Two soil hydrology formulations of ORCHIDEE (version Trunk.rev1311) tested for the Amazon basin

Matthieu Guimberteau, Philippe Ciais, Agnès Ducharne, Juan Pablo Boisier, Shushi Peng, M. De Weirdt, Hans Verbeeck2014 · DOI

TL;DR

This study compares two different soil hydrology models within the ORCHIDEE land surface model for the Amazon basin. While the overall water budget differences are minor, the more complex 11-layer model shows improved simulation of dry-season evapotranspiration and better captures soil water storage dynamics, particularly in drier sub-basins, suggesting its importance for future climate change impact assessments.

What Problem Does This Paper Address?

The Amazon basin's complex hydrology plays a crucial role in global climate. Accurately modeling this system requires robust land surface models (LSMs), and a key component of LSMs is how they represent soil hydrology. Different soil models can lead to varying simulations of water availability, evapotranspiration, and ultimately, vegetation response and carbon fluxes. This paper addresses the question of whether a more physically-based, multi-layered soil diffusion scheme (11-layer) offers significant improvements over a simpler bucket-type scheme (2-layer) in simulating the Amazonian water cycle and its interactions with vegetation, especially under varying dry-season conditions.

Methods

The study utilizes the ORCHIDEE (Organizing Carbon and Hydrology in Dynamic Ecosystems) land surface model. Two soil hydrology parameterizations are compared:

  • 2-layer soil scheme (2LAY): A simpler, bucket-type model with an evaporative layer.
  • 11-layer soil diffusion scheme (11LAY): A more complex model that simulates water diffusion based on physical processes (Fokker-Planck equation derived from Richards equation) across 11 discrete soil layers.

Both schemes are coupled with the same river routing module and the STOMATE module for phenology and carbon dynamics. The simulations cover the Amazon basin and five major sub-basins from 1980 to 2008. Model outputs are compared against various observational datasets for water budget components, total terrestrial water storage (TWS), evapotranspiration (ET), gross primary production (GPP), and leaf area index (LAI).

Datasets

The study uses the following datasets for evaluation:

  • Precipitation (P): ORE HYBAM dataset (independent from forcing data), daily in situ raingauge observations interpolated to 1°x1° resolution.
  • Evapotranspiration (ET): MTE-ET (Model Tree Ensemble-EvapoTranspiration) product, monthly estimates at 0.5°x0.5° resolution, derived from FLUXNET eddy-covariance measurements and remote sensing.
  • Gross Primary Production (GPP): MTE-GPP product, a global data-driven GPP product at 0.5°x0.5° spatial resolution and monthly temporal frequency.
  • Total Terrestrial Water Storage (TWS): GRACE satellite data (Level-2 Gravity Field Product), 1°x1° monthly data.
  • Leaf Area Index (LAI): Zhu et al. (2013) product, based on GIMMS NDVI3g and MODIS LAI, at 1/12° spatial resolution and 15-day temporal frequency.
  • River Discharge (Q): ORE HYBAM project data, from six river gauging stations representative of the main sub-basins.

Key Insights and Ideas

  • Dry season is critical: The length and intensity of the dry season are crucial for revealing differences between the two soil models, particularly in how they influence ET and vegetation water stress.
  • Soil water storage matters: The 11-layer model's ability to store more water in the soil, due to its higher water holding capacity and physically-based drainage, leads to more dynamic soil water storage variations and improved TWS simulations compared to GRACE.
  • Sustained ET in dry periods: The 11-layer model can sustain evapotranspiration for longer periods during the dry season, especially in southeastern sub-basins like the Xingu, which the 2-layer model struggles with.
  • Vegetation response: Lower plant water stress simulated by the 11-layer model leads to a better representation of the seasonal cycle of photosynthesis (GPP).
  • Sub-basin variability: While overall water budget differences are small, the 11-layer model shows more significant improvements in drier sub-basins compared to wetter ones.

Key Results

  • Water Budget: The 11-layer scheme did not significantly alter the overall Amazonian water budget, with only minor changes in evapotranspiration (+3.1%) and river discharge (-3.0%) compared to the 2-layer scheme.
  • Total Terrestrial Water Storage (TWS): The 11-layer model showed improved simulation of TWS variation when compared to GRACE satellite estimates, particularly in capturing higher soil water storage dynamics.
  • Dry-Season ET: The 11-layer model resulted in increased dry-season evapotranspiration (+0.5 mm d⁻¹, +17%) in southeastern sub-basins like the Xingu, sustaining ET throughout the dry season, unlike the 2-layer model.
  • Photosynthesis (GPP): The 11-layer model led to a better simulation of the seasonal GPP cycle due to lower simulated plant water stress.
  • Bias Reduction: The 11-layer scheme reduced biases in ET (up to -4%) and Q (up to -10%) compared to observations, especially in southern sub-basins.
  • LAI: Simulated LAI was higher with the 11-layer model (up to +0.4) but showed less variation than observed.

Limitations and Future Work

  • LAI Variation: The 11-layer model, while producing higher LAI, exhibited too low a variation compared to satellite-based datasets.
  • Floodplain Areas: The MTE-ET product might underestimate actual ET because it doesn't fully account for floodplain areas.
  • Root Depth: The authors suspect that extending the soil depth and rooting depth beyond 2m could further improve simulations, especially in areas with deep-rooted forests, as observed by Nepstad et al. (1994).
  • Further Research: The study highlights the need for more attention to improving the representation of soil hydrology and its relationship with vegetation dynamics in LSMs. Developing these relationships could improve simulations of dry-season precipitation and low river flows. The authors suggest that the current work is being extended to a global scale to identify similar signals in transition zones where soil moisture influences precipitation. Additional site-level simulations with flux tower measurements are recommended to identify missing modeling processes.

Summary updated

Abstract

This study analyzes the impact of the two soil model parameterizations of the Land Surface Model ORCHIDEE on their estimates of Amazonian hydrology and phenology for five major sub-basins (Xingu, Tapajós, Madeira, Solimões and Negro), during the 29 yr period 1980–2008. The two soil models are a simple 2 layer soil scheme with a bucket topped by an evaporative layer vs. an 11 layer soil diffusion scheme. The soil models were coupled with a river routing module and a process model of plant physiology, phenology and carbon dynamics. The simulated water budget and vegetation functioning components were compared with several datasets at sub-basin scale. The use of the 11 layer soil diffusion scheme did not significantly change the Amazonian water budget simulation when compared to the 2 layer soil scheme (+3.1 and −3.0% in evapotranspiration and river discharge, respectively). However, the higher water holding capacity of the soil and the physically based representation of runoff and drainage in the 11 layer soil diffusion, resulted in higher dynamics of soil water storage variation and improved simulation of the total terrestrial water storage when compared to GRACE satellite estimates. The greater soil water storage within the 11 layer soil diffusion scheme resulted in increased dry-season evapotranspiration (+0.5 mm d−1, +17%) and river discharge in the southeastern sub-basins such as the Xingu. Evapotranspiration over this sub-basin was sustained during the whole dry season with the 11 layer soil diffusion model, whereas the 2 layer soil scheme limited it at the end of the dry season. Lower plant water stress simulated by the 11 layer soil diffusion scheme, led to better simulation of the seasonal cycle of photosynthesis (GPP) when compared to a GPP data-driven model based upon eddy-covariance and satellite greenness measurements. Simulated LAI was consequently higher with the 11LAY (up to +0.4) but exhibited too low a variation when compared to a satellite-based dataset. The dry-season length between 4 and 7 months over the entire Amazon basin was found to be critical in distinguishing differences in hydrological feedbacks between the soil and the vegetation cover simulated by the two soil models. Overall, the 11 layer soil diffusion scheme provided little improvement in simulated hydrology on average over the wet tropical Amazonian sub-basins but a more significant improvement over the drier sub-basins. However, the use of the 11 layer soil diffusion scheme might become critical for assessments of future hydrological changes, especially in southern regions of the Amazon basin where longer dry season and more severe droughts are expected in the next century.
BibTeX
@misc{guimberteau2014two,
  author = {Matthieu Guimberteau and Philippe Ciais and Agnès Ducharne and Juan Pablo Boisier and Shushi Peng and M. De Weirdt and Hans Verbeeck},
  title = {{Two soil hydrology formulations of ORCHIDEE (version Trunk.rev1311) tested for the Amazon basin}},
  year = {2014},
  doi = {10.5194/gmdd-7-73-2014},
}
Figure