skip to main content

Title: A machine learning approach to emulation and biophysical parameter estimation with the Community Land Model, version 5
Abstract. Land models are essential tools for understanding and predicting terrestrial processes and climate–carbon feedbacks in the Earth system, but uncertainties in their future projections are poorly understood. Improvements in physical process realism and the representation of human influence arguably make models more comparable to reality but also increase the degrees of freedom in model configuration, leading to increased parametric uncertainty in projections. In this work we design and implement a machine learning approach to globally calibrate a subset of the parameters of the Community Land Model, version 5 (CLM5) to observations of carbon and water fluxes. We focus on parameters controlling biophysical features such as surface energy balance, hydrology, and carbon uptake. We first use parameter sensitivity simulations and a combination of objective metrics including ranked global mean sensitivity to multiple output variables and non-overlapping spatial pattern responses between parameters to narrow the parameter space and determine a subset of important CLM5 biophysical parameters for further analysis. Using a perturbed parameter ensemble, we then train a series of artificial feed-forward neural networks to emulate CLM5 output given parameter values as input. We use annual mean globally aggregated spatial variability in carbon and water fluxes as our emulation and calibration more » targets. Validation and out-of-sample tests are used to assess the predictive skill of the networks, and we utilize permutation feature importance and partial dependence methods to better interpret the results. The trained networks are then used to estimate global optimal parameter values with greater computational efficiency than achieved by hand tuning efforts and increased spatial scale relative to previous studies optimizing at a single site. By developing this methodology, our framework can help quantify the contribution of parameter uncertainty to overall uncertainty in land model projections. « less
Authors:
; ; ;
Award ID(s):
1947282
Publication Date:
NSF-PAR ID:
10233245
Journal Name:
Advances in Statistical Climatology, Meteorology and Oceanography
Volume:
6
Issue:
2
Page Range or eLocation-ID:
223 to 244
ISSN:
2364-3587
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract. The terrestrial carbon cycle plays a critical role in modulating the interactions of climate with the Earth system, but different models often make vastly different predictions of its behavior. Efforts to reduce model uncertainty have commonly focused on model structure, namely by introducing additional processes and increasing structural complexity. However, the extent to which increased structural complexity can directly improve predictive skill is unclear. While adding processes may improve realism, the resulting models are often encumbered by a greater number of poorly determined or over-generalized parameters. To guide efficient model development, here we map the theoretical relationship between model complexity and predictive skill. To do so, we developed 16 structurally distinct carbon cycle models spanning an axis of complexity and incorporated them into a model–data fusion system. We calibrated each model at six globally distributed eddy covariance sites with long observation time series and under 42 data scenarios that resulted in different degrees of parameter uncertainty. For each combination of site, data scenario, and model, we then predicted net ecosystem exchange (NEE) and leaf area index (LAI) for validation against independent local site data. Though the maximum model complexity we evaluated is lower than most traditional terrestrial biosphere models, themore »complexity range we explored provides universal insight into the inter-relationship between structural uncertainty, parametric uncertainty, and model forecast skill. Specifically, increased complexity only improves forecast skill if parameters are adequately informed (e.g., when NEE observations are used for calibration). Otherwise, increased complexity can degrade skill and an intermediate-complexity model is optimal. This finding remains consistent regardless of whether NEE or LAI is predicted. Our COMPLexity EXperiment (COMPLEX) highlights the importance of robust observation-based parameterization for land surface modeling and suggests that data characterizing net carbon fluxes will be key to improving decadal predictions of high-dimensional terrestrial biosphere models.« less
  2. Abstract The terrestrial carbon cycle is a major source of uncertainty in climate projections. Its dominant fluxes, gross primary productivity (GPP), and respiration (in particular soil respiration, R S ), are typically estimated from independent satellite-driven models and upscaled in situ measurements, respectively. We combine carbon-cycle flux estimates and partitioning coefficients to show that historical estimates of global GPP and R S are irreconcilable. When we estimate GPP based on R S measurements and some assumptions about R S :GPP ratios, we found the resulted global GPP values (bootstrap mean $${149}_{-23}^{+29}$$ 149 − 23 + 29 Pg C yr −1 ) are significantly higher than most GPP estimates reported in the literature ( $${113}_{-18}^{+18}$$ 113 − 18 + 18 Pg C yr −1 ). Similarly, historical GPP estimates imply a soil respiration flux (Rs GPP , bootstrap mean of $${68}_{-8}^{+10}$$ 68 − 8 + 10 Pg C yr −1 ) statistically inconsistent with most published R S values ( $${87}_{-8}^{+9}$$ 87 − 8 + 9 Pg C yr −1 ), although recent, higher, GPP estimates are narrowing this gap. Furthermore, global R S :GPP ratios are inconsistent with spatial averages of this ratio calculated from individual sites as well asmore »CMIP6 model results. This discrepancy has implications for our understanding of carbon turnover times and the terrestrial sensitivity to climate change. Future efforts should reconcile the discrepancies associated with calculations for GPP and Rs to improve estimates of the global carbon budget.« less
  3. Abstract Studies exploring long-term energy system transitions rely on resource cost-supply curves derived from estimates of renewable energy (RE) potentials to generate wind and solar power projections. However, estimates of RE potentials are characterized by large uncertainties stemming from methodological assumptions that vary across studies, including factors such as the suitability of land and the performance and configuration of technology. Based on a synthesis of modeling approaches and parameter values used in prior studies, we explore the implications of these uncertain assumptions for onshore wind and solar photovoltaic electricity generation projections globally using the Global Change Analysis Model. We show that variability in parametric assumptions related to land use (e.g. land suitability) are responsible for the most substantial uncertainty in both wind and solar generation projections. Additionally, assumptions about the average turbine installation density and turbine technology are responsible for substantial uncertainty in wind generation projections. Under scenarios that account for climate impacts on wind and solar energy, we find that these parametric uncertainties are far more significant than those emerging from differences in climate models and scenarios in a global assessment, but uncertainty surrounding climate impacts (across models and scenarios) have significant effects regionally, especially for wind. Our analysismore »suggests the need for studies focusing on long-term energy system transitions to account for this uncertainty.« less
  4. This paper describes the formation of, and initial results for, a new FLUXNET coordination network for ecosystem-scale methane (CH 4 ) measurements at 60 sites globally, organized by the Global Carbon Project in partnership with other initiatives and regional flux tower networks. The objectives of the effort are presented along with an overview of the coverage of eddy covariance (EC) CH 4 flux measurements globally, initial results comparing CH 4 fluxes across the sites, and future research directions and needs. Annual estimates of net CH 4 fluxes across sites ranged from −0.2 ± 0.02 g C m –2 yr –1 for an upland forest site to 114.9 ± 13.4 g C m –2 yr –1 for an estuarine freshwater marsh, with fluxes exceeding 40 g C m –2 yr –1 at multiple sites. Average annual soil and air temperatures were found to be the strongest predictor of annual CH 4 flux across wetland sites globally. Water table position was positively correlated with annual CH 4 emissions, although only for wetland sites that were not consistently inundated throughout the year. The ratio of annual CH 4 fluxes to ecosystem respiration increased significantly with mean site temperature. Uncertainties in annual CH 4more »estimates due to gap-filling and random errors were on average ±1.6 g C m –2 yr –1 at 95% confidence, with the relative error decreasing exponentially with increasing flux magnitude across sites. Through the analysis and synthesis of a growing EC CH 4 flux database, the controls on ecosystem CH 4 fluxes can be better understood, used to inform and validate Earth system models, and reconcile differences between land surface model- and atmospheric-based estimates of CH 4 emissions.« less
  5. Abstract

    Global estimates of the land carbon sink are often based on simulations by terrestrial biosphere models (TBMs). The use of a large number of models that differ in their underlying hypotheses, structure and parameters is one way to assess the uncertainty in the historical land carbon sink. Here we show that the atmospheric forcing datasets used to drive these TBMs represent a significant source of uncertainty that is currently not systematically accounted for in land carbon cycle evaluations. We present results from three TBMs each forced with three different historical atmospheric forcing reconstructions over the period 1850–2015. We perform an analysis of variance to quantify the relative uncertainty in carbon fluxes arising from the models themselves, atmospheric forcing, and model-forcing interactions. We find that atmospheric forcing in this set of simulations plays a dominant role on uncertainties in global gross primary productivity (GPP) (75% of variability) and autotrophic respiration (90%), and a significant but reduced role on net primary productivity and heterotrophic respiration (30%). Atmospheric forcing is the dominant driver (52%) of variability for the net ecosystem exchange flux, defined as the difference between GPP and respiration (both autotrophic and heterotrophic respiration). In contrast, for wildfire-driven carbon emissions modelmore »uncertainties dominate and, as a result, model uncertainties dominate for net ecosystem productivity. At regional scales, the contribution of atmospheric forcing to uncertainty shows a very heterogeneous pattern and is smaller on average than at the global scale. We find that this difference in the relative importance of forcing uncertainty between global and regional scales is related to large differences in regional model flux estimates, which partially offset each other when integrated globally, while the flux differences driven by forcing are mainly consistent across the world and therefore add up to a larger fractional contribution to global uncertainty.

    « less