Annual sums of carbon dioxide exchange over a heterogeneous urban landscape through machine learning based gap-filling

Abstract A small, but growing, number of flux towers in urban environments measure surface–atmospheric exchanges of carbon dioxide by the eddy covariance method. As in all eddy covariance studies, obtaining annual sums of urban CO 2 exchange requires imputation of data gaps due to low turbulence and non-stationary conditions, adverse weather, and instrument failures. Gap-filling approaches that are widely used for measurements from towers in natural vegetation are based on light and temperature response models. However, they do not account for key features of the urban environment including tower footprint heterogeneity and localized CO 2 sources. Here, we present a novel gap-filling modeling framework that uses machine learning to select explanatory variables, such as continuous traffic counts and temporal variables, and then constrains models separately for spatially classified subsets of the data. We applied the modeling framework to a three year time series of measurements from a tall broadcast tower in a suburban neighborhood of Minneapolis-Saint Paul, Minnesota, USA. The gap-filling performance was similar to that reported for natural measurement sites, explaining 64% to 88% of the variability in the fluxes. Simulated carbon budgets were in good agreement with an ecophysiological bottom-up study at the same site. Total annual carbon dioxide flux sums for the tower site ranged from 1064 to 1382 g C m −2  yr −1 , across different years and different gap-filling methods. Bias errors of annual sums resulting from gap-filling did not exceed 18 g C m −2  yr −1 and random uncertainties did not exceed ±44 g C m −2  yr −1 (or ±3.8% of the annual flux). Regardless of the gap-filling method used, the year-to-year differences in carbon exchange at this site were small. In contrast, the modeled annual sums of CO 2 exchange differed by a factor of two depending on wind direction. This indicated that the modeled time series captured the spatial variability in both the biogenic and anthropogenic CO 2 sources and sinks in a reproducible way. The gap-filling approach developed here may also be useful for inhomogeneous sites other than urban areas, such as logged forests or ecosystems under disturbance from fire or pests.

[1]  Radford M. Neal.  Pattern Recognition and Machine Learning , 2007, Technometrics.

[2]  T. Oke,et al.  Local Climate Zones for Urban Temperature Studies , 2012 .

[3]  Eytan Domany,et al.  Models of Neural Networks I , 1991 .

[4]  B O B B,et al.  Estimating diurnal to annual ecosystem parameters by synthesis of a carbon flux model with eddy covariance net ecosystem exchange observations , 2005 .

[5]  6 Bayesian Methods for Backpropagation Networks , 2022 .

[6]  T. Oke,et al.  Wind, temperature and stability conditions in an east-west oriented urban canyon , 1988 .

[7]  Lasse Møller-Jensen,et al.  Towards a spatial CO2 budget of a metropolitan region based on textural image classification and flux measurements , 2003 .

[8]  Andres Schmidt,et al.  Gap Filling and Quality Assessment of CO2 and Water Vapour Fluxes above an Urban Area with Radial Basis Function Neural Networks , 2008 .

[9]  C. Grimmond,et al.  Identification of Micro-scale Anthropogenic CO2, heat and moisture sources – Processing eddy covariance fluxes for a dense urban environment , 2012 .

[10]  George Cybenko,et al.  Approximation by superpositions of a sigmoidal function , 1989, Math. Control. Signals Syst..

[11]  Lena Weissert,et al.  A review of the current progress in quantifying the potential of urban forests to mitigate urban CO2 emissions , 2014 .

[12]  K. Taylor.  Summarizing multiple aspects of model performance in a single diagram , 2001 .

[13]  Kevin N. Gurney,et al.  An introduction to neural networks , 2018 .

[14]  Heekuck Oh,et al.  Neural Networks for Pattern Recognition , 1993, Adv. Comput..

[15]  J. William Munger,et al.  Measurements of carbon sequestration by long‐term eddy covariance: methods and a critical evaluation of accuracy , 1996 .

[16]  D. Hollinger,et al.  Uncertainty in eddy covariance measurements and its application to physiological models. , 2005, Tree physiology.

[17]  Volker C. Radeloff,et al.  Rural and Suburban Sprawl in the U.S. Midwest from 1940 to 2000 and Its Relation to Forest Fragmentation , 2005 .

[18]  D. Baldocchi.  ‘Breathing’ of the terrestrial biosphere: lessons learned from a global network of carbon dioxide flux measurement systems , 2008 .

[19]  Gene H. Golub,et al.  Matrix computations , 1983 .

[20]  Christopher M. Bishop,et al.  Neural networks and machine learning , 1998 .

[21]  Geoffrey E. Hinton,et al.  Learning internal representations by error propagation , 1986 .

[22]  A. Dolman,et al.  Inverse carbon dioxide flux estimates for the Netherlands , 2012 .

[23]  Natascha Kljun,et al.  Interpreting CO2 Fluxes Over a Suburban Lawn: The Influence of Traffic Emissions , 2011 .

[24]  Thomas Foken,et al.  Documentation and Instruction Manual of the Eddy-Covariance Software Package TK3 (update) , 2011 .

[25]  Timo Vesala,et al.  Fraction of natural area as main predictor of net CO2 emissions from cities , 2012 .

[26]  David S. Broomhead,et al.  Multivariable Functional Interpolation and Adaptive Networks , 1988, Complex Syst..

[27]  Matthias Roth,et al.  Cities as Net Sources of CO 2 : Review of Atmospheric CO2 Exchange in Urban Environments Measured by Eddy Covariance Technique , 2010 .

[28]  Timothy R. Oke,et al.  Flux and turbulence measurements at a densely built‐up site in Marseille: Heat, mass (water and carbon dioxide), and momentum , 2004 .

[29]  Christian Feigenwinter,et al.  Eddy Covariance Measurements Over Urban Areas , 2012 .

[30]  M. J. Narasimha,et al.  On the Computation of the Discrete Cosine Transform , 1978, IEEE Trans. Commun..

[31]  Thomas Foken,et al.  Quality control of CarboEurope flux data – Part 2: Inter-comparison of eddy-covariance software , 2007 .

[32]  C. Grimmond,et al.  Five years of carbon dioxide fluxes measurements in a highly vegetated suburban area , 2011 .

[33]  Eric A. Davidson,et al.  Spatial and temporal variability in forest–atmosphere CO2 exchange , 2004 .

[34]  Fei Yuan,et al.  Trends in Extreme Temperatures in Relation to Urbanization in the Twin Cities Metropolitan Area, Minnesota , 2009 .

[35]  A. Christen,et al.  Spatial source attribution of measured urban eddy covariance CO2 fluxes , 2015, Theoretical and Applied Climatology.

[36]  N. Coops,et al.  A systems approach to carbon cycling and emissions modeling at an urban neighborhood scale , 2013 .

[37]  D. Satterthwaite.  Cities' contribution to global warming: notes on the allocation of greenhouse gas emissions , 2008 .

[38]  Frans T. M. Nieuwstadt,et al.  Temperature measurement with a sonic anemometer and its application to heat and moisture fluxes , 1983 .

[39]  Ü. Rannik,et al.  Gap filling strategies for defensible annual sums of net ecosystem exchange , 2001 .

[40]  T. Foken,et al.  Tools for quality assessment of surface-based flux measurements , 1996 .

[41]  Lawrence E. Band,et al.  Beyond Urban Legends: An Emerging Framework of Urban Ecology, as Illustrated by the Baltimore Ecosystem Study , 2008 .

[42]  Min Liu,et al.  Uncertainty analysis of CO2 flux components in subtropical evergreen coniferous plantation , 2009 .

[43]  M. Heimann,et al.  Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes , 2007 .

[44]  H. Grassl,et al.  A search for large‐scale effects of ship emissions on clouds and radiation in satellite data , 2011 .

[45]  J. McFadden,et al.  Continuous measurements of net CO2 exchange by vegetation and soils in a suburban landscape , 2012 .

[46]  John Moncrieff,et al.  The propagation of errors in long‐term measurements of land‐atmosphere fluxes of carbon and water , 1996 .

[47]  Joseph P. McFadden,et al.  Seasonal contributions of vegetation types to suburban evapotranspiration , 2011 .

[48]  R. Valentini,et al.  A new assessment of European forests carbon exchanges by eddy fluxes and artificial neural network spatialization , 2003 .

[49]  T. Vesala,et al.  Towards a standardized processing of Net Ecosystem Exchange measured with eddy covariance technique: algorithms and uncertainty estimation , 2006 .

[50]  K. Seto,et al.  Global forecasts of urban expansion to 2030 and direct impacts on biodiversity and carbon pools , 2012, Proceedings of the National Academy of Sciences.

[51]  C. J. Moore.  Frequency response corrections for eddy correlation systems , 1986 .

[52]  Geoffrey E. Hinton,et al.  Evaluation of Gaussian processes and other methods for non-linear regression , 1997 .

[53]  Geoffrey E. Hinton,et al.  Learning representations by back-propagating errors , 1986, Nature.

[54]  E. Davidson,et al.  Spatial and temporal variability in forest–atmosphere CO2 exchange , 2004 .

[55]  Magnus R. Hestenes,et al.  Conjugate Direction Methods in Optimization , 1980 .

[56]  E. Nikinmaa,et al.  Seasonal and annual variation of carbon dioxide surface fluxes in Helsinki, Finland, in 2006–2010 , 2012 .

[57]  K. Davis,et al.  A multi-site analysis of random error in tower-based measurements of carbon and energy fluxes , 2006 .

[58]  D. Nowak,et al.  Local-scale fluxes of carbon dioxide in urban environments: methodological challenges and results from Chicago. , 2002, Environmental pollution.

[59]  Dario Papale,et al.  Eddy Covariance: A Practical Guide to Measurement and Data Analysis , 2012 .

[60]  G. Matheron.  Principles of geostatistics , 1963 .

[61]  T. A. Black,et al.  Sensitivity and uncertainty of the carbon balance of a Pacific Northwest Douglas-fir forest during an El Niño/La Niña cycle , 2004 .

[62]  Radford M. Neal.  Monte Carlo Implementation of Gaussian Process Models for Bayesian Regression and Classification , 1997, physics/9701026.

[63]  Ü. Rannik,et al.  Estimates of the annual net carbon and water exchange of forests: the EUROFLUX methodology , 2000 .

[64]  T. Vesala,et al.  On the separation of net ecosystem exchange into assimilation and ecosystem respiration: review and improved algorithm , 2005 .

[65]  E. Pardyjak,et al.  Toward understanding the behavior of carbon dioxide and surface energy fluxes in the urbanized semi-arid Salt Lake Valley, Utah, USA , 2011 .

[66]  David Hinkley,et al.  Bootstrap Methods: Another Look at the Jackknife , 2008 .

[67]  M. Reichstein,et al.  Random errors in carbon and water vapor fluxes assessed with Gaussian processes , 2013 .

[68]  rer. nat. Moffat,et al.  A new methodology to interpret high resolution measurements of net carbon fluxes between terrestrial ecosystems and the atmosphere , 2012 .

[69]  Carl E. Rasmussen,et al.  Gaussian Processes for Machine Learning (GPML) Toolbox , 2010, J. Mach. Learn. Res..

[70]  H. Schmid,et al.  A Simple Parameterisation for Flux Footprint Predictions , 2004 .

[71]  Wilhelm Kuttler,et al.  Carbon dioxide fluxes over an urban park area , 2010 .

[72]  D. Broomhead,et al.  Radial Basis Functions, Multi-Variable Functional Interpolation and Adaptive Networks , 1988 .

[73]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.

[74]  N. Coops,et al.  Validation of modeled carbon-dioxide emissions from an urban neighborhood with direct eddy-covariance measurements , 2011 .

[75]  E. K. Webb,et al.  Correction of flux measurements for density effects due to heat and water vapour transfer , 1980 .

[76]  Arain,et al.  Use of change-point detection for friction–velocity threshold evaluation in eddy-covariance studies , 2013 .

[77]  Eero Nikinmaa,et al.  Surface–atmosphere interactions over complex urban terrain in Helsinki, Finland , 2008 .