Inferring management and predicting sub-field scale C dynamics in UK grasslands using biogeochemical modelling and satellite-derived leaf area data

https://doi.org/10.1016/j.agrformet.2021.108466Get rights and content

Highlights

  • Earth observation data and process modelling are combined to estimate grassland C dynamics.

  • The model-data fusion algorithm infers grazing and cutting from leaf area index data.

  • The algorithm was implemented at 3 managed grasslands in England for 2015–2018.

  • 87.5% of harvests were identified and 83% of measured yields were simulated accurately.

  • The in-situ estimated and simulated grazed biomass had a r=0.8.

Abstract

Grasslands, natural and managed, cover a large part of the Earth’s surface and play an important role in the global carbon (C) cycle. Human management strongly affects grassland C budgets through grass cutting and removal, varied grazing intensities, and organic matter additions. Thus managed grassland C cycles are highly heterogeneous and challenging to quantify. In this study, we combine a process-based model of the grassland C cycle, validated against field data on C fluxes and pools, with satellite-derived data (Proba-V and Sentinel-2) on leaf area index (LAI) in order to quantify field-scale grassland productivity and C dynamics under climatic and management conditions typical of northwest Europe. Input data on the weekly vegetation canopy anomaly (estimated from Proba-V LAI) and meteorology are used to drive the grassland C model (DALEC-Grass) that is integrated into a Bayesian model-data fusion (MDF) framework. The novelty of the MDF algorithm is that it infers weekly livestock grazing and grass cutting events based on expected canopy growth estimated by the model, and constrained by LAI observations (estimated from Sentinel-2). The MDF approach also resolves observational, parametric, and input uncertainties on C cycling estimates. We analysed four years (2015–2018) of C dynamics at three variably-managed fields of the Rothamsted Research North Wyke Farm Platform (UK). Compared against independent field data, the MDF was able to (i) identify 87.5% of the harvest events that occurred, (ii) accurately predict the annual yields in 83% of the identified harvest years and (iii) reproduce the observed grazing intensity in each field (r=0.8, overlap = 90%). We demonstrate that the fusion of process modelling with earth observations is an effective method for monitoring biomass removals and quantifying management impacts on field-scale C balance, without the need for frequent and laborious ground measurements. This approach can support the delivery of more robust national greenhouse gas (GHG) accounting that takes account of grassland vegetation management.

Introduction

Grasslands cover more than 20% of northwest Europe (Ireland, United Kingdom, France, Netherlands, Belgium, Luxembourg) and represent around 50% of the region’s agricultural land (Smit et al., 2008). In the United Kingdom (UK), grasslands are the dominant type of land use (>60% of agricultural area), are part of the rural landscape and life, support a large livestock industry and are a critical, direct and indirect, agricultural greenhouse gas (GHG) contributor (DEFRA, 2020, Qi, Holland, Taylor, Richter, 2018). Grasslands in the UK are dominated by C3 plants whose physiology, combined with generally cool and wet climatic conditions, leads to relatively slow litter decomposition rates, which are beneficial for C accumulation in soils and highlight their potential for climate change mitigation through C sequestration (Gibson, 2010, Kipling, Virkajärvi, Breitsameter, Curnel, De Swaef, Gustavsson, Hennart, Höglind, Järvenranta, Minet, Nendel, Persson, Picon-Cochard, Rolinski, Sandars, Scollan, Sebek, Seddaiu, Topp, Twardy, Van Middelkoop, Wu, Bellocchi, 2016, Vertès, Delaby, Klumpp, Bloor, 2018).

Grassland management intensity, in the form of livestock grazing and grass cutting, defines a vegetation C stock that is much more dynamic than the soil C stock (Conant, Cerri, Osborne, Paustian, 2017, Erb, Kastner, Plutzar, Bais, Carvalhais, Fetzel, Gingrich, Haberl, Lauk, Niedertscheider, Pongratz, Thurner, Luyssaert, 2018, Soussana, Lemaire, 2014). Vegetation plays a key role in grassland C cycling as it supplies inputs to the soil C pool directly through the production of litter and indirectly through livestock excrement. Grassland biomass productivity and management are tightly coupled. The application of inputs, including fertilisers and lime, and the combination of sward composition, soil properties and climatic conditions control the productivity of a grassland (Qi et al., 2017). The timing and yield of grass harvests along with the timing and intensity of livestock grazing determine the final biomass production and the net C budget of a grassland.

The quantification of the C balance of managed grasslands across time and space is an important research area, particularly when considering the C-neutrality ambitions and Nationally Determined Contributions (NDC) of developed countries following the Paris agreement (Bell, Cloy, Topp, Ball, Bagnall, Rees, Chadwick, 2016, Committee on Climate Change, 2019). In this context, in 2019 the government of the UK has set a legally binding target of net zero GHG emissions by 2050 (UK Government, 2019). Existing approaches to quantifying the productivity and C budget of managed grasslands include purely statistical modelling tools, built on agricultural census and experimental farm data (Herrero, Havlík, Valin, Notenbaert, Rufino, Thornton, Bluemmel, Weiss, Grace, Obersteiner, 2013, Qi, Holland, Taylor, Richter, 2018, Qi, Murray, Richter, 2017, Smit, Metzger, Ewert, 2008). Nevertheless, statistical approaches do not provide an in-depth description of grassland functioning, limiting their utility for climate change and C cycling studies and for management scenario explorations. For these purposes, biogeochemical models that describe the processes controlling C accumulation and partitioning must be used. Such process models incorporate information on grassland management (i.e. grazing and cutting) and, if relevant data are available, can be implemented at any location (Chang, Ciais, Viovy, Vuichard, Sultan, Soussana, 2015, Chang, Viovy, Vuichard, Ciais, Wang, Cozic, Lardy, Graux, Klumpp, Martin, Soussana, 2013, Kipling, Virkajärvi, Breitsameter, Curnel, De Swaef, Gustavsson, Hennart, Höglind, Järvenranta, Minet, Nendel, Persson, Picon-Cochard, Rolinski, Sandars, Scollan, Sebek, Seddaiu, Topp, Twardy, Van Middelkoop, Wu, Bellocchi, 2016, Puche, Senapati, Flechard, Klumpp, Kirschbaum, Chabbi, 2019, Rolinski, Müller, Heinke, Weindl, Biewald, Bodirsky, Bondeau, Boons-Prins, Bouwman, Leffelaar, Roller, Schaphoff, Thonicke, 2018, Vuichard, Soussana, Ciais, Viovy, Ammann, Calanca, Clifton-Brown, Fuhrer, Jones, Martin, 2007). However, process models typically require local information on model parameters and exogenous drivers that can be challenging to obtain.

Identifying grassland management across time and space remains an outstanding challenge (Reinermann and Asam, 2020). Data on daily livestock density and the timings of harvests are available for a limited number of experimental farms but this information is uncertain across large areas (Chang et al., 2017). Earth observation (EO) satellite sensors, however, can provide spatially and temporally consistent data on grasslands at local to regional extents (Ali et al., 2016). Empirical models (e.g. Huang, He, Niu, 2013, Ullah, Si, Schlerf, Skidmore, Shafique, Iqbal, 2012), machine learning algorithms (e.g., Ali, Cawkwell, Dwyer, Barrett, Green, 2016, Ghosh, Behera, 2018, Wang, Wu, Deng, Tang, Wang, Sun, Shangguan, 2017) and physical-based retrieval approaches (e.g., He, Li, Yin, Nan, Bian, 2019, Punalekar, Verhoef, Quaife, Humphries, Bermingham, Reynolds, 2018) are typically used to produce estimates of biomass in forests and grasslands from EO data.

The recent availability of open-access EO data from the European Space Agency’s (ESA) Sentinel-2 dual-satellite constellation, with a spatial resolution as fine as 10 m and an average global revisit time of 5 days, has the potential to provide frequent observations of grassland vegetation at the sub-field scale. The fact that fields of managed grassland in the UK are enclosed by hedgerows and most of them cover an area between 5–8 ha highlights the importance of high resolution EO data for monitoring vegetation dynamics in UK grasslands (CEH, 2017, DEFRA, 2020, Deverell, McDonnell, Devlin, 2009, Guiomar, Godinho, Pinto-Correia, Almeida, Bartolini, Bezák, Biró, Bjørkhaug, Bojnec, Brunori, Corazzin, Czekaj, Davidova, Kania, Kristensen, Marraccini, Molnár, Niedermayr, O’Rourke, Ortiz-Miranda, Redman, Sipiläinen, Sooväli-Sepping, Šmane, Surová, Sutherland, Tcherkezova, Tisenkopfs, Tsiligiridis, Tudor, Wagner, Wästfelt, 2018, Robinson, Sutherland, 2002). In addition, for use on an operational basis the overall extent to which EO data can provide timely information on UK grassland dynamics is dependent on the frequency of available cloud-free imagery. Although there is typically an inverse relationship between temporal and spatial resolution, the use of EO-derived data at higher temporal resolutions is required for resolving fields of intensively managed grasslands in the UK (Gao, 2006).

The combined use of EO data and process modelling will produce more robust estimates of managed grassland C balance when compared to using each approach independently. This synergy of model and observational data, commonly known as model-data fusion (MDF), requires a framework in which ecosystem model parameters, fluxes and pools are adjusted accordingly to match observations (data assimilation) (Raupach et al., 2005). Probabilistic approaches for parameter optimisation and for quantifying predictive uncertainty in model-based studies are central to MDF (Ben Touhami, Bellocchi, 2015, Gottschalk, Wattenbach, Neftel, Fuhrer, Jones, Lanigan, Davis, Campbell, Soussana, Smith, 2007, Oenema, Burgers, van Keulen, van Ittersum, 2015, van Oijen, 2017, Patenaude, Milne, Van Oijen, Rowland, Hill, 2008). The dependence of MDF on process-modelling means that EO data for grasslands can be translated into estimates of C present in above and below-ground vegetation, removed via grazing or cutting, transferred to the soil and lost as CO2. In this respect, and in contrast to EO data processing methods, MDF provides a more holistic and informative approach to C balance quantification. Considering the increasing volume and quality of EO data, MDF has the potential to be used in order to monitor the C balance of managed grasslands across space. Model-data fusion has been used in studies focusing on different aspects of terrestrial ecosystem functioning in the past (Bloom, Williams, 2015, Fox, Williams, Richardson, Cameron, Gove, Quaife, Ricciuto, Reichstein, Tomelleri, Trudinger, Van Wijk, 2009, Keenan, Davidson, Moffat, Munger, Richardson, 2012, Kuppel, Peylin, Maignan, Chevallier, Kiely, Montagnani, Cescatti, 2014, Peaucelle, Bacour, Ciais, Vuichard, Kuppel, Peñuelas, Belelli Marchesini, Blanken, Buchmann, Chen, Delpierre, Desai, Dufrene, Gianelle, Gimeno-Colera, Gruening, Helfter, Hörtnagl, Ibrom, Joffre, Kato, Kolb, Law, Lindroth, Mammarella, Merbold, Minerbi, Montagnani, Šigut, Sutton, Varlagin, Vesala, Wohlfahrt, Wolf, Yakir, Viovy, 2019, Peylin, Bacour, MacBean, Leonard, Rayner, Kuppel, Koffi, Kane, Maignan, Chevallier, Ciais, Prunet, 2016, Scholze, Buchwitz, Dorigo, Guanter, Quegan, 2017, Smallman, Exbrayat, Mencuccini, Bloom, Williams, 2017, Wang, Trudinger, Enting, 2009, Xiao, Davis, Urban, Keller, 2014). However, the focus of studies that considered grasslands relied exclusively on plant functional type identification and they have not examined C cycling and the role of grassland management (Kuppel, Peylin, Maignan, Chevallier, Kiely, Montagnani, Cescatti, 2014, Peaucelle, Bacour, Ciais, Vuichard, Kuppel, Peñuelas, Belelli Marchesini, Blanken, Buchmann, Chen, Delpierre, Desai, Dufrene, Gianelle, Gimeno-Colera, Gruening, Helfter, Hörtnagl, Ibrom, Joffre, Kato, Kolb, Law, Lindroth, Mammarella, Merbold, Minerbi, Montagnani, Šigut, Sutton, Varlagin, Vesala, Wohlfahrt, Wolf, Yakir, Viovy, 2019, Peylin, Bacour, MacBean, Leonard, Rayner, Kuppel, Koffi, Kane, Maignan, Chevallier, Ciais, Prunet, 2016).

The aim of this study is to quantify C dynamics in fields of managed grassland in the UK using MDF. We use a process model of grassland C dynamics (DALEC-Grass) within a probabilistic MDF algorithm (CARDAMOM) to quantify the C impact of harvest and grazing events at three fields located on a experimental farm in Southwest England (UK) between 2015 and 2018. DALEC-Grass has been tested against an extensive set of field-measured data (chamber and tower-based) of C fluxes (including net ecosystem change, ecosystem respiration), C pools (above and below-ground biomass) and LAI collected at two variably-managed fields in the UK (Myrgiotis et al., 2020). In that study, daily livestock numbers and cutting dates were known. In the present study, the model is provided with a weekly time-series of vegetation reduction, which is derived from a high-temporal and low-spatial resolution EO-based dataset (Proba-V). These data, along with meteorological drivers, are used by DALEC-Grass to simulate weekly grass biomass removals via harvest and grazing. Temporally infrequent, but high spatial resolution, leaf area index (LAI) time series derived from EO-based data (Sentinel-2) are assimilated within CARDAMOM to (1) infer the simulated management from the vegetation reduction inputs according to biophysical and biogeochemical rules and (2) refine the distribution of DALEC-Grass parameters. The probabilistic MDF approach allows us to consider (i) the temporal uncertainties around the EO-derived vegetation reduction drivers, (ii) the model’s parametric uncertainty and (iii) the uncertainty around the assimilated LAI EO data. The combined impact of these uncertainties on MDF-estimates is quantified.

We compare MDF predictions to independent ground-based data on livestock density and harvest yields in order to assess the effectiveness of DALEC-Grass and CARDAMOM. The MDF-estimated net C balance and its different components are presented and discussed. Since livestock density is a key element of a grassland’s economic, ecological and environmental performance, we conducted a sensitivity analysis to assess how it affects the simulated C dynamics. As far as we are aware this study is the first example of a validated analysis of biomass removals and of C cycling in managed grasslands at field scale using MDF and relying exclusively on EO data (i.e. no in-situ data used). Based on the obtained results we discuss how monitoring the grassland C balance across time and space could be achieved by up-scaling the MDF framework.

Section snippets

Field data

The fields modelled in this study are part of the North Wyke Farm Platform, a systems scale research facility in Southwest England for investigating the sustainability of lowland ruminant production systems (Orr et al., 2016). Rothamsted Research’s North Wyke site is located near Okehampton, Devon (504610N, 305405W) and was established in 2010. The platform is 63 ha in size divided into 15 hydrologically isolated catchments across three 21 ha farmlets with five catchments in each. Six of

MDF performance

For Great Field, the CARDAMOM-produced LAI estimates show a good fit with the assimilated Sentinel-2 LAI data (overlap = 75%) (Table 3). The initial MDF implementation identified that this field was harvested only once, in 2017, and the harvest that occurred in 2018 was not captured. When implementing CARDAMOM using this information (i.e. annual harvest probability EDC) provided, the single harvest event was simulated in June 2017 even though it actually occurred a month earlier (May 2017). For

The performance of the MDF

Our results show that by using available EO datasets within a MDF framework we can determine with acceptable accuracy (1) the timing and yield of annual grass harvests and (2) the amount of grazing-induced biomass removals. DALEC-Grass depends on weekly vegetation reduction inputs in order to simulate biomass reduction due to grazing and cutting. Because the vegetation reduction data contain temporal uncertainties, time lags between the modelled and the actual timing of harvest events were

Conclusions

Climate change, increasing human population and changing demand for livestock products together with national C neutrality targets create the need for better estimates of the C balance of managed grasslands and of its spatial variation. This study demonstrated how estimates of grassland utilisation and C budget at the farm scale can be produced through the synergy of EO data and process modelling. The level of accuracy of the predicted annual utilisation rates (grazing/harvest) was shown to be

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors would like to thank the four anonymous reviewers for improving this article through their constructive comments. VM and MW devised the study concept. VM developed DALEC-Grass, implemented the model-data fusion, and undertook the analysis. PH and HS provided ground based data and other information from North Wyke. VM and MW led the analysis of the results. VM led the writing, with support from MW, AR, PH and HS. This study was supported by the Natural Environment Research Council

References (68)

  • V. Myrgiotis et al.

    Improving model prediction of soil N2O emissions through Bayesian calibration

    Sci. Total Environ.

    (2018)
  • J. Oenema et al.

    Stochastic uncertainty and sensitivities of nitrogen flows on dairy farms in The Netherlands

    Agric. Syst.

    (2015)
  • S.M. Punalekar et al.

    Application of Sentinel-2A data for pasture biomass monitoring using a physically based radiative transfer model

    Remote Sens. Environ.

    (2018)
  • A. Qi et al.

    Grassland futures in Great Britain—Productivity assessment and scenarios for land use change opportunities

    Sci. Total Environ.

    (2018)
  • A. Qi et al.

    Modelling productivity and resource use efficiency for grassland ecosystems in the UK

    Eur. J. Agron.

    (2017)
  • H.J. Smit et al.

    Spatial distribution of grassland productivity and land use in Europe

    Agric. Syst.

    (2008)
  • J.F. Soussana et al.

    Coupling carbon and nitrogen cycles for environmentally sustainable intensification of grasslands and crop-livestock systems

    Agric. Ecosyst. Environ.

    (2014)
  • T. Takahashi et al.

    Roles of instrumented farm-scale trials in trade-off assessments of pasture-based ruminant production systems

    Animal

    (2018)
  • S. Ullah et al.

    Estimation of grassland biomass and nitrogen using MERIS data

    Int. J. Appl. Earth Obs. Geoinform.

    (2012)
  • Y. Wang et al.

    A review of applications of model-data fusion to studies of terrestrial carbon fluxes at different scales

    Agric. Forest Meteorol.

    (2009)
  • Y. Wang et al.

    Prediction of aboveground grassland biomass on the Loess Plateau, China, using a random forest algorithm

    Sci. Rep.

    (2017)
  • J. Xiao et al.

    Uncertainty in model parameters and regional carbon fluxes: a model-data fusion approach

    Agric. Forest Meteorol.

    (2014)
  • I. Ali et al.

    Satellite remote sensing of grasslands: from observation to management

    J. Plant Ecol.

    (2016)
  • M.J. Bell et al.

    Quantifying N2O emissions from intensive grassland production: the role of synthetic fertilizer type, application rate, timing and nitrification inhibitors

    J. Agric. Sci.

    (2016)
  • A.A. Bloom et al.

    The decadal state of the terrestrial carbon cycle: global retrievals of terrestrial carbon allocation, pools, and residence times

    Proc. Natl. Acad. Sci. USA

    (2016)
  • A.A. Bloom et al.

    Constraining ecosystem carbon dynamics in a data-limited world: integrating ecological “common sense” in a model-data fusion framework

    Biogeosciences

    (2015)
  • A.M. Carswell et al.

    Impact of transition from permanent pasture to new swards on the nitrogen use efficiency, nitrogen and carbon budgets of beef and sheep production

    Agric. Ecosyst. Environ.

    (2019)
  • CEH

    Land Cover Map 2015, Dataset Documentation

    Technical Report

    (2017)
  • J. Chang et al.

    Future productivity and phenology changes in European grasslands for different warming levels: Implications for grassland management and carbon balance

    Carbon Balance Manag.

    (2017)
  • J. Chang et al.

    The greenhouse gas balance of European grasslands

    Global Change Biol.

    (2015)
  • J.F. Chang et al.

    Incorporating grassland management in ORCHIDEE: model description and evaluation at 11 eddy-covariance sites in Europe

    Geosci. Model Dev.

    (2013)
  • Committee on Climate Change

    Net Zero: The UK’s contribution to stopping global warming

    Technical Report

    (2019)
  • R.T. Conant et al.

    Grassland management impacts on soil carbon stocks: A new synthesis: A

    Ecol. Appl.

    (2017)
  • DEFRA

    Agriculture in the United Kingdom 2019

    Technical Report

    (2020)
  • Cited by (9)

    • Use of remote sensing and bio-geochemical models to estimate the net carbon fluxes of managed mountain grasslands

      2022, Ecological Modelling
      Citation Excerpt :

      This is particularly the case for optical sensors mounted onboard satellite systems, some of which have large spatial coverage and frequent revisiting time (McRoberts and Tomppo, 2007; White et al., 2016). Optical remotely sensed imagery, for example, can help detecting phenology or leaf area index to be then incorporated within models (You et al., 2019; Myrgiotis et al., 2021); the same imagery can provide effective estimates of the fraction of photosynthetically active radiation absorbed by vegetation (fAPAR), which are combined with ancillary data to obtain the ecosystem GPP through the so-called light use efficiency (LUE) approach (Veroustraete et al., 2002; Yu, 2020). This approach provides the theoretical foundation for the most widespread GPP prediction methods, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) GPP algorithm, which has been assessed in a number of cases (Heinsch et al., 2006; Yuan et al., 2014).

    • Evaluation of long-term carbon sequestration of biochar in soil with biogeochemical field model

      2022, Science of the Total Environment
      Citation Excerpt :

      These environmental stresses form a three-dimensional force field controlling the physical transport, crystallization-dissolution, synthesis-decomposition, oxidation-reduction, adsorption-desorption of elements, etc., and realize the simulation of the migration and transformation of elements in the natural environment (Deng et al., 2018a). As a result, the complex elemental fate in nature is transformed into a simple set of physical, thermodynamic and kinetic equations, while temperature, moisture, pH, Eh, etc. become the parameters or variables (Myrgiotis et al., 2021). In this study, Biogeochemical Field consists of four major modules, including (A) Soil Water/Heat Transfer, (B) Plant Growth, (C) Organic Matter Decomposition and (D) Fermentation/Nitrification/Denitrification (Fig. 2 and Fig. S1).

    View all citing articles on Scopus
    View full text