | Camille Meeussen7 | Sonia Merinero80  |
Rita de Cássia Guimarães Mesquita184  | Katrin Meusburger185  |
Filip J. R. Meysman186  | Sean T. Michaletz187  | Ann Milbau188  |
Dmitry Moiseev189 | Pavel Moiseev189  | Andrea Mondoni190  | Ruth Monfries21 |
   
|
 3
LEMBRECH TS ET aL.
Leonardo Montagnani191  | Mikel Moriana- Armendariz76  |
Umberto Morra di Cella192  | Martin Mörsdorf193  | Jonathan
R. Mosedale194  | Lena Muffler146  | Miriam Muñoz- Rojas195,196  |
Jonathan A. Myers197  | Isla H. Myers- Smith82  | Laszlo Nagy198 |
Marianna Nardino199  | Ilona Naujokaitis- Lewis200  | Emily Newling201 |
Lena Nicklas98  | Georg Niedrist202  | Armin Niessner203 | Mats B. Nilsson152  |
Signe Normand34,35  | Marcelo D. Nosetto204,205  | Yann Nouvellon127,128  |
Martin A. Nuñez92 ,206  | Romà Ogaya207,208  | Jérôme Ogée101  |
Joseph Okello39,209, 210  | Janusz Olejnik211  | Jørgen Eivind Olesen182  |
Øystein H. Opedal212  | Simone Orsenigo190  | Andrej Palaj37  |
Timo Pampuch213  | Alexey V. Panov214 | Meelis Pärtel215  | Ada Pastor178  |
Aníbal Pauchard111,113  | Harald Pauli165  | Marian Pavelka95  |
William D. Pearse216,217  | Matthias Peichl152  | Loïc Pellissier218,219 |
Rachel M. Penczykowski197  | Josep Penuelas207,208  | Matteo Petit Bon9,76 ,166  |
Alessandro Petraglia61  | Shyam S. Phartyal220  | Gareth K. Phoenix221  |
Casimiro Pio222  | Andrea Pitacco223  | Camille Pitteloud218,219  |
Roman Plichta180  | Francesco Porro190  | Miguel Portillo- Estrada1 |
Jérôme Poulenard224  | Rafael Poyatos63,225  | Anatoly S. Prokushkin30, 214  |
Radoslaw Puchalka226,227  | Mihai Pușcaș228,229,230  | Dajana Radujković1 |
Krystal Randall5, 231  | Amanda Ratier Backes65,131  | Sabine Remmele203 |
Wolfram Remmers232 | David Renault40,233  | Anita C. Risch234  |
Christian Rixen108,10 9  | Sharon A. Robinson5,231  | Bjorn J. M. Robroek235  |
Adrian V. Rocha236  | Christian Rossi237,238 | Graziano Rossi190  |
Olivier Roupsard239,240,241  | Alexey V. Rubtsov30  | Patrick Saccone165  |
Clotilde Sagot242 | Jhonatan Sallo Bravo243,24 4  | Cinthya C. Santos245  |
Judith M. Sarneel246  | Tobias Scharnweber213  | Jonas Schmeddes164 |
Marius Schmidt247  | Thomas Scholten248  | Max Schuchardt46  |
Naomi Schwartz249 | Tony Scott121  | Julia Seeber134,202  |
Ana Cristina Segalin de Andrade245 | Tim Seipel167  | Philipp Semenchuk250  |
Rebecca A. Senior251  | Josep M. Serra- Diaz252  | Piotr Sewerniak253  |
Ankit Shekhar26  | Nikita V. Sidenko214 | Lukas Siebicke161 |
Laura Siegwart Collier142, 254  | Elizabeth Simpson216  | David P. Siqueira255  |
Zuzana Sitková256  | Johan Six26  | Marko Smiljanic213  | Stuart W. Smith122, 257  |
Sarah Smith- Tripp258 | Ben Somers259  | Mia Vedel Sørensen122 |
José João L. L. Souza260  | Bartolomeu Israel Souza261  | Arildo Souza Dias245,262  |
Marko J. Spasojevic263  | James D. M. Speed264  | Fabien Spicher265  |
Angela Stanisci266  | Klaus Steinbauer165  | Rainer Steinbrecher267  |
Michael Steinwandter202  | Michael Stemkovski216  | Jörg G. Stephan268  |
4 
|
    LE MBRECHTS E T aL.
Christian Stiegler161  | Stefan Stoll232,269  | Martin Svátek180  |
Miroslav Svoboda10  | Torbern Tagesson28, 270  | Andrew J. Tanentzap107  |
Franziska Tanneberger271  | Jean- Paul Theurillat272,273  | Haydn J. D. Thomas82  |
Andrew D. Thomas274  | Katja Tielbörger59 | Marcello Tomaselli61  |
Urs Albert Treier34,35  | Mario Trouillier213  | Pavel Dan Turtureanu228,230,275  |
Rosamond Tutton276  | Vilna A. Tyystjärvi4,277  | Masahito Ueyama278  |
Karol Ujházy176  | Mariana Ujházyová72  | Domas Uogintas279  |
Anastasiya V. Urban180,214 | Josef Urban30,180  | Marek Urbaniak211  |
Tudor- Mihai Ursu280  | Francesco Primo Vaccari281  | Stijn Van de Vondel282  |
Liesbeth van den Brink59  | Maarten Van Geel283  | Vigdis Vandvik132  |
Pieter Vangansbeke7 | Andrej Varlagin284  | G. F. Veen285  |
Elmar Veenendaal286  | Susanna E. Venn287  | Hans Verbeeck288  |
Erik Verbrugggen1 | Frank G. A. Verheijen289  | Luis Villar290 | Luca Vitale291  |
Pascal Vittoz292  | Maria Vives- Ingla63  | Jonathan von Oppen34,35  |
Josefine Walz168  | Runxi Wang125  | Yifeng Wang276  | Robert G. Way276  |
Ronja E. M. Wedegärtner122  | Robert Weigel146  | Jan Wild9,159  |
Matthew Wilkinson42  | Martin Wilmking213  | Lisa Wingate101  |
Manuela Winkler165  | Sonja Wipf108,237  | Georg Wohlfahrt134  |
Georgios Xenakis293  | Yan Yang294  | Zicheng Yu295,296  | Kailiang Yu297  |
Florian Zellweger110  | Jian Zhang298  | Zhaochen Zhang298 | Peng Zhao152  |
Klaudia Ziemblińska211  | Reiner Zimmermann203,299  | Shengwei Zong300  |
Viacheslav I. Zyryanov214  | Ivan Nijs1 | Jonathan Lenoir265
1Research Group PLECO (Plant s and Ecosystems), University of Antwerp, Wilrijk, Belgium
2Department of Environmental Systems Science, Institute of Integrative Biology, ETH Zürich, Zürich, Switzerland
3Finnish Meteorological Institute, Helsinki, Finland
4Department of Geosciences and Geography, University of Helsinki, Finland
5Centre for Sustainable Ecosystem Solutions, School of Earth, Atmospheric and Life Sciences, University of Wollongong, Wollongong, New South Wales,
Australia
6Australian Museum, Sydney, Australia
7Forest & Nature Lab, Department of Environment, Ghent University, Melle- Gontrode, Belgium
8Geography Research Unit, University of Oulu, Oulu, Finland
9Institute of Botany of the Czech Academy of Sciences, Průhonice, Czech Republic
10Facult y of Forestry and Wood Sciences, Czech University of Life Sciences Prague, Prague 6 - Suchdol, Czech Republic
11Environment and Sustainability Institute, University of E xeter, Penr yn Campus, Penryn, UK
12Department of Geography, York St John University, York, UK
13Depar tment of Earth and Environmental Sciences, KU Leuven, Leuven, Belgium
14School of Natural Resources and Environment, University of Florida, Gainesville, Florida, USA
15Smithsonian Environmental Research Center, Edgewater, Maryland, USA
16Depar tment of Wildlife Ecology and Conservation, University of Florida, Gainesville, Florida, USA
17Depar tment of Natural Sciences and Environmental Health, University of South- Eastern Nor way, Bø, Norway
18Alpine Ecosystems Research Program, Institute of Ecology, Ilia State University, Tbilisi, Georgia
19Depar tment of Range Management, Faculty of Natural Resources and Marine Sciences, Tarbiat Modares University, Noor, Iran
20Depar tment of Ecological Science, Vrije Universiteit Amsterdam, The Netherlands
21Royal Botanic Garden Edinburgh, Edinburgh, UK
   
|
 5
LEMBRECH TS ET aL.
22Environmental Science Center, Qatar University, Doha, Qatar
23Depar tment of Environmental Systems Science, Institute of Integrative Biolog y, ETH Zurich, Zürich, Swit zerland
24Research group ECOBE, University of Ant werp, Wilrijk, Belgium
25Department of A groecology and Environment, A groscope Research Institute, Zürich, Switzerland
26Depar tment of Environmental Systems Science, ETH Zurich, Zurich, Switzerland
27UK Centre for Ecology and Hydrolog y, Penicuik, UK
28Depar tment of Physical Geography and Ecosystem Science, Lund University, Lund, Sweden
29European Commission, Joint Research Centre (JRC), Ispra, Italy
30Siberian Federal University, Krasnoyarsk, Russia
31Facultad de Ciencias Exactas y Naturales, Universidad Nacional de Cuyo, Mendoza, Argentina
32Instituto Argentino de Nivologiá, Glaciologiá y Ciencias Ambientales (IANIGLA), CONICET, CCT- Mendoza, Mendoza, Argentina
33Natural Histor y Museum, University of Oslo, Oslo, Norway
34Center for Sustainable Landscapes Under Global Change, Department of Biology, Aarhus University, Aarhus C, Denmark
35Center for Biodiversity Dynamics in a Changing World, Department of Biology, Aarhus University, Aarhus C, Denmark
36Ecological Plant Geography, Faculty of Geography, University of Marburg, Marburg, Germany
37Institute of Landscape Ecology Slovak Academy of Sciences, Bratislava, Slovakia
38Faculty of Environmental and Forest Sciences, Agricultural University of Iceland, Reykjaví k, Iceland
39Isotope Bioscience Laborator y - ISOF YS, Ghent University, Gent, Belgium
40Université de Rennes, CNRS, EcoBio (Ecosystèmes, biodiversité, évolution) - UMR 6553, Rennes, France
41Depar tment of Sustainable Agro- ecosystems and Bioresources, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy
42Forest Research, Alice Holt Lodge, Wrecclesham, Farnham, UK
43Department of Ecology, Pontificia Universidad Javeriana, Bogota, Colombia
44Jolube Consultor Botánico. C/Mariano R de Ledesma, Jaca, Huesca, Spain
45Institute of Landscape and Plant Ecology, Depar tment of Plant Ecology, University of Hohenheim, Stuttgart, Germany
46Disturbance Ecology, BayCEER, University of Bayreuth, Bayreuth, Germany
47Norwegian Institute for Nature Research, FR AM - High North Research Centre for Climate and the Environment, Tromsø, Norway
48Department of Earth Sciences, University of Gothenburg, Gothenburg, Sweden
49Gothenburg Global Biodiversity Centre, Gothenburg, Sweden
50Depar tment of Biological and Environmental Sciences, University of Gothenburg, Gothenburg, Sweden
51Department of Environmental Science, Polic y, and Management, University of California, Berkeley, California, USA
52Alfred Wegener Institute Helmholtz Center for Polar and Marine Research, Telegrafenberg A45, Potsdam, Germany
53Geography Department, Humboldt- Universität zu Berlin, Germany
54Pós- Graduação em Ciências de Florestas Tropicais, Instituto Nacional de Pesquisas da Amazônia, Manaus, Brasil
55UMR ECOSYS INRAE, Uinversité Paris Saclay, AgroParisTech, France
56Biological Dynamics of Forest Fragments Project, BDFFP, Instituto Nacional de Pesquisas da Amazônia, Manaus, Brazil
57Department of Forest Sciences, Federal University of Lavras, Lavras, Brazil
58Faculty of Arts and Sciences, Department of Molecular Biology and Genetics, Ordu University, Ordu, Turkey
59Plant Ecology Group, Department of Evolution and Ecology, University of Tübingen, Tübingen, Germany
60Department of Science and High Technology, Insubria University, Como, Italy
61Depar tment of Chemistry, Life Sciences and Environmental Sustainability, Universit y of Parma, Parma, Italy
62Depar tment of Evolutionary Biology, Ecology and Environmental Sciences, Biodiversity Research Institute (IRBio), University of Barcelona, Barcelona, Spain
63CREAF, E08193 Bellaterra (Cerdanyola del Vallès), Spain
64Laboratorio de Ecofisiología Vegetal y Cambio Climático, Laboratorio de Ecofisiología Vegetal y Cambio Climático, Departamento de Ciencias Veterinarias
y Salud Pública, Universidad Católica de Temuco, Campus Luis Rivas del Canto and Núcleo de Estudios Ambientales (NEA), Facultad de Recursos Naturales,
Universidad Católica de Temuco, Temuco, Chile
65German Centre for Integrative Biodiversity Research (iDiv) Halle- Jena- Leipzig, Leipzig, Germany
66Institute of Biology, Leipzig University, Leipzig, Germany
67Laboratory of Bioclimatology, Department of Ecolog y and Environmental Protection, Poznan University of Life Sciences, Poznan, Poland
68Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LECA, Grenoble, France
69Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LTSER Zone Atelier Alpes, Grenoble, France
70Securing Antarctica's Environmental Future, School of Biological Sciences, Monash University, Melbourne, Victoria, Australia
71Forest Ecology and Conservation Group, Department of Plant Sciences, University of Cambridge, Cambridge, UK
72Faculty of Ecology and Environmental Sciences, Technical University in Zvolen, Zvolen, Slovakia
73Millennium Institute Biodiversity of Antarctic and Subantarctic Ecosystems (BASE), University Austral of Chile, Valdivia, Chile
6 
|
    LE MBRECHTS E T aL.
74Cape Horn International Center (CHIC), Puer to Williams, Chile
75British Antarctic Survey, NERC , High Cross, Cambridge, UK
76Department of Arctic and Marine Biology, Faculty of Biosciences Fisheries and Economics, UiT- The Arctic University of Norway, Tromsø, Norway
77Climate Change Unit, Environmental Protection Agency of Aosta Valley, Italy
78Depar tment of Biological Sciences, University of Notre Dame, Notre Dame, Indiana, USA
79Department of Science, University of Roma Tre, Rome, It aly
80Department of Ecology, Environment and Plant Sciences and Bolin Centre for Climate Research, Stockholm University, Stockholm, Sweden
81The Count y Administrative Board of Västra Götaland, Gothenburg, Sweden
82School of GeoSciences, University of Edinburgh, Edinburgh, UK
83Department of Geology, Geography and Environment, University of Alcalá, Madrid, Spain
84Chair of Geoinformatics, Technische Universität Dresden, Dresden, Germany
85Vegetation Ecology, Institute of Natural Resource Sciences (IUNR), ZHAW Zurich University of Applied Sciences, Wädenswil, Switzerland
86Plant Ecology, Bayreuth Center of Ecology and Environmental Research (BayCEER), University of Bayreuth, Bayreuth, Germany
87VITO- TAP, Mol, Belgium
88Swiss Federal Research Institute WSL, Birmensdorf, Switzerland
89Majella Seed Bank, Majella National Park, Colle Madonna, Lama dei Peligni, Italy
90Depar tment of Life, Health and Environmental Sciences, University of L'Aquila, L'Aquila, Italy
91Grupo de Ecología de Poblaciones de Insectos, IFAB (INTA - CONICET), Bariloche, Argentina
92Department of Biology and Biochemistry, Universit y of Houston, Houston, Texas, USA
93Faculty of Science, Department of Botany, University of South Bohemia, České Budějovice, Czech Republic
94Climate Impacts Research Centre, Department of Ecology and Environmental Science, Umeå University, Abisko, Sweden
95Global Change Research Institute, Academy of Sciences of the Czech Republic, Czech Republic
96School of Biological Sciences, The Universit y of Western Australia, Crawley, Western Australia, Australia
97Kings Park Science, Department of Biodiversity, Conservation and Attractions, Kings Park, Australia
98Depar tment of Botany, Faculty of Biology, Universit y of Innsbruck, Innsbruck, Austria
99Imperial College London, Ascot, UK
100Operation Wallacea, Lincolnshire, UK
101INRAE, Bordeaux Sciences Agro, UMR 1391 ISPA, Villenave d'Ornon, France
102Department of Life and Environmental Sciences, Universit y of Cagliari, Cagliari, Italy
103Department of Botany, University of Granada, Granada, Spain
104IMIB – Biodiversity Research Institute, University of Oviedo, Mieres, Spain
105Institute for Plant Science and Microbiology, University of Hamburg, Hamburg, Germany
106Dartmouth College, Hanover, New Hampshire, USA
107Ecosystems and Global Change Group, Department of Plant Sciences, University of Cambridge, Cambridge, UK
108WSL Institute for Snow and Avalanche Research SLF, Davos Dorf, Switzerland
109Climate Change, Extremes and Natural Hazards in Alpine Regions Research Center CERC, Davos Dor f, Switzerland
110Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Birmensdorf, Switzerland
111Laboratorio de Invasiones Biológicas (LIB), Facultad de Ciencias Forestales, Universidad de Concepción, Concepción, Chile
112School of Education and Social Sciences, Adventist University of Chile, Chile
113Instituto de Ecología y Biodiversidad (IEB), Santiago, Chile
114Pyrenean Institute of Ecolog y (CSIC), Zaragoza, Spain
115Biodiversity and Landscape, TERRA Research Centre, Gembloux Agro- Bio Tech, University of Liège, Gembloux, Belgium
116Department of Geo- information in Environmental Management , Mediterranean Agronomic Institute of Chania, Chania, Greece
117Department of Environmental Management and Policy, Georgian Institute of Public Affairs, Tbilisi, Georgia
118Flemish Institute for Technological Research, Mol, Belgium
119Department of Earth and Environmental Science, Faculty of BioScience Engineering, KULeuven, Belgium
120Depar tment of Biogeochemical Signals, Max Planck Institute for Biogeochemistry, Jena, Germany
121Sustainable Agricultural Sciences Department, Rothamsted Research, Harpenden, UK
122Department of Biology, Norwegian University of Science and Technology, Trondheim, Nor way
123Biodiversity, Wildlife and Ecosystem Health, Biomedical Sciences, University of Edinburgh, Edinburgh, UK
124Department of Ecology, Swedish University of Agricultural Sciences, Uppsala, Sweden
125School of Biological Sciences, The University of Hong Kong, Hong Kong SAR, China
126Department of Theoretical and Applied Sciences, Insubria University, Varese, Italy
   
|
 7
LEMBRECH TS ET aL.
127CIRAD, UMR Eco&Sols, Montpellier, France
128Eco&Sols, Univ Montpellier, CIRAD, INR AE, IRD, Montpellier SupAgro, Montpellier, France
129Senckenberg Research Institute and Natural History Museum Frankfurt, Gelnhausen, Germany
130Faculty of Biology, University of Duisburg- Essen, Essen, Germany
131Institute of Biology / Geobotany and Botanical Garden, Martin Luther University Halle- Wittenberg, Halle (Saale), Germany
132Depar tment of Biological Sciences and Bjerknes Centre for Climate Research, University of Bergen, Bergen, Norway
133Centre for Biodiversity and Taxonomy, Department of Botany, Universit y of Kashmir, Srinagar, India
134Department of Ecology, University of Innsbruck, Innsbruck, Austria
135INRAE, Univ. Bordeaux, BIOGECO, Cestas, France
136The Heathland Centre, Alver, Nor way
137TERR A Teaching and Research Center, Faculty of Gembloux Agro- Bio Tech, University of Liege, Gembloux, Belgium
138UK Centre for Ecology and Hydrology, Penicuik, UK
139Vegetation Ecolog y, Institute of Natural Resource Sciences, ZHAW Zurich University of Applied Sciences, Grüental, Switzerland
140Institute for Botany, University of Natural Resources and Life Sciences Vienna (BOKU), Vienna, Austria
141Centre for Agrometeorological Research (Z AMF), German Meteorological Ser vice (DWD), Braunschweig, Germany
142Dept of Biology, Memorial University, St. John's, Newfoundland, Canada
143Department of Biological Sciences, Simon Fraser University, Burnaby, British Columbia, Canada
144Department of Geography, University of Zaragoza, Zaragoza, Spain
145Facult y of Resource Management, HAWK University of Applied Sciences and Arts, Göttingen, Germany
146Plant Ecology, Albrecht- von- Haller- Institute for Plant Sciences, Georg- August University of Göttingen, Göttingen, Germany
147Department of Ecoscience and Arctic Research Centre, Aarhus University, Rønde, Denmark
148Department of Geography, Faculty of Science, Masar yk University, Brno, Czech Republic
149Department of Environmental Science, Shinshu University, Matsumoto, Japan
150Department of Ecoscience and Arctic Research Centre, Aarhus University, Roskilde, Denmark
151INR AE, Universit y of Bordeaux, BIOGECO, Cestas, France
152Department of Forest Ecology and Management, Swedish University of A gricultural Sciences, Umeå, Sweden
153Depar tment of Silviculture and Forest Tree Genetics, Forest Research Institute, Raszyn, Poland
154Bayreuth Center of Ecolog y and Environmental Research, Bayreuth, Germany
155ARAID/IPE- CSIC, Pyrenean Institute of Ecology, Avda. Llano de la Victoria, Spain
156Life and Environmental Sciences, University of Iceland, Reykjavík, Iceland
157School of Biological Sciences, University of Bristol, Bristol, UK
158Biological and Environmental Sciences, Faculty of Natural Sciences, University of Stirling, Scotland
159Faculty of Environmental Sciences, Czech University of Life Sciences Prague, Prague 6 - Suchdol, Czech Republic
160Centre for Environmental and Climate Science, Lund University, Lund, Sweden
161Bioclimatology, University of Göttingen, Göttingen, Germany
162Environment Agency Austria, Vienna, Austria
163Centre for Ecological Research, Institute of Ecology and Botany, Vácrátót, Hungar y
164Experimental Plant Ecology, Institute of Botany and Landscape Ecology, University of Greifswald, Greifswald, Germany
165GLORIA Coordination, Institute for Interdisciplinary Mountain Research, Austrian Academy of Sciences (ÖAW) & Department of Integrative Biology and
Biodiversity Research, University of Natural Resources and Life Sciences, Vienna, Austria
166Department of Arctic Biology, The University Centre in Svalbard (UNIS), Long yearbyen, Svalbard, Norway
167Department of Land Resources and Environmental Sciences, Montana State University, Bozeman, Montana, USA
168Climate Impacts Research Centre, Department of Ecology and Environmental Sciences, Umeå University, Abisko, Sweden
169Centre for Polar Ecology, Faculty of Science, University of South Bohemia, České Budějovice, Czech Republic
170School of Biological Sciences, Monash University, Melbourne, Vic toria, Australia
171Terrestrial Ecolog y Unit, Depar tment of Biolog y, Ghent University, Gent, Belgium
172Finnish Meteorological Institute, Climate System Research, Helsinki, Finland
173INAR Institute for Atmospheric and Earth System Research/Physics, Faculty of Science, University of Helsinki, Finland
174Interuniversity Institute for Earth System Research, University of Granada, Granada, Spain
175CNR Institute for Agricultural and Forestr y Systems in the Mediterranean, Portici (Napoli), Italy
176Faculty of Forestry, Technical Universit y in Zvolen, Zvolen, Slovakia
177School of Pure & Applied Sciences, Environmental Conservation and Management Programme, Open University of Cyprus, Latsia, Cyprus
178Department of Biology, Aarhus University, Aarhus C , Denmark
8 
|
    LE MBRECHTS E T aL.
179Aarhus Institute of Advanced Studies, AIAS Høegh- Guldbergs Gade 6B, Aarhus, Denmark
180Department of Forest Botany, Dendrology and Geobiocoenology, Faculty of Forestr y and Wood Technology, Mendel University in Brno, Brno, Czech
Republic
181Regional Centre for Integrated Environmental Monitoring, Odesa National I.I. Mechnikov University, Odesa, Ukraine
182Depar tment of Agroecology, Aarhus University, Tjele, Denmark
183NGO New Energy, Kharkiv, Ukraine
184Biological Dynamics of Forest Fragments Project , Coordenação de Dinâmica Ambiental, Instituto Nacional de Pesquisas da Amazônia, Manaus, Brazil
185Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Birmensdorf, Switzerland
186Department of Biology, University of Antwerp, Wilrijk, Belgium
187Depar tment of Botany and Biodiversity Research Centre, Universit y of British Columbia, Vancouver, British Columbia, Canada
188Department of Environment, Province of Antwerp, Antwerpen, Belgium
189Institute of Plant and Animal Ecology of Ural Division of Russian Academy of Science, Ekaterinburg, Russia
190Department of Earth and Environmental Sciences, Universit y of Pavia, Pavia, Italy
191Faculty of Science and Technology, Free University of Bolzano, Bolzano, Italy
192Climate Change Unit, Environmental Protection Agency of Aosta Valley, Saint- Christophe, Italy
193Chair of Geobotany, University of Freiburg, Freiburg, Germany
194Environment and Sustainability Institute, University of Exeter, Penryn C ampus, Cornwall, UK
195Centre for Ecosystem Science, School of Biological, Earth and Environmental Sciences, UNSW Sydney, Sydney, New South Wales, Australia
196Department of Plant Biology and Ecology, University of Seville, Seville, Spain
197Department of Biology, Washington Universit y in St. Louis, St. Louis, Missouri, USA
198Department of Animal Biology, Institute of Biology, University of Campinas, Campinas, Brazil
199CNR Institute of BioEconomy, Bologna, Italy
200National Wildlife Research Centre, Environment and Climate Change Canada, Carleton University, Ottawa, Ontario, Canada
201School of Life and Environmental Sciences, Deakin University, Burwood, Victoria, Australia
202Institute for Alpine Environment, Eurac Research, Bozen/Bolzano, Italy
203Institute of Biology, Department of Molecular Botany, University of Hohenheim, Stuttgart, Germany
204Instituto de Matemática Aplicada San Luis, IMASL, CONICET and Universidad Nacional de San Luis, San Luis, Argentina
205Cátedra de Climatología Agrícola (FCA- UNER), Entre Ríos, Argentina
206Grupo de Ecología de Invasiones, INIBIOMA, CONICET/ Universidad Nacional del Comahue, Bariloche, Argentina
207CSIC, Global Ecolog y Unit CREAF- CSIC- UAB, Bellaterra, Spain
208CREAF, Spain
209Mountains of the Moon University, Fort Portal, Uganda
210National Agricultural Research Organisation, Mbarara Zonal Agricultural Research and Development Institute, Mbarara, Uganda
211Laborator y of Meteorolog y, Department of Construction and Geoengineering, Faculty of Environmental Engineering and Mechanical Engineering, Poznan
University of Life Sciences, Poznan, Poland
212Depar tment of Biology, Lund University, Lund, Sweden
213Institute of Botany and L andscape Ecology, University Greifswald, Greifswald, Germany
214V.N. Sukachev Institute of Forest SB RAS, Krasnoyarsk, Russia
215Institute of Ecology and Earth Sciences, Universit y of Tartu, Tartu, Estonia
216Department of Biology and Ecology Center, Utah State Universit y, Logan, Utah, USA
217Department of Life Sciences, Imperial College, Ascot, Berkshire, UK
218Landscape Ecology, Institute of Terrestrial Ecosystems, Depar tment of Environmental Systems Science, ETH Zürich, Zürich, Switzerland
219Unit of Land Change Science, Swiss Federal Research Institute WSL, Birmensdorf, Switzerland
220School of Ecology and Environment Studies, Nalanda University, Rajgir, India
221School of Biosciences, University of Sheffield, Sheffield, UK
222CESAM & Department of Environment, University of Aveiro, Aveiro, Portugal
223Department of Agronomy, Food, Natural resources, Animals and Environment - University of Padua, Legnaro, Italy
224Univ. Savoie Mont Blanc, CNRS, Univ. Grenoble Alpes, EDYTEM, Chambéry, France
225Universitat Autònoma de Barcelona, Spain
226Department of Ecolog y and Biogeography, Faculty of Biological and Veterinary Sciences, Nicolaus Copernicus University, Toruń, Poland
227Centre for Climate Change Research, Nicolaus Copernicus University, Toruń, Poland
228A. Borza Bot anic Garden, Babeș- Bolyai University, Cluj- Napoca, Romania
229Faculty of Biolog y and Geology, Department of Taxonomy and Ecology, Babeș- Bolyai University, Cluj- Napoca, Romania
230E. G. Racoviț ă Institute, Babeș- Bolyai University, Cluj- Napoca, Romania
   
|
 9
LEMBRECH TS ET aL.
231Securing Antarctica's Environmental Future, School of Earth, Atmospheric and Life Sciences, University of Wollongong, Wollongong, New South Wales,
Australia
232University of Applied Sciences Trier, Environmental Campus Birkenfeld, Birkenfeld, Germany
233Institut Universitaire de France, Paris, France
234Swiss Federal Institute for Forest, Snow and Landscape Research WSL , Birmensdorf, Switzerland
235Aquatic Ecology and Environmental Biology, Radboud Institute for Environmental and Biological Sciences, Radboud University Nijmegen, Nijmegen, The
Netherlands
236Depar tment of Biological Sciences and the Environmental Change Initiative, University of Notre Dame, Notre Dame, Indiana, USA
237Swiss National Park, Chastè Planta- Wildenberg, Zernez, Switzerland
238Remote Sensing L aboratories, Department of Geography, University of Zurich, Zurich, Switzerland
239CIRAD, UMR Eco&Sols, Dakar, Senegal
240Eco&Sols, Univ Montpellier, CIRAD, INRAE, IRD, Institut Agro, Montpellier, France
241LMI IESOL, Centre IRD- ISR A de Bel Air, Dakar, Senegal
242Parc national des Ecrins - Domaine de Charance, France
243Universidad Nacional de San Antonio Abad del Cusco, Cusco, Perú
244Centro de Investigación de la Biodiversidad Wilhelm L. Johannsen, Cusco, Perú
245Biological Dynamics of Forest Fragments Project, PDBFF, Instituto Nacional de Pesquisas da Amazônia, Manaus, Brazil
246Department of Ecology and Environmental Science, Umeå Universit y, Umeå, Sweden
247Institute of Bio- and Geosciences (IBG- 3): A grosphere, Forschungszentrum Jülich GmbH, Jülich, Germany
248Chair of Soil Science and Geomorphology, Department of Geosciences, University of Tuebingen, Tuebingen, Germany
249Department of Geography, The University of British Columbia, Vancouver, British Columbia, Canada
250Depar tment of Botany and Biodiversity Research, Vienna, Austria
251Princeton School of Public and International Affairs, Princeton University, Princeton, New Jersey, USA
252Université de Lorraine, AgroParisTech, INRAE, Nancy, France
253Department of Soil Science and L andscape Management, Faculty of Earth Sciences and Spatial Management, Nicolaus Copernicus University, Toruń, Poland
254Terra Nova National Park, Parks Canada Agency, Glovertown, Newfoundland, Canada
255Universidade Estadual do Norte Fluminense Darcy Ribeiro, Rio de Janeiro, Brazil
256National Forest Centre, Forest Research Institute Zvolen, Zvolen, Slovakia
257Depar tment of Physical Geography, Stockholm University, Stockholm, Sweden
258Depar tment of Geography, University of British Columbia, Vancouver, British Columbia, Canada
259Department of Earth and Environmental Sciences, Leuven, Belgium
260Soil Science Department, Federal University of Viçosa, Viçosa- MG, Brazil
261Departamento de Geociências. Cidade Universitária, Universidade Federal da Paraíba, João Pessoa - PB, Brasil
262Depar tment of Physical Geography, Goethe- Universität Frankfurt, Frankfurt am Main, Germany
263Department of Evolution, Ecology, and Organismal Biology, University of California Riverside, Riverside, California, USA
264Department of Natural History, NTNU University Museum, Norwegian University of Science and Technology, Trondheim, Norway
265UMR 7058 CNRS ‘Ecologie et Dynamique des Systèmes Anthropisés’ (EDYSAN), Univ. de Picardie Jules Verne, Amiens, France
266EnvixL ab, Dipartimento di Bioscienze e Territorio, Università degli Studi del Molise, Termoli, Italy
267Institute of Meteorology and Climate Research (IMK), Department of Atmospheric Environmental Research (IFU), Karlsruhe Institute of Technology (KIT),
Garmisch- Partenkirchen, Germany
268Swedish University of A gricultural Sciences, SLU Swedish Species Information Centre, Uppsala, Sweden
269Faculty for Biology, University Duisburg- Essen, Essen, Germany
270Department of Geosciences and Natural Resource Management, University of Copenhagen, Copenhagen, Denmark
271Experimental Plant Ecology, Institute of Botany and Landscape Ecolog y, University of Greifswald, partner in the Greifswald Mire Centre, Greifswald,
Germany
272Foundation J.- M. Aubert, Champex- Lac, Switzerland
273Dépar tement de Botanique et Biologie végétale, Université de Genève, Chambésy, Switzerland
274Department of Geography and Ear th Sciences, Aberystwyth University, Wales, UK
275Center for Systematic Biology, Biodiversity and Bioresources - 3B, Babeș- Bolyai University, Cluj- Napoca, Romania
276Nor thern Environmental Geoscience Laboratory, Department of Geography and Planning, Queen's Universit y, Kingston, Ontario, Canada
277Finnish Meteorological Inst, Helsinki, Finland
278Graduate School of Life and Environmental Sciences, Osaka Prefecture University, Japan
279Nature Research Centre, Vilnius, Lithuania
280Institute of Biological Research Cluj- Napoca, National Institute of Research and Development for Biological Sciences, Bucharest, Romania
10 
|
    LE MBRECHTS E T aL.
281CNR Institute for BioEconomy, Firenze, Italy
282The Ecosystem Management Research Group (ECOBE), University of Antwerp, Wilrijk (Antwerpen), Belgium
283Plant Conservation and Population Biology, Department of Biology, KU Leuven, Heverlee, Belgium
284A.N. Sever tsov Institute of Ecology and Evolution, Russian Academy of Sciences, Moscow, Russia
285Netherlands Institute of Ecology, Wageningen, the Netherlands
286Plant Ecology and Nature Conser vation Group, Wageningen University, Wageningen, the Netherlands
287Centre for Integrative Ecology, School of Life and Environmental Sciences, Deakin University, Burwood, Victoria, Australia
288CAVElab - Computational and Applied Vegetation Ecology, Department of Environment , Ghent University, Gent, Belgium
289Earth Surface Processes Team, Centre for Environmental and Marine Studies (CESAM), Department of Environment and Planning, University of Aveiro,
Aveiro, Por tugal
290Instituto Pirenaico de Ecología, IPE- CSIC. Av. Llano de la Victoria, Jaca (Huesca), Spain
291CNR - Institute for Agricultural and Forestry Systems in the Mediterranean, Portici, Italy
292Institute of Ear th Sur face Dynamics, Faculty of Geosciences and Environment, Universit y of Lausanne, Géopolis, Switzerland
293Forest Research, Northern Research Station, Roslin, UK
294Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu, P.R. China
295MOE Key Laboratory of Geographical Processes and Ecological Security in Changbai Mountains, School of Geographical Sciences, Nor theast Normal
University, Changchun, China
296Department of Earth and Environmental Sciences, Lehigh University, Bethlehem, Pennsylvania, USA
297High Meadows Environmental Institute, Princeton University, New Jersey, USA
298Zhejiang Tiantong Forest Ecosystem National Observation and Research Station, School of Ecological and Environmental Sciences, East China Normal
University, Shanghai, China
299Ecological- Botanical Gardens, University of Bayreuth, Bayreuth, Germany
300Key Laboratory of Geographical Processes and Ecological Securit y in Changbai Mountains, Ministry of Education, School of Geographical Sciences,
Northeast Normal University, Changchun, China
Abstract
Research in global change ecology relies heavily on global climatic grids derived from
estimates of air temperature in open areas at around 2 m above the ground. These cli-
matic grids do not reflect conditions below vegetation canopies and near the ground
surface, where critical ecosystem functions occur and most terrestrial species reside.
Here, we provide global maps of soil temperature and bioclimatic variables at a 1- km2
resolution for 0– 5 and 5– 15 cm soil depth. These maps were created by calculating
the difference (i.e. offset) between in situ soil temperature measurements, based on
time series from over 1200 1- km2 pixels (summarized from 8519 unique tempera-
ture sensors) across all the world's major terrestrial biomes, and coarse- grained air
temperature estimates from ERA5- Land (an atmospheric reanalysis by the European
Centre for Medium- Range Weather Forecasts). We show that mean annual soil tem-
perature differs markedly from the corresponding gridded air temperature, by up to
10°C (mean = 3.0 ± 2.1°C), with substantial variation across biomes and seasons. Over
the year, soils in cold and/or dry biomes are substantially warmer (+3.6 ± 2.3°C) than
gridded air temperature, whereas soils in warm and humid environments are on aver-
age slightly cooler (−0.7 ± 2.3 °C). The obser ved subst antial and biome- specific offsets
emphasize that the projected impacts of climate and climate change on near- surface
biodiversity and ecosystem functioning are inaccurately assessed when air rather
than soil temperature is used, especially in cold environments. The global soil- related
Correspondence
Jonas J. Lembrechts, Research Group
PLECO (Plant s and Ecosystems),
University of Antwerp, 2610 Wilrijk,
Belgium.
Email: jonas.lembrechts@uantwerpen.be
Jonathan Lenoir, UMR 7058 CNRS
‘Ecologie et Dynamique des Systèmes
Anthropisés’ (EDYSAN), Univ. de Picardie
Jules Verne, Amiens, France.
Email: jonathan.lenoir@u-picardie.fr
Funding information
Fonds Wetenschappelijk Onderzoek,
Grant/Award Number: 12P1819N, and
G018919N, W001919N; for full list of
funders see acknowledgements
   
|
  11
LEMBRECH TS ET aL.
1 |  INTRODUCTION
With the rapidly increasing availability of big data on species dis-
tributions, functional traits and ecosystem functioning (Bond-
Lamberty & Thomson, 2018; Bruelheide et al., 2018; Kattge et al.,
2019; Kissling et al., 2018; Lenoir et al., 2020), we can now study
biodiversity and ecosystem responses to global changes in unprec-
edented detail (Antão et al., 2020; van den Hoogen et al., 2019;
Senior et al., 2019; Steidinger et al., 2019). However, despite this
increasing availability of ecological data, most spatially explicit
studies of ecological, biophysical and biogeochemical processes
still have to rely on the same global gridded temperature data (Du
et al., 2020; van den Hoogen et al., 2019; Soudzilovskaia et al.,
2015). Thus far, these global gridded products are based on mea-
surements from standard meteorological stations that record free-
air temperature inside well- ventilated protective shields placed up
to 2 m above- ground in open, shade- free habitats, where abiotic
conditions may differ substantially from those actually experienced
by most organisms (Lembrechts et al., 2020; World Meteorological
Organization, 2008).
Ecological patterns and processes often relate more directly
to below- canopy soil temperature rather than to well- ventilated
air temperature inside a weather station. Near- surface, rather
than air, temperature better predicts ecosystem functions like
biogeochemical cycling (e.g. organic matter decomposition, soil
respiration and other aspects of the global carbon balance) (Davis
et al., 2020; Gottschall et al., 2019; Hursh et al., 2017; Jian et al.,
2021; Perera- Castro et al., 2020; Pleim & Gilliam, 2009; Portillo-
Estrada et al., 2016; Schimel et al., 2004). Similarly, the use of soil
temperature in correlative analyses or predictive models may im-
prove predictions of climate impacts on organismal physiology
and behaviour, as well as on population and community dynam-
ics and species distributions (Ashcroft et al., 2008; Berner et al.,
2020; Kearney et al., 2009; Körner & Paulsen, 2004; Opedal et al.,
2015; Scherrer et al., 2011; Schimel et al., 2004; Zellweger et al.,
2020). Given the key role of soil- related processes for both abo-
veground and belowground parts of the ecosystem and their
feedbacks to the atmosphere (Crowther et al., 2016), adequate
soil temperature data are critical for a broad range of fields of
study, such as ecology, biogeography, biogeochemistry, agronomy,
soil science and climate system dynamics. Nevertheless, existing
global soil temperature products such as those from ERA5- Land
(Copernicus Climate Change Service (C3S), 2019), with a resolu-
tion of 0.08 × 0.08 degrees (≈9 × 9 km at the equator) remain too
coarse for most ecological applications.
The direction and magnitude of the difference or offset between
in situ soil temperature and coarse- gridded air temperature prod-
ucts result from a combination of two factors: (i) the (vertical) mi-
croclimatic difference between air and soil temperature and (ii) the
(horizontal) mesoclimatic difference between air temperature in flat,
cleared areas (i.e. where meteorological stations are located) and air
temperature within different vegetation types (e.g. below a dense
canopy of trees) or topographies (e.g. within a ravine or on a ridge)
(De Frenne et al., 2021; Lembrechts et al., 2020). In essence, the
offset is thus the combination of both the vertical and horizontal dif-
ferences that result from factors affecting the energy budget at the
Earth's surface, principally radiative energy: the ground absorbs ra-
diative energy, which is transferred to the air by convective heat ex-
change, evaporation and spatial variation in net radiation, and lower
convective conductance near the Earth's surface results in horizon-
tal and vertical variation in temperature (Geiger, 1950; Richardson,
1922). Both these vertical and horizontal differences in temperature
vary significantly across the globe and in time as a result of environ-
mental conditions affecting the radiation budget (e.g. as a result of
topographic orientation, canopy cover or surface albedo), convec-
tive heat exchange and evaporation (e.g. foliage density, variation in
the degree of wind shear caused by surface friction) and the capacity
for the soil to store and conduct heat (e.g. water content and soil
structure and texture) (De Frenne et al., 2019; Geiger, 1950; Way &
Lewkowicz, 2018; Zhang et al., 2008).
Although the physics of soil temperatures have long been well
unders to od (Geiger, 1950; Richardson , 1922), the cre ation of high-
resolution global gridded soil temperature products has not been
feasible before, partially due to the absence of detailed global
in situ soil temperature measurements (Lembrechts et al., 2020;
Lembrechts & Lenoir, 2019). Recently, however, the call for mi-
croclimate temperature data representative of in situ conditions
(i.e. microhabitat) as experienced by organisms living close to the
ground surface or in the soil has become more urgent (Bramer
et al., 2018). In this paper, we address this issue by generating
bioclimatic variables provided here are an important step forward for any application
in ecology and related disciplines. Nevertheless, we highlight the need to fill remain-
ing geographic gaps by collecting more in situ measurements of microclimate con-
ditions to further enhance the spatiotemporal resolution of global soil temperature
products for ecological applications.
KEYWORDS
bioclimatic variables, global maps, microclimate, near- surface temperatures, soil- dwelling
organisms, soil temperature, temperature offset, weather stations
12 
|
    LE MBRECHTS E T aL.
global gridded maps of below- canopy and near- surface soil tem-
perature at 1- km² resolution (in line with most existing global air
temperature products). These maps are more representative of
the habitat conditions as experienced by organisms living under
vegetation canopies, in the topsoil or near the soil surface. They
were created using the abovementioned offset between gridded
air temperature data and in situ soil temperature measurements.
We expect these soil temperature maps to be substantially more
representative of actual microclimatic conditions than existing
products as they capture relevant near- surface and belowground
abiotic conditions where ecosystem functions and processes
operate (Bramer et al., 2018; Daly, 2006; Körner & Hiltbrunner,
2018). Indeed, the offset between free- air (macroclimate) and soil
(microclimate) temperature, and between cleared areas and other
habitats, can easily reach up to ±10°C annually, even at the 1- km2
spatial resolution used here (Lembrechts et al., 2019; Wild et al.,
2019; Zhang et al., 2018).
To create the global gridded soil temperature maps introduced
above, we used over 8500 time series of soil temperature mea-
sured in situ across the world's major terrestrial biomes, which are
compiled and stored in the SoilTemp database (Lembrechts et al.,
2020) (Figure 1a, Figure S1) and averaged into 1200 (or 1000 for
the second soil layer) unique 1- km2 pixels. First, to illustrate the
magnitude of the studied effect, we visualized the global and
biome- specific patterns in the mean annual offset between in situ
soil temperature (0– 5 cm and 5– 15 cm depth) and coarse- scale
interpolated air temperature from ERA5- Land using the average
within 1 × 1 km grid cells. Hereafter, we refer to this difference
between soil temperature and air temperature as the temperature
offset (or offset), sensu (De Frenne et al., 2021); elsewhere called
the surface offset (Smith & Riseborough, 1996, 2002). Secondly, we
used a machine learning approach with 31 environmental predictor
variables (including macroclimate, soil, topography, reflectance,
vegetation and anthropogenic variables) to model the spatial vari-
ation in monthly temperature offsets at a 1 × 1 km resolution for
all continents except Antarctica (as not covered by many of the
used predictor variable layers). Using these offsets, we then cal-
culated relevant soil- related bioclimatic variables (SBIO), mirroring
the existing global bioclimatic variables for air temperature. Finally,
we compared the modelled mean annual temperature (SBIO1, top-
soil layer) with a similar product based on monthly ERA5L topsoil
(0–7 cm) temperature with a spatial resolution of 0.08 × 0.08 de-
grees (≈9 × 9 km at the equator).
2 |  METHODS
2.1  |  Data acquisition
Analyses are based on SoilTemp, a global database of microclimate
time series (Lembrechts et al., 2020). We compiled soil temperature
measurements from 9362 unique sensors (mean duration 2.9 years,
median duration 1.0 year, ranging from 1 month to 41 years) from
60 countries, using both published and unpublished data sources
(Figure 1, Figure S1). Each sensor corresponds to one independent
time series.
We us ed time seri es spanning a minimum of 1 month , wit h a tem-
poral resolution of 4 h or less. Sensors of any type were included
(Table S1), as long as they measured in situ. Sensors in experimen-
tally manipulated plots, that is, plots in which microclimate has been
manipulated, such as in open top chambers, were excluded. Most
data (>90%) came from low- cost rugged microclimate loggers such
as iButtons (Maxim Integrated, USA) or TMS4- sensors (Wild et al.,
2019), with measurement errors of around 0. 5– 1°C (note that we are
using degree Celsius over Kelvin throughout, for ease of understand-
ing), while in a minority of cases sensors with higher meteorological
specifications such as industrial or scientific- grade thermocouples
and thermistors (measurement errors of less than 0.5°C) were used.
Contributing data sets mostly consisted of short- term regional net-
works of microclimate measurements, yet also included a set (<5%)
of soil temperature sensors from long- term research networks
equipped with weather stations (e.g. Pastorello et al., 2017). By
combining these two types of data, a much higher spatial density of
sensors and broader distribution of microhabitats could be obtained
than by using weather station data only.
About 68% of sensors were deployed between 2010 and 2020
and 93% between 2000 and 2020; we, thus, focus on the latter
period in our analyses. Additionally, given the relatively short time
frame covered by most individual sensors and thus the lack of spa-
tially unbiased long- term time series, we were not able to test for
systematic differences in the temperature offset between old and
recent data sets, and thus we did not correct for this in our models.
We strongly urge future studies to assess such temporal dynamics in
the offset once long- term microclimate data have become sufficient
and more available.
For each of the individual 9362 time series, we calculated
monthly mean, minimum (5% percentile of all monthly values) and
maximum (95% percentile) temperature, after checking all time series
FIGURE 1 Temperature offsets between soil and air temperature differed significantly among biomes. (a) Distribution of in situ
measurement locations across the globe, coloured by the mean annual temperature offset (in °C) between in situ measured soil temperature
(topsoil, 0– 5 cm depth) and gridded air temperature (ERA5- Land). Offsets were averaged per hexagon, each with a size of approximately
70,000 km². Mollweide projection. (b) Mean annual temperature offsets per Whittaker biome (adapted from Whittaker 1970, based on
geographic location of sensors averaged at 1 km2; 0– 5 cm depth), ordered by mean temperature offset and coloured by mean annual
precipitation. (c– d) Distribution of sensors in 2D climate space for the topsoil (c, 0– 5 cm depth, N = 4530) and the second layer (d, 5– 15 cm
depth, N = 3989). Colours of hexagons indicate the number of sensors at each climatic location, with a resolution of 1.2°C (x- axis) and 100
mm (y- axis). Grey dots in the background represent the global variation in climatic space (obtained by sampling 1,000,000 random locations
from the CHELSA world maps). Overlay with grey lines depicts a delineation of Whittaker biomes
   
|
  13
LEMBRECH TS ET aL.
for plausibility and erroneous data. These monthly values, while per-
haps not fully intercomparable between the Northern and Southern
Hemisphere, are those that have traditionally been used to calculate
bioclimatic variables (Fick & Hijmans, 2017). Months with more than
1 day of missing data, either at the beginning or end of the measure-
ment period, or due to logger malfunctioning during measurement,
were excluded, resulting in a final subset of 380,676 months of soil
temperature time series that were used for further analyses. For
14 
|
    LE MBRECHTS E T aL.
each sensor with more than 12 months of data, we calculated mov-
ing averages of annual mean temperature, using each consecutive
month as a starting month and calculating the mean temperature
including the next 11 months. We used these moving averages to
make maximal use of the full temporal extent covered by each sen-
sor because each time series spanned a different time period, often
including parts of calendar years only.
The selected data set contained sensors installed strictly below-
ground, measuring temperature at depths between 0 and 200 cm
below the ground surface. Sensors recording several measurements
at the same site but located at different (vertical) depths were in-
cluded separately (the 9362 unique sensors thus came from 7251
unique loggers).
Sensors were grouped in different soil depth categories (0– 5, 5–
15, 15– 30, 30– 60, 60– 100, 100– 200 cm, Table S2) to incorporate the
effects of soil temperature dampening associated with vertical stratifi-
cation. We limited our analyses to the topsoil (0– 5 cm) and the second
soil layer (5– 15 cm), as we currently lack sufficient global coverage to
make accurate models at deeper soil depths (8519 time series, about
91%, came from the two upper depth layers). Due to uncertainty in the
identification of these soil depths between studies (e.g. due to litter
layers), no finer categorization is used.
We tested for potential bias in temporal resolution (i.e. mea-
surement interval) by calculating mean, minimum and maximum
temperature for a selection of 2000 months for data measured
every 15 min, and the same data aggregated to 30, 60, 90, 120 and
240 min. Monthly mean, minimum and maximum temperature cal-
culated with any of the aggregated data sets differed on average
less than 0.2°C from the ones with the highest temporal resolution.
We were, thus, confident that pooling data with different temporal
resolutions of 4 h or finer would not significantly affect our results.
2.2  |  Temperature offset calculation
For each monthly value at each sensor location (see Table S3 for
number of data points per month), we extracted the correspond-
ing monthly means of the 2 m air temperature from the European
Centre for Medium- Range Weather (ECMWF) Forecast's 5th reanal-
ysis (ERA5) (from 1979 to 1981) and ERA5- Land from 1981 to 2020
(Copernicus Climate Change Service (C3S), 2019), hereafter called
ERA5L. The latter data set models the global climate with a spatial
resolution of 0.08 × 0.08 degrees (≈9 × 9 km at the equator) with an
hourly resolution, converted into monthly means using daily means
for the whole month. Similarly, monthly minima and maxima were
obtained from TerraClimate (Abatzoglou et al., 2018) for the period
2000 to 2020 at a 0.0 4 × 0.04 degrees (≈4 × 4 km at the equa-
tor) resolution. Monthly means for TerraClimate were not available,
and we therefore estimated them by averaging the monthly minima
and maxima. Finally, we also obtained monthly mean temperatures
from CHELSA (Karger et al., 2017a, 2017b) for the period 2000 to
2013 at a 30 × 30 arc second (≈1 × 1 km at the equator) resolu-
tion. In our modelling exercises (see section 2.5 Modelling below),
we opted to use the mean temperature offsets as calculated based
on ERA5L rather than on CHELSA. While CHELSA’s higher spatial
resolution is definitely an advantage, its time period (stopping in
2013) insufficiently overlapped with the time period covered by our
in situ measurements (2000– 2020), soil temperature offsets based
on the CHELSA data set were only used for comparative purposes.
We used TerraClimate to model offsets in monthly minimum and
maximum temperature.
We calculated moving annual averages of the gridded air tem-
perature data in the same way as for soil temperature. These were
used to create annual temperature offset values following the same
approach as above.
The offset between the in situ measured soil temperature in the
SoilTemp database and the 2 m free- air temperature obtained from
the air- temperature grids (ERA5L, TerraClimate and CHELSA, hereaf-
ter called ‘gridded air temperature’) was calculated by subtracting the
monthly or annual mean air temperature from the monthly or annual
mean soil temperature. Positive offset values indicate a measured soil
temperature higher than gridded air temperature, whereas negative off-
set values represent cooler soils. Similarly, monthly minimum and max-
imum air temperature were subtracted from minimum and maximum
soil temperature, respectively. Monthly minima and maxima of the soil
temperature were calculated as, respectively, the 5% lowest and highest
instantaneous measurement in that month, to correct for outliers, which
can be especially pronounced at the soil surface (Speak et al., 2020). As a
result, patterns in minima and maxima are more conservative estimates
than if we had used the absolute lowest and highest values.
Importantly, the temperature offset calculated here is a result of
three key groups of drivers: (1) height effects (2 m versus 0– 15 cm
below the soil surface); (2) environmental or habitat effects (e.g. spatial
variability in vegetation, snow or topography); and (3) spatial scale ef-
fects (resolution of gridded air temperature) (Lembrechts et al., 2020).
We investigated the potential role of scale effects by comparing grid-
ded air temperature data sources with different resolutions (ERA5L,
TerraClimate and CHELSA, see below; Figures S2-S3). Height effects
and environmental effects are, however, not disentangled here, as the
offset we propose incorporates both the difference between air and
soil temperature (vertically), as well as the difference between free- air
macroclimate and in situ microclimate (horizontally) in one measure
(Lembrechts et al., 2020). While it can be argued that it would be bet-
ter to treat both vertical and horizontal effects separately, this would
require a similar database of coupled in situ air and soil temperature
measurements, which is not yet available. Using in situ measured air
temperature could also solve spatial mismatches (i.e. spatially averaged
air temperature represents the whole 1 to 81 km2 pixel, depending on
pixel size, not only the exact location of the sensor). However, coupled
air and soil temperature measurements are not only rare, but the air
temperature measurements also have large measurement errors, es-
pecially in open habitats (Maclean et al., 2021). These errors can be up
to several degrees in open habitats when using non- standardized sen-
sors, loggers and shielding (Holden et al., 2013; Maclean et al., 2021;
Terando et al., 2017). Hence, using in situ measured air temperature
without correcting for these measurement errors would be misleading.
   
|
  15
LEMBRECH TS ET aL.
2.3  |  Global and biome- level analyses
For the purpose of visualization, annual offsets were first averaged in
hexagons with a resolution of approximately 70,000 km2, using the
dggr idR- pack age (ver sion 2. 0.4) in R (Barnes et al., 2017) (Fig ure 1). Next,
we plotted mean, minimum and maximum annual soil temperature as
a function of corresponding gridded air temperature from ERA5L,
TerraClimate and CHELSA and used generalized additive models
(GAMs, package mgcv 1.8- 31; Wood, 2012) to visualize deviations from
the 1:1- line (i.e. temperature offsets deviating from zero, Figures S4– S5).
All annual and monthly values within each soil depth category
and falling within the same 1- km2 pixel were aggregated as a mean,
resulting in a total of c. 1200 unique pixels at 0– 5 cm, and c. 1000
unique pixels at 5– 15 cm each month, across the globe (Tables S3–
S5). This averaging includes summarizing the data over space, that is,
multiple sensors within the same 1- km² pixel, and time, that is, data
from multi- year time serie s from a cer tain sensor, to reduce spatial and
temporal autocorrelation and sampling bias. We assigned these 1- km2
averages to the corresponding Whittaker biome of their georefer-
enced location, using the package plotbiomes (version 0.0.0.9901) in R
(Figure 1c,d, Tables S4– S5 (Stefan & Levin, 2018)). We ranked biomes
based on their offset and compared this with the mean annual precip-
itation in each biome (Figure 1b). This was done separately for each
air temperature data source (ERA5L, TerraClimate and CHELSA), soil
depth (0– 5 cm, 5– 15 cm) and time frame (ERA5L 1979– 2020, 2000–
2020), as well as for the offset between monthly minimum and max-
imum soil temperature and the minimum and maximum gridded air
temperature from TerraClimate. Our analyses showed that patterns
were robust to variation in spatial resolution, sensor depth, climate
interpolation method and temporal scale (Figures S2– S5).
2.4  |  Acquisition of global predictor variables
To create spatial predictive models of the offset between in situ soil
temperature and gridded air temperature, we first sampled a stack of
global map layers at each of the logger locations within the data set.
These layers included long- term macroclimatic conditions, soil tex-
ture and physiochemical information, vegetation, radiation and topo-
graphic indices as well as anthropogenic variables. Details of all layers,
including descriptions, units and source information, are described in
Supplementary Data S1. In short, information about soil texture, struc-
ture and physiochemical properties was obtained from SoilGrids (ver-
sion 1 [Hengl et al., 2017]), limited to the upper soil layer (top 5 cm).
Long- term averages of macroclimatic conditions (i.e. monthly mean,
maximum and minimum temperature, monthly precipitation) was ob-
tained from CHELSA (version 2017 [Karger et al., 2017a]), which in-
cludes climate data averaged across 1979– 2013, and from WorldClim
(version 2 [Fick & Hijmans, 2017]). Monthly snow probability is based
on a pixel- wise frequency of snow occurrence (snow cover >10%) in
MODIS daily snow cover products (MOD10A1 & MYD10A1 [Hall
et al., 2002]) in 2001– 2019. Spectral vegetation indices (i.e. averaged
MODIS NDVI product MYD13Q1) and surface reflectance data (i.e.
MODIS MCD43A4) were obtained from the Google Earth Engine
Data Catalog (developers.google.com/earth- engine/datasets) and av-
eraged from 2015 to 2019. Landcover and topographic information
were obtained from EarthEnv (Amatulli et al., 2018). Aridity index (AI)
and potential evapotranspiration layers were obtained from CGIAR
(Zomer et al., 2008). Anthropogenic information (population density)
was obtained from the EU JRC (ghsl.jrc.ec.europa.eu/ghs_pop2019.
php). Aboveground biomass data were obtained from GlobBiomass
(Santoro, 2018). RESOLVE ecoregion classifications were used to cat-
egorize sampling locations into biomes (Dinerstein et al., 2017). With
this set of predictor variables, we included information on all differ-
ent categories of drivers of soil temperature. An important variable
that had to be excluded was snow depth, due to the lack of a relevant
1- km2 resolution global product. The final set of predictor variables
included 24 ‘static’ variables and eight monthly layers (i.e. maximum,
mean, and minimum temperature, precipitation, cloud cover, solar ra-
diation, water vapour pressure and snow cover). As cloud cover es-
timates were not available for high- latitude regions in the Northern
Hemisphere in January and December due to a lack of daylight, we
excluded cloud cover as an explanatory variable for these months (i.e.
‘EarthEnvCloudCover_MODCF_monthlymean_XX’, with XX repre-
senting the months in two- digit form Supplementary Data S1).
All variable map layers were reprojected and resampled to a
unified pixel grid in EPSG:4326 (WGS84) at 30 arc- sec resolution
(≈1 × 1 km at the equator). Areas covered by permanent snow or
ice (e.g. the Greenland ice cap or glaciated mountain ranges, iden-
tified using SoilGrids) were excluded from the analyses. Antarctic
sampling points were excluded from the modelling data set owing to
the limited coverage of several covariate layers in the region.
2.5  |  Modelling
To generate global maps of monthly temperature offsets (Figure 2), we
trained Random Forest (RF) models for each month, using the temper-
ature offsets as the response variables and the global variable layers as
predic tors (Breiman, 20 01; Hen gl et al ., 2018). We used a geospatial RF
modelling pipeline as developed by van den Hoogen et al. (2021). RF
models are machine learning models that combine many classification
trees using randomized subsets of the data, with each tree iteratively
dividing data into groups of most closely related data points (Hengl
et al., 2018). They are particularly valuable here due to their capacity
to uncover nonlinear relationships (e.g. due to increased decoupling of
soil from air temperature in colder and thus snow- covered areas) and
their ability to capture complex interactions among covariates (e.g. be-
tween snow and vegetation cover) (Olden et al., 2008). Furthermore,
they may currently have advantages over mechanistic microclimate
models for global modelling (Maclean & Klinges, 2021), as the latter
require highly detailed physical input parameters for calibration, and
current computational barriers preclude global assessments at a 1 km2
resolution and over multiple decades. Nevertheless, we urge future
endeavours to compare and potentially improve our results with esti-
mates based on such mechanistic models.
16 
|
    LE MBRECHTS E T aL.
We performed a grid search procedure to tune the RF models across
a range of 52 hyperparameter settings (variables per split: 2– 14, mini-
mum leaf population: 2– 5, in all combinations adding up to 52 models,
each time with 250 trees). During this procedure, we assessed each of
the 52 model's performance using k- fold cross- validation (k = 10; folds
assigned randomly, stratified per biome). The models’ mean and standard
FIGURE 2 Global modelled temperature offsets between soil and air temperature show strong spatiotemporal variation across months.
Modelled annual (a) and monthly (b– m) temperature offset (in °C) between in situ measured soil temperature (topsoil, 0– 5 cm) and gridded
air temperature. Positive (red) values indicate soils that are warmer than the air. Dark grey represents regions outside the modelling area
   
|
  17
LEMBRECH TS ET aL.
deviation values were the basis for choosing the best of all evaluated
models. This procedure was repeated for each month separately for the
two soil depth layers (0– 5 cm, 5– 15 cm), for offsets in mean, minimum
and maximum temperature. The importance of predictor variables was
assessed using the variable importance and ordered by mean variable
importance across all models. This variable importance adds up the
decreases in the impurity criterion (i.e. the measure on which the local
optimal condition is chosen) at each split of a node for each individual
variable over all trees in the forest (van den Hoogen et al., 2021).
2.6  |  Soil bioclimatic variables
The resulting global maps of the annual and monthly offsets be-
tween mean, minimum and maximum soil and air temperature
were used to calculate relevant bioclimatic variables following the
definition used in CHELSA, BIOCLIM, ANUCLIM and WorldClim
(Booth et al., 2014; Fick & Hijmans, 2017; Karger et al., 2017a; Xu
& Hutchinson, 2011) (Table 1, Figures 3– 4). First, we calculated
monthly soil mean, maximum and minimum temperature by adding
monthly temperature offsets to the respective CHELSA monthly
mean, maximum and minimum temperature (Karger et al., 2017a).
Next, we used these soil temperature layers to compute 11 soil
bioclimatic layers (SBIO, Table 1) (O’Donnell & Ignizio, 2012).
Wettest and driest quarters were identified for each pixel based
on CHELSA’s monthly values.
2.7  |  Model uncertainty
To assess the uncertainty in the monthly models, we performed
a stratified bootstrapping procedure, with total size of the boot-
strap samples equal to the original training data (van den Hoogen
et al., 2021). Using biomes as a stratification category, we ensured
the samples included in each of the bootstrap training collections
were proportionally representative of each biome's total area. Next,
we trained RF models (with the same hyperparameters as selected
during the grid- search procedure) using each of 100 bootstrap
iterations. Each of these trained RF models was then used to clas-
sify the predictor layer stack, to generate per- pixel 95% confidence
intervals and standard deviation for the modelled monthly offsets
(Figure 5a, Figure S6a). The mean R² value of the RF models for the
monthly mean temperature offset was 0.70 (from 0.64 to 0.78) at
0– 5 cm and 0.76 (0.63– 0.85) at 5 to 15 cm across all 12 monthly
models. Mean RMSE of the models was 2.20°C (1.94– 2.51°C) at
0– 5 cm, and 2.06°C (1.67– 2.35°C) at 5– 15 cm.
Importantly, model uncertainty as reported in Figure 5a and
Figure S6a comes on top of existing uncertainties in (1) in situ soil
temperature measurements and (2) the ERA5L macroclimate models
as used in our models. However, both of those are usually under 1°C
(Copernicus Climate Change Service (C3S), 2019; Wild et al., 2019).
To assess the spatial extent of extrapolation, which is necessary
due to the incomplete global coverage of the training data, we first per-
formed a principal component analysis (PCA) on the full environmental
space covered by the monthly training data, including all explanatory
variables as used in the models, and then transformed the composite
image into the same principal comonents' (PC) spaces as of the sampled
data (van den Hoogen et al., 2019). Next, we created convex hulls for
each of the bivariate combinations from the first 10 to 12 PC s, cove ring
at least 90% of the sample space variation, with the number of PCs de-
pending on the month. Using the coordinates of these convex hulls, we
assessed whether each pixel fell within or outside each of these convex
hulls and calculated the percentage of bivariate combinations for which
this was the case (Figure 5b, Figure S6b). This process was repeated for
each month and for each of the two soil depths separately.
These uncertainty maps are important because one should
be careful with extrapolation beyond the range of conditions
covered by the environmental variables included in the original
calibration data set, especially in the case of non- linear patterns
such as modelled here. The maps are provided as spatial masks to
remove or reduce the weighting of the pixels for which predic-
tions are beyond the ra nge of valu es covere d by the mod els du ring
calibration. To assess this further, we used a spatial leave- one- out
cross- validation analysis to test for spatial autocorrelation in the
data set (Figure S7) (van den Hoogen et al., 2021). This approach
trains a model for each sample in the data set on all remaining
Bioclimatic variable Meaning
SBIO1 annual mean temperature
SBIO2 mean diurnal range (mean of monthly (max temp - min temp))
SBIO3 isothermality (SBIO2/SBIO7) (×100)
SBIO4 temperature seasonality (standard deviation ×100)
SBIO5 max temperature of warmest month
SBIO6 min temperature of coldest month
SBIO7 temperature annual range (SBIO5- SBIO6)
SBIO8 mean temperature of wettest quar ter
SBIO9 mean temperature of driest quarter
SBIO10 mean temperature of warmest quarter
SBIO11 mean temperature of coldest quarter
TAB LE 1  Overview of soil bioclimatic
variables as calculated in this study
18 
|
    LE MBRECHTS E T aL.
samples, excluding data points that fall within an increasingly
large buffer around that focal sample. Results show lowest confi-
dence for May to September at 5– 15 cm, likely driven by uneven
global coverage of data points.
Finally, we compared the modelled mean annual temperature
(SBIO1, topsoil layer) with a similar product based on monthly ERA5L
topsoil (0– 7 cm) temperature with a spatial resolution of 0.08 × 0.08 de-
grees (≈9 × 9 km at the equator, Copernicus Climate Change Service
(C3S), 2019). The corresponding SBIO1 based on ERA5L was calculated
using the means of the monthly averages for each month over the pe-
riod 1981 to 2016, and averaging these 12 monthly values into one
annual product. We then visualized spatial differences between SBIO1
and ERA5, as well as differences across the macroclimatic gradient, to
identify mismatches between both data sets.
All geospatial modelling was performed using the Python API in
Google Earth Engine (Gorelick et al., 2017). The R statistical soft-
ware, version 4.0.2 (R Core Team, 2020), was used for data visu-
alizations. All maps were plotted using the Mollweide projection,
FIGURE 3 Soil bioclimatic variables. Global maps of bioclimatic variables for topsoil (0– 5 cm depth) climate, calculated using the maps of
the monthly offsets between soil and air temperature (see Figure 2), and the bioclimatic variables for air temperature from CHELSA
   
|
  19
LEMBRECH TS ET aL.
which preserves relative areas, to avoid large distortions at high
latitudes.
2.8  |  Sources of uncertainty
The temporal mismatch between the period covered by CHELSA (1979–
2013) and our in situ measurements (2000– 2020) prevented us from
directly using CHELSA climate to calculate the temperature offsets used
in our models. This temporal mismatch might affect the offsets calculated
here because the relationship between temperature offset and macrocli-
mate will change through time as the climate warms. Similarly, inter- annual
differences in offsets due to specific weather conditions cannot be im-
plemented in the used approach. However, we are confident that, at the
relatively coarse spatial (1 km2) and temporal (monthly averages) resolution
we are working at, our results are sufficiently robust to withstand these
FIGURE 4 Mean annual soil temperature shows significantly lower spatial variability than air temperature. (a) Global map of mean
annual topsoil temperature (SBIO1, 0– 5 cm depth, in °C), created by adding the monthly offset between soil and air temperature for the
period 2000– 2020 (Figure 2) to the monthly air temperature from CHELSA. A black mask is used to exclude regions where our models are
extrapolating (i.e. interpolation values in Figure 5 are <0.9, 18% of pixels). Dark grey represents regions outside the modelling area. (b– c)
Density plots of mean annual soil temperature across the globe (b) and for each Whittaker biome separately (c) for SBIO1 (dark grey, soil
temperature), compared with BIO1 from CHELSA (light grey, air temperature), created by extracting 1,000,000 random points from the
1- km² gridded bioclimatic products. The numbers in (c) represent the standard deviations of air temperature (light grey) and soil temperature
(dark grey). Biomes are ordered according to the median annual soil temperature values (vertical black line) from the highest temperature
(subtropical desert) to the lowest (tundra)
20 
|
    LE MBRECHTS E T aL.
temporal issues, given that we found high consistency in offset patterns
between the different time frames and air temperature data sets examined
(Figures S2– S5). Nevertheless, we strongly urge future research to disen-
tangle these potential temporal dynamics, especially given the increasing
rate at which the climate is warming (GISTEMP Team, 2021; Xu et al., 2018).
Similarly, a potential bias could result from the mismatch in method
and resolution between ERA5L— used to calculate the temperature
offsets— and CHELSA, which was used to create the bioclimatic vari-
ables. However, even though temperature offsets have slightly larger
variation when based on the coarser- grained ERA5L- data than on the
finer- grained CHELSA- data, Figures S2– S5 show that relationships be-
tween soil and air temperature are largely consistent in all biomes and
across the whole global temperature gradient. Therefore, the larger
offsets created additional random scatter, yet no consistent bias.
Finally, we acknowledge that the 1- km² resolution gridded prod-
ucts might not be representative of conditions at the in situ mea-
surement locations within each pixel. This issue could be particularly
significant for different vegetation types (here proxied at the pixel
level using total aboveground biomass (unit: tons/ha i.e., Mg/ha,
for the year 2010; Santoro, 2018) and NDVI (MODIS NDVI product
MYD13Q1, averaged over 2015– 2019). To verify this, we compared
a pixel's estimated aboveground biomass with the dominant in situ
habitat (forest versus open) surrounding the sensors in that pixel
(Table S6). Importantly, all sensors installed in forests fell indeed in
pixels with more than 1 ton/ha aboveground biomass. Similarly, 75%
or more of sensors in open terrain fell in pixels with biomass estimates
of less than 1 ton/ha. Only in the temperate woodland biome was the
match between in situ habitat estimates and pixel- level aboveground
biomass lower, with less than 95% of sensors in forested locations cor-
rectly placed in pixels with more than 1 ton/ha biomass, and less than
50% of open terrain sensors in pixels with less than 1 ton/ha biomass.
While our predictions will thus not be accurate for locations within a
pixel that largely deviate from average conditions (e.g. open terrain
in pixels identified as largely forested, or vice versa), they should be
largely representative for those pixel- level averages.
3 |  RESULTS
3.1  |  Biome- wide patterns in the temperature
offset
We found positive and negative temperature offsets of up to 10°C
between in situ measured mean annual topsoil temperature and grid-
ded air temperature (mean = 3.0 ± 2.1°C standard deviation, Figure 1,
0– 5 cm, depth; 5– 15 cm is available in Figures S2, S5). The magnitude
FIGURE 5 Models of the temperature offset between soil and air temperature have low standard deviations and good global coverage.
Analyses for the temperature offset between in situ measured topsoil (0– 5 cm depth) temperature and gridded air temperature. (a) Standard
deviation (in °C) over the predictions from a cross- validation analysis that iteratively varied the set of covariates (explanatory data layers) and
model hyperparameters across 100 models and evaluated model strength using 10- fold cross- validation, for January (left) and July (right), as
examples of the two most contrasting months. (b) The fraction of axes in the multidimensional environmental space for which the pixel lies
inside the range of data covered by the sensors in the database. Low values indicate increased extrapolation
   
|
  21
LEMBRECH TS ET aL.
and direction of these temperature offsets varied considerably within
and across biomes. Mean annual topsoil temperature was on aver-
age 3.6 ± 2.3°C higher than gridded air temperature in cold and/or
dry biomes, namely tundra, boreal forests, temperate grasslands
and subtropical deserts. In contrast, offsets were slightly negative in
warm and wet biomes (tropical savannas, temperate forests and tropi-
cal rainforests) where soils were, on average, 0.7 ± 2.7°C cooler than
gridded air temperature (Figure 1b, Figures S2 and S5; note, however,
the lower spatial coverage in these biomes in Figure 1a,c,d, Table S4).
Temperature offsets in annual minimum and maximum temperature
amounted to c. 10°C maximum. While annual soil temperature minima
were on average higher than corresponding gridded air temperature
minima in all biomes, temperature offsets of annual maxima followed
largely the same biome- related trends as seen for the annual means,
albeit with the higher variability expected for temperature extremes
(Figures S2g, S2h, S4g, S4h). Using different air temperature data
sources did not alter the annual temperature offset and biome- related
patterns (see Methods and Figures S2– S5).
Soils in the temperate seasonal forest biome were on average
0.8°C (±2.2°C) cooler than air temperature within 1- km2 grid cells
of forested habitats, and 1.0°C (±4.0°C) warmer than the air within
1 - k m 2 grid cells of non- forested habitats, resulting in a biome- wide
average of 0.5°C (Table S7). Similar patterns were obser ved in other
biomes.
3.2  |  Temporal and spatial variation in
temperature offsets
Our RF outputs highlighted a strong seasonality in monthly temper-
ature offsets, especially towards higher latitudes (Figure 2). High-
latitude soils were found to be several degrees warmer than the
air (monthly offsets of up to 25°C) during their respective winter
months, and cooler (up to 10°C) in summer months, both at 0– 5 cm
(Figure 2) and 5– 15 cm (Figure S8) soil depths. In the tropics and
subtropics, soils in dry biomes (e.g. in the Sahara Desert or southern
Africa) were predicted to be warmer than air throughout most of
the year, while soils in mesic biomes (e.g. tropical biomes in South
America, central Africa and Southeast Asia) were modelled to be
consistently cooler, at both soil depths. These global gridded prod-
ucts were then used to create temperature- based global bioclimatic
variables for soils (SBIO, Figure 3, Figure S9).
3.3  |  Global variation in soil temperature
We observed 17% less spatial variation in mean annual soil tem-
perature globally (expressed by the standard deviation) than in air
temperature, largely driven by the positive offset between soil
and air temperature in cold environments (Figure 4). Importantly,
our machine learning models slightly (up to 1°C, or around 10% of
variation) underestimated temperature offsets at both extremes
of the temperature gradient at the 1- km² resolution (Figure S10)
and likely even more in comparison with finer- resolution prod-
ucts. Estimates of the reduction in variation across space are thus
conservative, especially in the coldest biomes. The reduction in
spatial temperature variation was observed in all cold and cool
biomes, with tundra and boreal forests having both a significant
positive me an te mperatur e offset and a reduction of 20% and 22%
in variation, respectively (Figure 4c). In the warmest biomes (e.g.
tropical savanna and subtropical desert), however, we found an
increase in variation of, on average, 10%.
Our bootstrap approach to validate modelled monthly offsets in-
dicated high consistency among the outcomes of 100 bootstrapped
models (Figure 5, Figure S6a), with standard deviations in most months
and across most parts of the globe around or below ±1°C. One excep-
tion to this was the temperature offset at high latitudes of the Northern
Hemisphere during winter months (standard deviation up to ±5°C in the
0– 5 cm layer). Predictive performance was comparable across biomes,
although with large variation in data availability (Figure S11).
The importance of predictor variables in the RF models was
largely consistent across months. Macroclimatic variables such as in-
coming solar radiation as well as long- term averages in air tempera-
ture and precipitation were by far the most influential explanatory
variables in the spatial models of the monthly temperature offset
(Figures S12 and S13).
We highlight that the current availability of in situ soil tempera-
ture measurements is significantly lower in the tropics (Table S5),
where our model had to extrapolate temperatures beyond the range
used to calibrate the model (Figure 5b, Figure S6b).
Finally, our comparison with a mean annual soil temperature product
derived from the coarse- resolution ERA5L topsoil temperature showed
that spatial variability, for example, driven by topographic heterogeneity,
is muc h bet ter captur ed here th an in the coar ser resol ution of the ER A5L-
based product (Figure 6c- e). Nevertheless, our predictions at the coarse
scale showed to be condensed within a 5°C range of values from the
ERA5L- predictions, for more than 95% of pixels globally. Noteworthy,
our predictions resulted in consistently cooler soil temperature predic-
tions than topsoil conditions provided by ERA5L across large areas, such
as the boreal and tropical forest biomes (Figure 6a,b). Additionally, our
models predicted lower values for SBIO1 than ERA5L in all regions with
mean annual soil temperature below 0°C, except for a few locations
around Greenland and Svalbard (Figure 6a,b).
4 |  DISCUSSION
4.1  |  Global patterns in soil temperature
We observed large spatiotemporal heterogeneity in the global offset
between soil and air temperature, often in the order of several degrees
annually and up to more than 20°C during winter months at high lati-
tudes. These values are in line with empirical data from regional studies
(Lembrechts et al., 2019; Obu et al., 2019; Zhang et al., 2018). Both
annual and monthly offsets showed clear discrepancies between cold
and dry versus warm and wet biomes. The modelled monthly offsets
22 
|
    LE MBRECHTS E T aL.
covaried strongly negatively with both long- term averages in free- air
temperature and solar radiation, linking to the well- known decoupling
of soil from air temperature due to snow (for cold extremes in cold and
cool biomes) (Grundstein et al., 2005). However, the secondary impor-
tance of variables related to precipitation and soil structure hints to the
additional distinction between wet and dry biomes at the warm end of
the temperature gradient. There, buffering due to shading, evapotran-
spiration and the specific heat of water (mostly against warm extremes
in warm and wet biomes) results in cooler soil temperature (De Frenne
et al., 2013; Geiger, 1950; Grünberg et al., 2020; Grundstein et al.,
2005; Hennon et al., 2010; Wang & Dickinson, 2012), while such buff-
ering is not as strong in warm and dry biomes due to the lower water
availability (Greiser et al., 2018; Wang & Dickinson, 2012; Zhou et al.,
2021). As such, these results highlight strong macroclimatic impacts
on the soil microclimate across the globe (see also De Frenne et al.,
2019), yet with soil temperature importantly non- linearly related to air
temperature at the global scale. This confirms that the latter is not suf-
ficient as a proxy for temperature conditions near or in the soil. With
our soil- specific global bioclimatic products, we have provided the
means to correct for these important region- specific, non- linear dif-
ferences between soil and air temperature at an unprecedented spatial
resolution.
4.2  |  Drivers of the temperature offset
Our empirical modelling approach enabled us to accurately map
global patterns in soil temperature. In doing so we did not aim to
FIGURE 6 The mean annual soil temperature (SBIO1, 1 x 1 km resolution) modelled here is consistently cooler than ERA5L (9 x 9 km)
soil temperature in forested areas. (a) Spatial representation of the difference between SBIO1 based on our model and based on ERA5L
soil temperature data. Negative values (blue colours) indicate areas where our model predicts cooler soil temperature. Dark grey areas
(Greenland and Antarctica) are excluded from our models. Asterisk in Scandinavia indicates the highlighted area in panels d to f (see below).
(b) Distribution of the difference between SBIO1 and ERA5L along the macroclimatic gradient (represented by SBIO1 itself) based on a
random subsample of 50,000 points from the map in a). Red line from a Generalized Additive Model (GAM) with k = 4. (c- e) High- resolution
zoomed panels of an area of high elevational contrast in Norway (from 66.0– 66.4°N, 15.0– 16.0°E) visualizing SBIO1 (c), ERA5L (d) and their
difference (e), to highlight the higher spatial resolution as obtained with SBIO1
   
|
  23
LEMBRECH TS ET aL.
disentangle the mechanisms governing the temperature offset:
such an endeavour would require modelling the biophysics of en-
ergy exchange at the soil surface across biomes (Kearney et al.,
2019; Maclean & Klinges, 2021; Maclean et al., 2019). Importantly,
many of the predictor variables used in our study (e.g. long- term
averages in macroclimatic conditions or solar radiation) are un-
likely to represent direct causal relationships underlying the
temperature offset, but may rather indirectly relate to many en-
suing factors that affect the functioning of ecosystems at fine
spatial scales which, in turn, feedback on local temperature off-
sets, such as energy and water balances, snow cover, wind inten-
sity and vegetation cover (De Frenne et al., 2021). For example,
while increased solar radiation itself would theoretically result in
soils warming more than the air, high solar radiation at the global
scale often coincides with high vegetation cover blocking radia-
tion input to the soil, thus correlating with relatively cooler soils
(De Frenne et al., 2021). Our results highlight, however, that the
complex relationship between microclimatic soil temperature and
macroclimatic air temperature is predictable across large spatial
extents thanks to broad scale patterns, even if this is governed
by a multitude of local- scale factors involving fine spatiotempo-
ral resolutions. Nevertheless, the predictive quality of our models
was lower in high latitude regions, where high variation in the in
situ measured offsets— likely driven by the interactions between
snow, local topography and vegetation— reduced predictive
power of the models at the 1- km2 resolution (Greiser et al., 2018;
Grünberg et al., 2020; Myers- Smith et al., 2020; Niittynen et al.,
2020; Way & Lewkowicz, 2018).
4.3  |  Implications for microclimate warming
Our results highlight clear biome- specific differences in mean annual
temperature between air and soil temperatures, as well as a signifi-
cant reduction in the spatial variation in temperature in the soil or
near the soil surface, especially in cold and cool biomes (Figure 4).
These patterns remain even despite the presence of often strongly
opposing monthly offset trends (Figure 2). The observed correlation
between long- term averages in macroclimatic conditions and the an-
nual temperature offset illustrates that soil temperature is unlikely
to warm at the same rate as air temperature when macroclimate
warms. Indeed, one degree of air temperature warming could result
in either a bigger or smaller soil temperature change, depending on
where along the macroclimatic gradient this is happening. These ef-
fects migh t be se en in col d bio me so ils most stro ngl y, as th ey not on ly
experience the largest (positive) temperature offsets and reductions
in climate range compared to air temperature (Figure 4b,c), but they
are also expected to experience the strongest magnitude of macro-
climate warming (Chen et al., 2021; Cooper, 2014; GISTEMP Team,
2021; Overland et al., 2014). As a result, mean annual temperatures
in cold climate soils can be expected to warm slower than the cor-
responding macroclimate as offsets shrink with increasing macrocli-
mate warming.
Contrastingly, predicted climate warming in hot and dry bi-
omes could be amplified in the topsoil, where we show soils to
become increasingly warmer than the air at higher tempera-
tures. Similarly, changes in precipitation regimes— and thus soil
moisture— can significantly alter the relationship between air
and soil temperature, with critical implications for soil moisture-
atmosphere feedbacks, especially in hot biomes (Zhou et al.,
2021). Indeed, as precipitation decreases, offsets could turn
more positive and soil temperatures might warm even faster than
the observed macroclimate warming. Therefore, future research
should not only use soil temperature data as provided here to
study belowground ecological processes (De Frenne et al., 2013;
Lembrechts et al., 2020), it should also urgently investigate fu-
ture scenarios of soil climate warming in light of changing air tem-
perature and precipitation, at ecologically relevant spatial and
temporal resolutions to incorporate the non- linear relationships
exposed so far (Lembrechts & Nijs, 2020).
4.4  |  Within- pixel heterogeneity
We chose to use a 1- km² resolution spatial grid to model mismatches
between soil and air temperature, aggregating all values from dif-
ferent microhabitats within the same 1- km2 grid cell (e.g. sensors in
forested versus open patches) as well as all daily and diurnal varia-
tion within a month. Additionally, we used coarse- grained free- air
temperature rather than in situ measured air temperatures. We are
aware that higher spatiotemporal resolutions would likely reveal
the importance of locally heterogeneous variables. Finer- scale fac-
tors that affect the local radiation balance and wind (e.g. topogra-
phy, snow and vegetation cover, urbanization) at the landscape to
local scales and those that directly affect neighbouring locations
(e.g. topographic shading and cold- air drainage, Ashcroft & Gollan,
2012; Lembrechts et al., 2020; Whiteman, 1982) would probably
have emerged as more important drivers at regional scales and with
higher spatiotemporal resolutions than those used here (Figure S12).
The latter is illustrated by the multi- degree Celsius difference in
mean annual temperature between forested and non- forested loca-
tions within the same biome (Table S7), as well as the lower accuracy
obtained during winter months at high latitudes, where and when
fine- scale spatial heterogeneity in snow cover and depth probably
lowers models’ predictability at the 1- km2 resolution. In situ meas-
urements were largely from areas with a representative vegetation
type, supporting the reliability of our predictions for the dominant
habitat type within a pixel. However, improved accuracy at high
latitudes will depend on the future development of high- resolution
snow depth and/or snow water equivalent estimates (Luojus et al.,
2010).
The SoilTemp database (Lembrechts et al., 2020) will facili-
tate the necessary steps towards mapping soil temperature at
higher spatiotemporal resolutions in the future, with its geo-
referenced time series of in situ measured soil and near- surface
temperature and associated metadata. Nevertheless, compared
24 
|
    LE MBRECHTS E T aL.
with existing soil temperature products such as those from
ERA5L (Copernicus Climate Change Service (C3S), 2019), we
emphasize that the increased resolution of our data products
already provides a major technical advance, even though sub-
stantial finer within- pixel variation is still lost through spatio-
temporal aggregation.
5 |  CONCLUSIONS
The spatial (biome- specific) and temporal (seasonally variable) off-
sets between air and soil temperature quantified here likely bias
predictions of current and future climate impacts on species and
ecosystems (Bergstrom et al., 2021; Cooper, 2014; Graae et al.,
2018; Kearney et al., 2009; Körner & Paulsen, 2004; Opedal et al.,
2015; Zellweger et al., 2020). Temperature in the topsoil rather than
in the air ultimately defines the distribution and performance of
most terrestrial species, as well as many ecosystem functions at or
below the soil surface (Gottschall et al., 2019; Hursh et al., 2017;
Pleim & Gilliam, 2009; Portillo- Estrada et al., 2016). As many eco-
system functions are highly correlated with temperature (yet often
non- lineary, Johnston et al., 2021), soil temperature rather than air
temperature should in those instances be the preferred predictor for
estimating their rates and temperature thresholds (Coûteaux et al.,
1995; Rosenberg et al., 1990; Schimel et al., 1996). Correcting for
the non- linear relationship between air and soil temperature identi-
fied here is, thus, vital for all fields investigating abiotic and biotic
processes relating to terrestrial environments (White et al., 2020).
Indeed, soil temperature, macroclimate and land- use change will in-
teract to define the future climate as experienced by organisms, and
high- resolution soil temperature data are needed to tackle current
and future challenges.
By making our global soil temperature maps and the underlying
monthly offset data openly available, we offer gridded soil tempera-
ture data for climate research, ecology, agronomy and other life and
environmental sciences. Future research has the important task of
further improving the spatial and temporal resolution of global micro-
climate products as microclimate operates at much higher temporal
resolutions, with temporal variation over hours, days, seasons and
years (Bütikofer et al., 2020; Potter et al., 2013), as well as to confirm
accuracy of predictions in undersampled regions in the underlying
maps (Lembrechts et al., 2021). However, we are convinced that the
maps presented here bring us one step closer to having accessible
climate data exactly where it matters most for many terrestrial or-
ganisms (Ashcroft et al., 2014; Kearney & Porter, 2009; Lembrechts
& Lenoir, 2019; Niittynen & Luoto, 2018; Pincebourde et al., 2016).
We, never theless, highlight that there is still a long way to go towards
global soil microclimate data with an optimal spatiotemporal resolu-
tion. We, therefore, urge all scientists to submit their microclimate
time series to the SoilTemp database to fill data gaps and help to in-
crease the spatial resolution until it matches with the scale at which
ecological processes take place (Bütikofer et al., 2020; Lembrechts
et al., 2020).
ACKNOWLEDGEMENTS
JJL received funding from the Research Foundation Flanders
(grant nr. 12P1819N). The project received funding from the
Research Foundation Flanders (grants nrs, G018919N,
W0 01919N) . JVDH and T WC receive d funding from DOB Ecolo gy.
JA received funding from the University of Helsinki, Faculty of
Science (MICROCLIM, grant nr. 7510145) and Academy of Finland
Flagship (grant no. 337552). PDF, CM and PV received funding
from the European Research Council (ERC) under the European
Union’s Horizon 2020 research and innovation programme (ERC
Starting Grant FORMICA 757833). JK received funding from the
Arctic Interactions at the University of Oulu and Academy of
Finland (318930, Profi 4), Maaja vesitekniikan tuki ry., Tiina and
Antti Herlin Foundation, Nordenskiöld Samfundet and Societas
pro Fauna et Flora Fennica. MK received funding from the Czech
Science Foundation (grant nr. 20- 28119S) and the Czech Academy
of Sciences (grant nr. RVO 67985939). TWC received funding from
National Geographic Society grant no. 9480 - 14 and WW- 240 R- 17.
MA received funding from CISSC (program ICRP (grant nr:2397)
and INSF (grant nr: 96005914). The Royal Botanic Garden
Edinburgh is supported by the Scottish Government’s Rural and
Environment Science and Analytical Services Division. JMA re-
ceived funding from the Funding Org. Qatar Petroleum (grant nr.
QUEX- CAS- QP- RD- 18/19). JMA received funding from the
European Union’s Horizon 2020 research and innovation program
(grant no. 678841) and from the Swiss National Science Foundation
(grant no. 31003A _176044). JA was supported by research grants
LTAUSA19137 (program INTER- EXCELLENCE, subprogram
INTER- ACTION) provided by Czech Ministry of Education, Youth
and Sports and 20- 05840Y of the Czech Science Foundation. AA
was sup ported by the Ministr y of Science and Higher Education of
the Russian Federation (grant FSRZ- 2020- 0014). SN, UAT, JJA,
and JvO received funding from the Independent Research Fund
Denmark (7027- 00133B). LvdB, KT, MYB and RC acknowledge
funding from the German Research Foundation within the Priority
Program SPP- 1803 ‘EarthShape: Earth Surface Shaping by Biota’
(grant TI 338/14- 1&2 and BA 3843/6- 1). PB was supported by
grant project VEGA of the Ministry of Education of the Slovak
Republic and the Slovak Academy of Sciences No. 2/0132/18.
Forest Research received funding from the Forestry Commission
(climate change research programme). JCB acknowledges the sup-
port of Universidad Javeriana. JLBA received funding from the
Dirección General de Cambio Climático del Gobierno de Aragón;
JLBA acknowledges fieldwork assistance by Ana Acín, the Ordesa
y Monte Perdido National Park, and the Servicio de Medio
Ambiente de Soria de la Junta de Castilla y León. RGB and MPB
received funding from BECC - Biodiversity and Ecosystem ser-
vices in a Changing Climate. MPB received funding from The
European Union’s Horizon 2020 research and innovation program
under the Marie Skłodowska- Curie Grant Agreement No. 657627
and The Swedish Research Council FORMAS – future research
leaders No. 2016- 01187. JB received funding from the Czech
Academy of Sciences (grant nr. RVO 67985939). NB received
   
|
  25
LEMBRECH TS ET aL.
funding from the SNF (grant numbers 40FA40_154245,
20FI21_148992, 20FI20_173691, 407340_172433) and from the
EU (contract no. 774124). ICOS EU research infrastructure. EU
FP7 NitroEurope. EU FP7 ECLAIRE. The authors from Biological
Dynamics of Forest Fragments Project, PDBFF, Instituto Nacional
de Pesquisas da Amazônia, Brazil were supported by the MCTI/
CNPq/FNDCT – Ação Transversal n°68/2013 – Programa de
Grande Escala da Biosfera- Atmosfera na Amazônia – LBA; Project
‘Como as florestas da Amazônia Central respondem às variações
climáticas? Efeitos sobre dinâmica florestal e sinergia com a frag-
mentação florestal’. This is the study 829 of the BDFFP Technical
Series. to The EUCFLUX Cooperative Research Program and
Forest Science and Research Institute- IPEF. NC acknowledges
funding by Stelvio National Park. JC was funded by the Spanish
government grant CGL2016- 78093- R. ANID- FONDECYT
1181745 AND INSTITUTO ANTARTICO CHILENO (INACH FR-
0418). SC received funding from the German Research Foundation
(grant no. DFG– FZT 118, 202548816). The National Science
Foundation, Poland (grant no. UMO- 2017/27/B/ST10/02228),
within the framework of the ‘Carbon dioxide uptake potential of
sphagnum peatlands in the context of atmospheric optical param-
eters and climate changes’ (KUSCO2) project. SLC received fund-
ing from the South African National Research Foundation and the
Australian Research Council. FM, MČ, KU and MU received fund-
ing from Slovak Research and Development Agency (no. APVV-
19- 0319). Instituto Antartico Chileno (INACH_RT- 48_16),
Iniciativa Científica Milenio Núcleo Milenio de Salmónidos
Invasores INVASAL, Institute of Ecology and Biodiversity (IEB),
CONICYT PIA APOYO CCTE AFB170008. PC is supported by
NERC core funding to the BAS ‘Biodiversity, Evolution and
Adaptation Team. EJC received funding from the Norwegian
Research Council (grant number 230970). GND was supported by
NERC E3 doctoral training partnership grant (NE/L002558/1) at
the University of Edinburgh and the Carnegie Trust for the
Universities of Scotland. Monitoring stations on Livingston Island,
Antarctica, were funded by different research projects of the
Gobern of Spain (PERMAPLANET CTM2009- 10165- E;
ANTARPERMA CTM2011- 15565- E; PERMASNOW CTM2014-
52021- R), and the PERMATHERMAL arrangement between the
University of Alcalá and the Spanish Polar Committee. GN re-
ceived funding from the Autonomous Province of Bolzano (ITA).
The infrastructure, part of the UK Environmental Change Network,
was funded historically in part by ScotNature and NERC National
Capability LTS- S: UK- SCAPE; NE/R016429/1). JD was supported
by the Czech Science Foundation (GA17- 19376S) and MSMT
(LTAUSA18007). ED received funding from the Kempe Foundation
(JCK- 1112 and JCK- 1822). The infrastructure was supported by
the Minist r y of Educatio n, Youth and Spor ts of the Cze ch Republic
within the Nati onal Sus t ainabili t y Pro gra mme I (N PU I), gra nt num-
ber LO1415 and by the project for national infrastructure support
CzeCOS/ICOS Reg. No. LM2015061. NE received funding from
the German Research Foundation (DFG– FZT 118, 202548816).
BE received funding from the GLORIA- EU project no
EVK2- CT2000- 00056, the Autonomous Province of Bolzano
(I TA ), fro m the Tirol er Wiss ens cha ftsfonds an d from the Univ e r sit y
of Innsbruck. RME was supported by funding to the SAFE Project
from the Sime Darby Foundation. OF received funding from the
German Research Foundation (DFG– FZT 118, 202548816). EFP
was supported by the Jardín Botánico Atlántico (SV- 20- GIJON-
JB A). MF was fu nded by th e German Fe deral Minis try of Ed ucati on
and Research (BMBF) in the context of The Future Okavango
(Grant No. 01LL0912) and SASSCAL (01LG1201M; 01LG1201N)
pr ojects. EF L recei ved fu nding fr o m ANID PIA / BASA L FB2100 0 6.
RAG received funding from Fondecyt 11170516, CONICYT PIA
AFB170008 and ANID PIA / BASAL FB210006. MBG received
funding from National Parks (DYNBIO, #1656/2015) and The
Spanish Research Agenc y ( VULBIMON, #CGL2017- 90 040- R <