Grassland ecosystem services in a changing environment: The potential of hyperspectral monitoring

https://doi.org/10.1016/j.rse.2019.111273Get rights and content

Highlights

  • First hyperspectral analysis within a free air carbon dioxide enrichment study

  • Performances of different hyperspectral normalisation techniques were investigated.

  • Good to excellent predictive performances for 14 traits related to ecosystem services.

  • Accurate monitoring of plant traits even under varying carbon dioxide concentrations.

  • Highlighting of capabilities to derive high resolution spatio-temporal canopy data.

Abstract

Provisioning services from grassland ecosystems are strongly linked to physical and chemical grassland traits, which are affected by atmospheric CO2 concentrations ([CO2]s). The influences of increased [CO2]s ([eCO2]s) are typically investigated in Free Air Carbon dioxide Enrichment (FACE) studies via destructive sampling methods. This traditional approach is restricted to sampling plots and harvest dates, while hyperspectral approaches provide new opportunities as they are rapid, non-destructive and cost-effective. They further allow a high temporal resolution including spatially explicit information. In this study we investigated the hyperspectral predictability of 14 grassland traits linked to forage quality and quantity within a FACE experiment in central Germany with three plots under ambient atmospheric [CO2]s, and three plots at [eCO2]s (∼20% above ambient [CO2]s). We analysed the suitability of various normalisation and feature selection techniques to link comprehensive laboratory analyses with two years of hyperspectral measurements (spectral range 600–1600 nm). We applied partial least squares regression and found good to excellent predictive performances (0.49 ≤ leave one out cross-validation R2≤ 0.94), which depended on the normalisation method applied to the hyperspectral data prior to model training. Noteworthy, the models' predictive performances were not affected by the different [CO2]s, which was anticipated due to the altered plant physiology under [eCO2]s. Thus, an accurate monitoring of grassland traits under different [CO2]s (present-day versus future, or within a FACE facility) is promising, if appropriate predictors are selected. Moreover, we show how hyperspectral predictions can be used e.g., within a future phenotyping approach, to monitor the grassland on a spatially explicit level and on a higher temporal resolution compared to conventional destructive sampling techniques. Based on the information during the vegetation period we show how hyperspectral monitoring might be used e.g., to adapt harvest practices or gain deeper insights into physiological plant alterations under [eCO2]s.

Introduction

Grasslands provide multiple important ecosystem services such as the supply of forage for livestock and substrate for biogas production, provision of recreational space, and carbon sequestration (White et al., 2000, Herrero et al., 2013). Grassland ecosystem services are strongly linked to physical and chemical plant traits, which are affected by species morphology and distribution parameters (Milcu et al., 2016, Isbell et al., 2017, Dietrich et al., 2017, Meyer et al., 2018, Weisser et al., 2017, Baert et al., 2018, Fischer et al., 2019), and environmental conditions (e.g., CO2 concentration, air temperature, water availability). Increasing atmospheric CO2 concentration ([eCO2]) under climate change conditions affect plant physiology through increasing the photosynthesis rate and the water use efficiency, and thus increases the aboveground biomass productivity in C3 grasslands (Morgan et al., 2004, Nowak et al., 2004, Ainsworth and Long, 2005, Ainsworth and Rogers, 2007, Lee et al., 2010, McGranahan and Yurkonis, 2018, Campbell and Stafford Smith, 2000, dec), which is accompanied by a decreased forage N content (Cotrufo et al., 1998, Nowak et al., 2004, Ainsworth and Long, 2005, Campbell and Stafford Smith, 2000, dec, Dumont et al., 2015, Augustine et al., 2018). From an agroeconomic perspective, forage digestibility may either remain unchanged (Dumont et al., 2015) or decreased (Morgan et al., 2004, Augustine et al., 2018), and forage quality (indicated by crude protein availability) might either be decreased (Soussana and Lüscher, 2007) or increased (McGranahan and Yurkonis, 2018). Such influences of [eCO2]s on grassland traits, and thus, ecosystem services, are mainly investigated within free air carbon dioxide enrichment (FACE) experiments.

To estimate the ecosystem status and potential outcomes of grassland ecosystems in general and on FACE experiments, in-situ measurements of plant traits are usually performed by destructive vegetation samples, such as labour- and cost-intensive biomass cuttings for subsequent time-consuming laboratory analysis (e.g., Buchmann et al., 2018, Brookshire and Weaver, 2015, Andresen et al., 2018, Hovenden et al., 2014). Consequently, traditional sampling methods constrain the analysis of the grassland traits under different [CO2]s to certain harvest dates and sampling plots. To overcome these problems, multi- and hyperspectral approaches have proven to be advantageous due to their rapid and non-destructive sampling, allowing for high temporal resolutions with spatially explicit information at high accuracy and a reasonable price (Walter et al., 2012, Walter et al., 2015, Humphreys et al., 2006). While multi- or hyperspectral applications to grassland FACE experiments have not yet been considered, various studies have proven the feasibility of the optical delineation of physical and chemical grassland traits such as aboveground biomass (Marabel and Alvarez-Taboada, 2013, Xiaoping et al., 2011, Kawamura et al., 2008, Zhao et al., 2007), nitrogen content (Ramoelo et al., 2013), chlorophyll content (Darvishzadeh et al., 2008), leaf area index (Darvishzadeh et al., 2008), crude protein (Biewer et al., 2009, Suzuki et al., 2008, Kawamura et al., 2008, Zhao et al., 2007, Pullanagari et al., 2012a,Pullanagari et al., 2012b, Pullanagari et al., 2013), crude lipids (Pullanagari et al., 2012b, Pullanagari et al., 2013), crude ash (Biewer et al., 2009, Pullanagari et al., 2012a,Pullanagari et al., 2012b, Pullanagari et al., 2013, neutral and acid detergent fibre (Zhao et al., 2007, Biewer et al., 2009, Kawamura et al., 2008, Pullanagari et al., 2012a,Pullanagari et al., 2012b, Pullanagari et al., 2013), enzyme-soluble organic matter (Pullanagari et al., 2012a, Pullanagari et al., 2012b), and metabolizable energy, which is required to derive the potential of energy extraction for ruminants (Pullanagari et al., 2013). Most of these studies applied different methods of spectral transformations, to minimize the effect of external perturbing factors e.g., soil background, illumination, and viewing geometry and/or to enhance the spectral absorption features in hyperspectral data.

To date, hyperspectral monitoring of different grassland traits under varying [CO2]s and accompanying plant physiological alterations (e.g., within a FACE facility) have not been tested with respect to their feasibility. Therefore, it is not clear whether the transfer functions between spectral indices and grassland traits derived from plants grown under present-day [CO2] conditions will accurately predict grassland traits under future atmospheric [CO2] conditions. Difficulties may arise since the spectral delineation of plant traits might be affected by different CO2 concentrations as a function of the altered physiology of plants under increasing [CO2]s. For instance, an increased photosynthesis under [eCO2]s may lead to higher biomass accumulation for plants under [eCO2] compared to plants grown under today's ambient [CO2], despite a similar chlorophyll content (the latter is well detectable by optical sensors; e.g., Gitelson et al., 2003, Haboudane et al., 2002, Maccioni et al., 2001, Daughtry et al., 2000, le Maire et al., 2008).

We combine advanced hyperspectral measurements and advanced data processing techniques within a FACE facility in central Germany, to set up a non-invasive monitoring approach for the most important grassland traits under different CO2 concentrations. Here, a careful selection of methods and predictors is mandatory to enable the hyperspectral monitoring of grassland traits. The latter shall help to overcome the sampling restriction of invasive procedures to certain plots and dates, which may deliver new insights on the spatio-temporal dynamics of grassland traits under different CO2 concentrations and weather conditions.

Consequently, we hypothesize that: (1) Specific spectral transformations for individual canopy traits enable an accurate trait prediction by means of hyperspectral data, (2) the hyperspectral predictability of different grassland traits is biased under different [CO2]s due to physiological alterations, and (3) higher spatial and temporal resolutions of grassland trait values (hyperspectral predictions compared to destructive sampling) enable knowledge gains e.g., to improve management practices and the understanding of biophysical plant alterations under ambient and elevated [CO2]s.

Section snippets

Study area and sampling

Field samplings were conducted at the Environmental Monitoring and Climate Impact Research Station Linden located near Giessen, Germany (50° 32 N and 8° 41 E; 172 m a.s.l.). Here, a FACE experiment comprising six rings with 8 m in diameter is in operation on an extensively managed, species-rich grassland under moderate climate conditions (Jäger et al., 2003, Andresen et al., 2018). The non-grazed grassland has been mown twice a year, fertilized with only 40 kg N ha−1 yr−1, and never ploughed

Results

The main PFT in the ambient as well as in the elevated rings, as indicated by the relative biomass proportions, was grasses followed closely by forbs (Table 2). Legumes showed only minor contributions to the total biomass of the rings in the ambient as well as in the elevated rings. No significant differences in the proportions of PFT-wise biomasses between the treatments were observed by paired samples t-test (for grasses and forbs) and paired samples Wilcoxon test (for legumes).

The summary

Treatment-wise grassland characteristics

The high proportions of the forbs PFT can be explained by the management practices (low fertilization rates and the harvest regime) that has lead to a steady increase of the forbs PFT over the long-term period of the experimental operation (Andresen et al., 2018). Noteworthy to say, this long-term increase in the proportions of the forbs PFT was observed in the ambient as well as in the elevated rings. Similarly, no significant differences in the relative biomass proportions of the different

Conclusion

Our results clearly show that the hyperspectral prediction, and thus a non-invasive monitoring, of various grassland traits in the investigated grassland is in general feasible. However, careful creations and selections of appropriate predictor variables are needed for each canopy trait as shown by the large differences in the predictive performance of the different feature spaces. Interestingly, our results show that [eCO2] does not lead to biases in the hyperspectral predictions of grassland

Acknowledgements

The contribution of the following individuals to the initiation, construction, installation and long-term, ongoing maintenance of the Giessen FACE experiment is gratefully acknowledged: H.-J. Jäger (deceased 2013), L. Grünhage, S. Schmidt, J. Senkbeil, W. Stein, B. Lenz, J. Franz, T. Strohbusch, G. Mayer and A. Brück. Furthermore the authors want to thank S. Achilles and A. Bendix for their ongoing contributions to field sampling and S. Achilles for the help with technical aspects. This

References (1)

    Cited by (39)

    • A hybrid model to predict nitrogen concentration in heterogeneous grassland using field spectroscopy

      2023, Remote Sensing of Environment
      Citation Excerpt :

      A common method of monitoring heterogeneous grassland characteristics is by using captured spectral measurements alongside empirical-statistical algorithms (Darvishzadeh et al., 2014; Quan et al., 2017; Proctor et al., 2017). For example, The use of Partial Least Squares Regression (PLSR) alongside grassland hyperspectral data is very popular, achieving different levels of accuracy (Foster et al., 2017; Pullanagari et al., 2012; Obermeier et al., 2019). PLSR transforms the collected spectral information into new predictor components with a lower feature space to explain target variables (N%) (Wold et al., 2001; Berger et al., 2020b,a).

    • Improving the integrated efficacy of ecosystem restoration efforts by linking land degradation neutrality to ecosystem service enhancement from a spatial association perspective

      2022, Ecological Engineering
      Citation Excerpt :

      Similarly, ecosystem degradation in specific areas tends to result in ecosystem deterioration in the surrounding areas (Chen et al., 2020). However, although associations between ESs and their determinants have been investigated using multiple statistical approaches, such as gray correlation degree analysis (Zhang et al., 2020), logistic regression (Zhang et al., 2021), partial least squares (Cebrián-Piqueras et al., 2020; Catherine et al., 2021), random forest models (Wang et al., 2021b), machine learning (Liu et al., 2021b), and multiple regression models (Obermeier et al., 2019; Shifaw et al., 2019), spatial attributes (i.e., spatial location information), and spillover effects of independent and dependent variables have not been sufficiently considered in these traditional approaches (Chi and Ho, 2018). Therefore, the limitations of these statistical approaches constrain the identification of spatial determinants of ESs, soil loss, and their spillover effects (Chen et al., 2019, 2020).

    View all citing articles on Scopus
    View full text