An official website of the United States government
Abstract Evapotranspiration (ET) is a significant ecosystem flux, governing the partitioning of energy at the land surface. Understanding the seasonal pattern and magnitude ofETis critical for anticipating a range of ecosystem impacts, including drought, heat‐wave events, and plant mortality. In this study, we identified the relative controls of seasonal variability inET, and how these controls vary among ecosystems. We used overlapping AmeriFlux and PhenoCam time series at a daily timestep from 20 sites to explore these linkages (# site‐years >100), and our study area covered a broad climatological aridity gradient in the U.S. and Canada. We focused on disentangling the most important controls of bulk surface conductance (Gs) and evaporative fraction (EF = LE/[H + LE]), whereLEandHrepresent latent and sensible heat fluxes, respectively. Specifically, we investigated how vegetation phenology varied in importance relative to meteorological variables (vapor pressure deficit and antecedent precipitation) as a driver ofGsandEFusing path analysis, a framework for quantifying and comparing the causal linkages among multiple response and explanatory variables. Our results revealed that the drivers ofGsandEFseasonality varied significantly between energy‐ and water‐limited ecosystems. Specifically, precipitation had a much higher effect in water‐limited ecosystems, while seasonal patterns in canopy greenness emerged as a stronger control in energy‐limited ecosystems. Given that phenology is expected to shift under future climate, our findings provide key information for understanding and predicting how phenology may impact 21st‐century hydroclimate regimes and the surface‐energy balance.
more »
« less
Hybrid modeling of evapotranspiration: inferring stomatal and aerodynamic resistances using combined physics-based and machine learning
Abstract The process of evapotranspiration transfers liquid water from vegetation and soil surfaces to the atmosphere, the so-called latent heat flux ( Q LE ), and modulates the Earth’s energy, water, and carbon cycle. Vegetation controls Q LE by regulating leaf stomata opening (surface resistance r s in the Big Leaf approach) and by altering surface roughness (aerodynamic resistance r a ). Estimating r s and r a across different vegetation types is a key challenge in predicting Q LE . We propose a hybrid approach that combines mechanistic modeling and machine learning for modeling Q LE . The hybrid model combines a feed-forward neural network which estimates the resistances from observations as intermediate variables and a mechanistic model in an end-to-end setting. In the hybrid modeling setup, we make use of the Penman–Monteith equation in conjunction with multi-year flux measurements across different forest and grassland sites from the FLUXNET database. This hybrid model setup is successful in predicting Q LE , however, this approach leads to equifinal solutions in terms of estimated physical parameters. We follow two different strategies to constrain the hybrid model and therefore control for the equifinality that arises when the two resistances are estimated simultaneously. One strategy is to impose an a priori constraint on r a based on mechanistic assumptions (theory-driven strategy), while the other strategy makes use of more observational data and adds a constraint in predicting r a through multi-task learning of both latent and sensible heat flux ( Q H ; data-driven strategy) together. Our results show that all hybrid models predict the target variables with a high degree of success, with R 2 = 0.82–0.89 for grasslands and R 2 = 0.70–0.80 for forest sites at the mean diurnal scale. The predicted r s and r a show strong physical consistency across the two regularized hybrid models, but are physically implausible in the under-constrained hybrid model. The hybrid models are robust in reproducing consistent results for energy fluxes and resistances across different scales (diurnal, seasonal, and interannual), reflecting their ability to learn the physical dependence of the target variables on the meteorological inputs. As a next step, we propose to test these heavily observation-informed parameterizations derived through hybrid modeling as a substitute for ad hoc formulations in Earth system models.
more »
« less
- Award ID(s):
- 2019625
- PAR ID:
- 10430172
- Date Published:
- Journal Name:
- Environmental Research Letters
- Volume:
- 18
- Issue:
- 3
- ISSN:
- 1748-9326
- Page Range / eLocation ID:
- 034039
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract. Land surface modellers need measurable proxies toconstrain the quantity of carbon dioxide (CO2) assimilated bycontinental plants through photosynthesis, known as gross primary production(GPP). Carbonyl sulfide (COS), which is taken up by leaves through theirstomates and then hydrolysed by photosynthetic enzymes, is a candidate GPPproxy. A former study with the ORCHIDEE land surface model used a fixedratio of COS uptake to CO2 uptake normalised to respective ambientconcentrations for each vegetation type (leaf relative uptake, LRU) tocompute vegetation COS fluxes from GPP. The LRU approach is known to havelimited accuracy since the LRU ratio changes with variables such asphotosynthetically active radiation (PAR): while CO2 uptake slows underlow light, COS uptake is not light limited. However, the LRU approach hasbeen popular for COS–GPP proxy studies because of its ease of applicationand apparent low contribution to uncertainty for regional-scaleapplications. In this study we refined the COS–GPP relationship andimplemented in ORCHIDEE a mechanistic model that describes COS uptake bycontinental vegetation. We compared the simulated COS fluxes againstmeasured hourly COS fluxes at two sites and studied the model behaviour andlinks with environmental drivers. We performed simulations at a global scale,and we estimated the global COS uptake by vegetation to be −756 Gg S yr−1,in the middle range of former studies (−490 to −1335 Gg S yr−1). Basedon monthly mean fluxes simulated by the mechanistic approach in ORCHIDEE, wederived new LRU values for the different vegetation types, ranging between0.92 and 1.72, close to recently published averages for observed values of1.21 for C4 and 1.68 for C3 plants. We transported the COS using the monthlyvegetation COS fluxes derived from both the mechanistic and the LRUapproaches, and we evaluated the simulated COS concentrations at NOAA sites.Although the mechanistic approach was more appropriate when comparing tohigh-temporal-resolution COS flux measurements, both approaches gave similarresults when transporting with monthly COS fluxes and evaluating COSconcentrations at stations. In our study, uncertainties between these twoapproaches are of secondary importance compared to the uncertainties in theCOS global budget, which are currently a limiting factor to the potential ofCOS concentrations to constrain GPP simulated by land surface models on theglobal scale.more » « less
-
Abstract Forest leaf area has enormous leverage on the carbon cycle because it mediates both forest productivity and resilience to climate extremes. Despite widespread evidence that trees are capable of adjusting to changes in environment across both space and time through modifying carbon allocation to leaves, many vegetation models use fixed carbon allocation schemes independent of environment, which introduces large uncertainties into predictions of future forest responses to atmospheric CO2fertilization and anthropogenic climate change. Here, we develop an optimization‐based model, whereby tree carbon allocation to leaves is an emergent property of environment and plant hydraulic traits. Using a combination of meta‐analysis, observational datasets, and model predictions, we find strong evidence that optimal hydraulic–carbon coupling explains observed patterns in leaf allocation across large environmental and CO2concentration gradients. Furthermore, testing the sensitivity of leaf allocation strategy to a diversity in hydraulic and economic spectrum physiological traits, we show that plant hydraulic traits in particular have an enormous impact on the global change response of forest leaf area. Our results provide a rigorous theoretical underpinning for improving carbon cycle predictions through advancing model predictions of leaf area, and underscore that tree‐level carbon allocation to leaves should be derived from first principles using mechanistic plant hydraulic processes in the next generation of vegetation models.more » « less
-
In this paper we present a hybrid neural network augmented physics-based modeling (APBM) framework for Bayesian nonlinear latent space estimation. The proposed APBM strategy allows for model adaptation when new operation conditions come into play or the physics-based model is insufficient (or incomplete) to properly describe the latent phenomenon. One advantage of the APBMs and our estimation procedure is the capability of maintaining the physical interpretability of estimated states. Furthermore, we propose a constraint filtering approach to control the neural network contributions to the overall model. We also exploit assumed density filtering techniques and cubature integration rules to present a flexible estimation strategy that can easily deal with nonlinear models and high-dimensional latent spaces. Finally, we demonstrate the efficacy of our methodology by leveraging a target tracking scenario with nonlinear and incomplete measurement and acceleration models, respectively.more » « less
-
We investigated the climatic and ecohydrological controls of the monthly methane emission fluxes from freshwater wetlands across the globe. Fluxes of methane, photosynthetically active radiation (PAR), soil temperature (TS), atmospheric pressure, latent heat flux (LE), wind speed (WS), friction velocity, vapor pressure deficit (VPD), soil water content (SWC), water table depth, and precipitation were obtained from 32 FLUXNET wetland sites. Multivariate pattern recognition techniques of principal component and factor analyses were utilized to classify and group climatic and ecological variables based on their similarity as drivers, examining their interrelation patterns across the different sites. Partial least squares regression models were developed to estimate the relative linkages of methane emission fluxes with the climatic and ecohydrological drivers. When the wetlands were flooded (i.e., positive water table depth relative to the ground), PAR, LE, VPD, and TS had the strongest controls on the methane emission fluxes. However, in the absence of flooding (i.e., negative water table depth), the methane emission fluxes were mainly controlled by SWC and WS. For the wetland sites with unavailable water table depth data, PAR, TS, and WS had the strongest controls on the methane emissions and subsequent transport. Our findings provided important knowledge and insights for predicting and managing methane emissions in freshwater wetlands at a global scale.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1088/1748-9326/acbbe0
