EGU2020-15936
https://doi.org/10.5194/egusphere-egu2020-15936
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Which hydro-meteorological variables control large-scale photosynthesis?

Wantong Li, Mirco Migliavacca, Yunpeng Luo, and René Orth
Wantong Li et al.
  • Department of Biogeochemical Integration, Max Planck Institute for Biogeochemistry, Jena, Germany (wantong@bgc-jena.mpg.de)

Vegetation dynamics are determined by a multitude of hydro-meteorological variables, and this interplay changes in space and time. Due to its complexity, it is still not fully understood at large spatial scales. This knowledge gap contributes to increased uncertainties in future climate projections because large-scale photosynthesis is influencing the exchange of energy and water between the land surface and the atmosphere, thereby potentially impacting near-surface weather. In this study, we explore the relative importance of several hydro-meteorological variables for vegetation dynamics. For this purpose, we infer the correlations of anomalies in temperature, precipitation, soil moisture, VPD, surface net radiation and surface downward solar radiation with respective anomalies of photosynthetic activity as inferred from Sun-Induced chlorophyll Fluorescence (SIF). To detect changing hydro-meteorological controls across different climate conditions, this global analysis distinguishes between climate regimes as determined by long-term mean aridity and temperature. The results show that soil moisture was the most critical driver with SIF in the simultaneous correlation with dry and warm conditions, while temperature and VPD was both influential on cold and wet regimes during the study period 2007-2018. We repeat our analysis by replacing the SIF data with NDVI, as a proxy for vegetation greenness, and find overall similar results, except for surface net radiation expanding controlled regions on cold and wet regimes. As the considered hydro-meteorological variables are inter-related, spurious correlations can occur. We test different approaches to investigate and account for this phenomenon. The results can provide new insight into mechanisms of vegetation-water-energy interactions and contribute to improve dynamic global vegetation models.

How to cite: Li, W., Migliavacca, M., Luo, Y., and Orth, R.: Which hydro-meteorological variables control large-scale photosynthesis?, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-15936, https://doi.org/10.5194/egusphere-egu2020-15936, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 1 – uploaded on 30 Apr 2020
  • AC1: Comments from the live chat, Wantong Li, 07 May 2020
    Tim van Emmerik (WUR, co-convener) (09:21) Thanks Wantong! To what extent do you think your results (mean drivers) are an artifact of your machine learning recipe?
    J Green LSCE (convener) (09:22) I was also wondering if you had tried using any other soil moisture products-- as I know that there are some pretty large differences between those?
    Brianna Pagán UGent (co-convener) (09:24) @Wantong NDVI isn't shown but I am wondering how large a difference you see between the coarse SIF aggregating at bi-weekly intervals and NDVI?
    Wantong Li (MPI-BGC) (09:24) @Thank you, J Green, Random foest can nicely calculate the importance of drivers. Though it could not perfectly deal with problem like conearility, while it give more significant results.
    Wantong Li (MPI-BGC) (09:25) @Tim, @ J Green, we also tested different correlative methods, which showed similar parrterns as we found in RF. Besides, adjusting optimized parameters (optimized in grid cell samples) in RF gives little change in the prediction results.
    Wantong Li (MPI-BGC) (09:26) @J Green, we not yet tested other SM products. That is on our list. One problem is, we distinguished depths of SM, which is hard to be provided by other products.
    JC Calvet CNRM Toulouse (audience) (09:26) @Wantong. Are there seasonal variations in the sensitivity to different soil layers?
    Wantong Li (MPI-BGC) (09:27) @Brianna, nice question, we found many regions like in north latitudes, NDVI and SIF showed similar results, while in some regions of tropics, they are different. The uncertainty in SIF and NDVI in tropics need to be further tested.
    Wantong Li (MPI-BGC) (09:28) @JC Calvet, we already removed the seasonality of data.
    J Green LSCE (convener) (09:28) Yes I would think that NDVI would also saturate in the tropics
  • CC1: Comment on EGU2020-15936, Thomas Lees, 07 May 2020

    Great study thanks! How did you control for the colinearity of the input features? For example the Soil Levels 1-4 are highly correlated variables. Disentangling the importance of each feature separately can be difficult when included in the same model. How did you overcome this issue?

    Great work really cool!

    • AC2: Reply to CC1, Wantong Li, 07 May 2020

      Thank you for your question! We didn’t handle the potential problem of collinearity in Random forest, while we did contrast methods based on machine learning and correlation analysis, and they showed similar results. This support the patterns of main drivers that we identified by RF.

    • AC4: Reply to CC1, Wantong Li, 08 May 2020

      I'd like to give some complements:

      We cannot fully overcome the problem of collinearity is a limitation of our study.
      We try to mitigate the problem by using state-of-the-art machine learning (random forests), and removing the seasonal cycles in our next step.

  • CC2: Why replace potential evapotranspiration in calculation of Aridity?, Thomas Lees, 07 May 2020

    In your slides you mention: "Here we use unit-adjusted net radiation to replace potential evapotranspiration" why is that? Potential evaporation is available as a field in ERA5?

    • AC3: Reply to CC2, Wantong Li, 07 May 2020

      Nice question! There are many ways to calculate aridity index. We simply choose the method of using united-adjusted solar radiation devided by precipitation, because we already have solar radiation and precipitation data in global ERA5 datasets, and also the uncertainties of different aridity index hardly impact our main analysis.

    • AC5: Reply to CC2, Wantong Li, 08 May 2020

      Another point is that, potential evapotranspiration is hard to validate, so the net radiation from ERA5 datasets can be more reliable that the calculation of PET.