CH<sub>4</sub> parameter estimation in CLM4.5bgc using surrogate global optimization

Abstract

Abstract. Over the anthropocene methane has increased dramatically. Wetlands are one of the major sources of methane to the atmosphere, but the role of changes in wetland emissions is not well understood. The Community Land Model (CLM) of the Community Earth System Models contains a module to estimate methane emissions from natural wetlands and rice paddies. Our comparison of CH 4 emission observations at 16 sites around the planet reveals, however, that there are large discrepancies between the CLM predictions and the observations. The goal of our study is to adjust the model parameters in order to minimize the root mean squared error (RMSE) between model predictions and observations. These parameters have been selected based on a sensitivity analysis. Because of the cost associated with running the CLM simulation (15 to 30 min on the Yellowstone Supercomputing Facility), only relatively few simulations can be allowed in order to find a near-optimal solution within an acceptable time. Our results indicate that the parameter estimation problem has multiple local minima. Hence, we use a computationally efficient global optimization algorithm that uses a radial basis function (RBF) surrogate model to approximate the objective function. We use the information from the RBF to select parameter values that are most promising with respect to improving the objective function value. We show with pseudo data that our optimization algorithm is able to make excellent progress with respect to decreasing the RMSE. Using the true CH 4 emission observations for optimizing the parameters, we are able to significantly reduce the overall RMSE between observations and model predictions by about 50 %. The methane emission predictions of the CLM using the optimized parameters agree better with the observed methane emission data in northern and tropical latitudes. With the optimized parameters, the methane emission predictions are higher in northern latitudes than when the default parameters are used. For the tropics, the optimized parameters lead to lower emission predictions than the default parameters.

Bio-inspired computation: Where we stand and what's next

“…These two interpolation models are among the most popular ones for feasibility analysis and optimization by virtue of their capability to provide a quantitative measure of prediction uncertainty. This has allowed these models to prevail in many applications, such as design simulation [196] and pharmaceutical process simulations [197] in the case of Kriging, and parameter estimation [198] or water pumping optimization [199] in the case of RBF.…”

Efficient surrogate modeling methods for large-scale Earth system models based on machine-learning techniques

“…Improving predictive understanding of Earth system variability and change requires datamodel integration. For example, Bilionis et al (2015) improved Community Land Model (CLM) prediction of crop productivity after model calibration; Müller et al (2015) improved the CLM prediction of methane emission after parameter optimization; and Fox et al (2009) and Lu et al (2017) improved the terrestrial ecosystem model predictive credibility of carbon fluxes after uncertainty quantification. However, data-model integration methods are usually computationally expensive involving a large ensemble of model simulations, which prohibits their application to complex Earth system models (ESMs) with lengthy simulation time.…”

Calibration of the E3SM Land Model Using Surrogate‐Based Global Optimization

“…The parameter optimization in ELMv0 faces several challenges due to its model complexity, strong model nonlinearity, and significant computational requirements. The strong model complexity and nonlinearity requires a global optimization, as several local optima may exist (Lu et al, ; Müller et al, ). A local optimizer, such as most gradient‐based optimization methods, is sensitive to the initial parameter value, and generally stops at a local optimum if the initial value is not close to the global one.…”

“…However, the use of surrogates in the calibration of land surface models is less common. Müller et al () applied a surrogate global optimization method to estimate 11 CH 4 parameters in the CLM4.5. First, they used radial basis functions to build a surrogate system based on a set of initial samples.…”