Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms

Gianluca Tramontana, Martin Jung, Christopher R Schwalm, Kazuhito Ichii, Gustau Camps-Valls, Botond Ráduly, Markus Reichstein, M. Altaf Arain, Alessandro Cescatti, Gerard Kiely, Lutz Merbold, Penelope Serrano-Ortiz, Sven Sickert, Sebastian Wolf, Dario Papale

Research output: Contribution to journalArticle

82 Citations (Scopus)

Abstract

Spatio-temporal fields of land-atmosphere fluxes derived from data-driven models can complement simulations by process-based land surface models. While a number of strategies for empirical models with eddy-covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we performed a cross-validation experiment for predicting carbon dioxide, latent heat, sensible heat and net radiation fluxes across different ecosystem types with 11 machine learning (ML) methods from four different classes (kernel methods, neural networks, tree methods, and regression splines). We applied two complementary setups: (1) 8-day average fluxes based on remotely sensed data and (2) daily mean fluxes based on meteorological data and a mean seasonal cycle of remotely sensed variables. The patterns of predictions from different ML and experimental setups were highly consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R2 < 0.5), ecosystem respiration (R2 > 0.6), gross primary production (R2 > 0.7), latent heat (R2 > 0.7), sensible heat (R2 > 0.7), and net radiation (R2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well (R2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted (R2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). The evaluated large ensemble of ML-based models will be the basis of new global flux products.

Original languageEnglish (US)
Pages (from-to)4291-4313
Number of pages23
JournalBiogeosciences
Volume13
Issue number14
DOIs
StatePublished - Jul 29 2016
Externally publishedYes

Fingerprint

energy flux
artificial intelligence
carbon dioxide
energy
heat
net radiation
net ecosystem exchange
methodology
eddy covariance
temperate zones
meteorological data
neural networks
primary production
primary productivity
land surface
tropics
complement
method
climate
prediction

ASJC Scopus subject areas

  • Earth-Surface Processes
  • Ecology, Evolution, Behavior and Systematics

Cite this

Tramontana, G., Jung, M., Schwalm, C. R., Ichii, K., Camps-Valls, G., Ráduly, B., ... Papale, D. (2016). Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms. Biogeosciences, 13(14), 4291-4313. https://doi.org/10.5194/bg-13-4291-2016

Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms. / Tramontana, Gianluca; Jung, Martin; Schwalm, Christopher R; Ichii, Kazuhito; Camps-Valls, Gustau; Ráduly, Botond; Reichstein, Markus; Arain, M. Altaf; Cescatti, Alessandro; Kiely, Gerard; Merbold, Lutz; Serrano-Ortiz, Penelope; Sickert, Sven; Wolf, Sebastian; Papale, Dario.

In: Biogeosciences, Vol. 13, No. 14, 29.07.2016, p. 4291-4313.

Research output: Contribution to journalArticle

Tramontana, G, Jung, M, Schwalm, CR, Ichii, K, Camps-Valls, G, Ráduly, B, Reichstein, M, Arain, MA, Cescatti, A, Kiely, G, Merbold, L, Serrano-Ortiz, P, Sickert, S, Wolf, S & Papale, D 2016, 'Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms', Biogeosciences, vol. 13, no. 14, pp. 4291-4313. https://doi.org/10.5194/bg-13-4291-2016
Tramontana, Gianluca ; Jung, Martin ; Schwalm, Christopher R ; Ichii, Kazuhito ; Camps-Valls, Gustau ; Ráduly, Botond ; Reichstein, Markus ; Arain, M. Altaf ; Cescatti, Alessandro ; Kiely, Gerard ; Merbold, Lutz ; Serrano-Ortiz, Penelope ; Sickert, Sven ; Wolf, Sebastian ; Papale, Dario. / Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms. In: Biogeosciences. 2016 ; Vol. 13, No. 14. pp. 4291-4313.
@article{77c22f335de846a4adcfc49b9164791e,
title = "Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms",
abstract = "Spatio-temporal fields of land-atmosphere fluxes derived from data-driven models can complement simulations by process-based land surface models. While a number of strategies for empirical models with eddy-covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we performed a cross-validation experiment for predicting carbon dioxide, latent heat, sensible heat and net radiation fluxes across different ecosystem types with 11 machine learning (ML) methods from four different classes (kernel methods, neural networks, tree methods, and regression splines). We applied two complementary setups: (1) 8-day average fluxes based on remotely sensed data and (2) daily mean fluxes based on meteorological data and a mean seasonal cycle of remotely sensed variables. The patterns of predictions from different ML and experimental setups were highly consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R2 < 0.5), ecosystem respiration (R2 > 0.6), gross primary production (R2 > 0.7), latent heat (R2 > 0.7), sensible heat (R2 > 0.7), and net radiation (R2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well (R2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted (R2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). The evaluated large ensemble of ML-based models will be the basis of new global flux products.",
author = "Gianluca Tramontana and Martin Jung and Schwalm, {Christopher R} and Kazuhito Ichii and Gustau Camps-Valls and Botond R{\'a}duly and Markus Reichstein and Arain, {M. Altaf} and Alessandro Cescatti and Gerard Kiely and Lutz Merbold and Penelope Serrano-Ortiz and Sven Sickert and Sebastian Wolf and Dario Papale",
year = "2016",
month = "7",
day = "29",
doi = "10.5194/bg-13-4291-2016",
language = "English (US)",
volume = "13",
pages = "4291--4313",
journal = "Biogeosciences",
issn = "1726-4170",
publisher = "European Geosciences Union",
number = "14",

}

TY - JOUR

T1 - Predicting carbon dioxide and energy fluxes across global FLUXNET sites with regression algorithms

AU - Tramontana, Gianluca

AU - Jung, Martin

AU - Schwalm, Christopher R

AU - Ichii, Kazuhito

AU - Camps-Valls, Gustau

AU - Ráduly, Botond

AU - Reichstein, Markus

AU - Arain, M. Altaf

AU - Cescatti, Alessandro

AU - Kiely, Gerard

AU - Merbold, Lutz

AU - Serrano-Ortiz, Penelope

AU - Sickert, Sven

AU - Wolf, Sebastian

AU - Papale, Dario

PY - 2016/7/29

Y1 - 2016/7/29

N2 - Spatio-temporal fields of land-atmosphere fluxes derived from data-driven models can complement simulations by process-based land surface models. While a number of strategies for empirical models with eddy-covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we performed a cross-validation experiment for predicting carbon dioxide, latent heat, sensible heat and net radiation fluxes across different ecosystem types with 11 machine learning (ML) methods from four different classes (kernel methods, neural networks, tree methods, and regression splines). We applied two complementary setups: (1) 8-day average fluxes based on remotely sensed data and (2) daily mean fluxes based on meteorological data and a mean seasonal cycle of remotely sensed variables. The patterns of predictions from different ML and experimental setups were highly consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R2 < 0.5), ecosystem respiration (R2 > 0.6), gross primary production (R2 > 0.7), latent heat (R2 > 0.7), sensible heat (R2 > 0.7), and net radiation (R2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well (R2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted (R2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). The evaluated large ensemble of ML-based models will be the basis of new global flux products.

AB - Spatio-temporal fields of land-atmosphere fluxes derived from data-driven models can complement simulations by process-based land surface models. While a number of strategies for empirical models with eddy-covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we performed a cross-validation experiment for predicting carbon dioxide, latent heat, sensible heat and net radiation fluxes across different ecosystem types with 11 machine learning (ML) methods from four different classes (kernel methods, neural networks, tree methods, and regression splines). We applied two complementary setups: (1) 8-day average fluxes based on remotely sensed data and (2) daily mean fluxes based on meteorological data and a mean seasonal cycle of remotely sensed variables. The patterns of predictions from different ML and experimental setups were highly consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R2 < 0.5), ecosystem respiration (R2 > 0.6), gross primary production (R2 > 0.7), latent heat (R2 > 0.7), sensible heat (R2 > 0.7), and net radiation (R2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well (R2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted (R2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). The evaluated large ensemble of ML-based models will be the basis of new global flux products.

UR - http://www.scopus.com/inward/record.url?scp=84980351826&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84980351826&partnerID=8YFLogxK

U2 - 10.5194/bg-13-4291-2016

DO - 10.5194/bg-13-4291-2016

M3 - Article

VL - 13

SP - 4291

EP - 4313

JO - Biogeosciences

JF - Biogeosciences

SN - 1726-4170

IS - 14

ER -