Loading…

Two Random Forest Models for the Non‐Iterative Parametrization of Surface‐Layer Turbulent Fluxes

This study investigated two random forest (RF) models for the non‐iterative parametrization of surface‐layer turbulent fluxes: (a) the RF scheme, a calculation model that is directly trained using correlated variables, and (b) the RF_Li10 scheme, a random forest correction model based on the Li10 sc...

Full description

Saved in:
Bibliographic Details
Published in:Geophysical research letters 2023-11, Vol.50 (21), p.n/a
Main Authors: Yu, Yingxin, Gao, Chloe Yuchao, Li, Yubin, Gao, Zhiqiu
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study investigated two random forest (RF) models for the non‐iterative parametrization of surface‐layer turbulent fluxes: (a) the RF scheme, a calculation model that is directly trained using correlated variables, and (b) the RF_Li10 scheme, a random forest correction model based on the Li10 scheme (Li et al., 2010, https://doi.org/10.1007/s10546-010-9523-y). A comparison between these two new models and the Li10 scheme against an iterative scheme revealed the hierarchy of maximum relative errors in estimating stability parameters, as well as momentum and heat transfer coefficients. This hierarchy is as follows: the Li10 scheme is the greatest, followed by the RF scheme, with the RF_Li10 scheme exhibiting the least errors. Plain Language Summary The computation module for surface‐layer turbulent fluxes is an essential component of numerical weather prediction models. Based on the Monin‐Obukhov Similarity Theory, many parameterization schemes for surface fluxes have been proposed. With the advancement of artificial intelligence, machine learning methods have been applied in meteorology. This study applies the RF model in machine learning to the parametrization of surface‐layer turbulent fluxes. The RF scheme directly calculates the stability parameter after training, and the RF_Li10 scheme is designed to refine the stability parameter derived from the Li10 scheme, by utilizing the RF algorithm for this correction process. In addition, the two new schemes have also been used to calculate the momentum and heat transfer coefficients. The values calculated from the Li10 scheme, RF scheme, and RF_Li10 scheme are compared with the values calculated from the iterative scheme. Under different surface roughness conditions, the average relative errors of the stability parameter obtained from the Li10 scheme, RF scheme, and RF_Li10 scheme are 3.17%, 3.42%, and 0.17%, respectively; the maximum average relative errors of the stability parameter are 5.67%, 3.52%, and 0.52% respectively. Key Points Two Random Forest models (RF and RF_Li10) for the non‐iterative parametrization of near‐surface turbulent fluxes are proposed Compared to the iterative schemes and existing parameterization schemes, the RF_Li10 scheme exhibits the lowest calculation error Compared to the iterative schemes, the RF scheme and RF_Li10 scheme reduce the computation time by 91.0% and 78.4%, respectively
ISSN:0094-8276
1944-8007
DOI:10.1029/2023GL105923