Estimation of the Surface Fluxes for Heat and Momentum in Unstable Conditions with Machine Learning and Similarity Approaches for the LAFE Data Set

Abstract

Measurements of three flux towers operated during the land atmosphere feedback experiment (LAFE) are used to investigate relationships between surface fluxes and variables of the land–atmosphere system. We study these relations by means of two machine learning (ML) techniques: multilayer perceptrons (MLP) and extreme gradient boosting (XGB). We compare their flux derivation performance with Monin–Obukhov similarity theory (MOST) and a similarity relationship using the bulk Richardson number (BRN). The ML approaches outperform MOST and BRN. Best agreement with the observations is achieved for the friction velocity. For the sensible heat flux and even more so for the latent heat flux, MOST and BRN deviate from the observations while MLP and XGB yield more accurate predictions. Using MOST and BRN for latent heat flux, the root mean square errors (RMSE) are 107 Wm$$^{-2}$$ - 2 and 121 Wm$$^{-2}$$ - 2 , respectively, as well as the intercepts of the regression lines are $$\approx 110$$ ≈ 110 Wm$$^{-2}$$ - 2 . For the ML methods, the RMSEs reduce to 31 Wm$$^{-2}$$ - 2 for MLP and 33 Wm$$^{-2}$$ - 2 for XGB as well as the intercepts to just 4 Wm$$^{-2}$$ - 2 for MLP and $$-1$$ - 1 Wm$$^{-2}$$ - 2 for XGB with slopes of the regression lines close to 1, respectively. These results indicate significant deficiencies of MOST and BRN, particularly for the derivation of the latent heat flux. In fact, in contrast to the established theories, feature importance weighting demonstrates that the ML methods base their improved derivations on net radiation, the incoming and outgoing shortwave radiations, the air temperature gradient, and the available water contents, but not on the water vapor gradient. The results imply that further studies of surface fluxes and other turbulent variables with ML techniques provide great promise for deriving advanced flux parameterizations and their implementation in land–atmosphere system models.

The land–atmosphere feedback observatory: a new observational approach for characterizing land–atmosphere feedback

“…Operational measurements aim for long time series datasets to investigate statistics like the demonstrated correlation between MLH and surface fluxes. Furthermore, the long-term datasets from the EC stations are interesting to use for machine learning approaches to investigate new SL relationships, as demonstrated by Wulfmeyer et al (2022). During IOPs, non-automated instruments complement the operational instrumentation for extended analysis, e.g., analysis of sensible and latent heat fluxes in the ABL (with humidity and temperature measurements from ARTHUS) or SL observations of humidity (with the scanning capability of the WV DIAL) to relate atmospheric moisture distribution to soil water structures.…”

“…Instrument, Soil Vegetation Atmosphere Fluxes Radiation Aerosols Clouds mode DIAL, q(z), dq(z)/dz, q (z), β par (z) vertical q 2 , q 3 DIAL, RHI 2D q, dq/dz 2D 2D scan vertical wind, humidity, and temperature (w * , q * , and θ * ) are determined by combining the scanning lidars with MOST (Späth et al, 2022a) or EC station measurements with MOST (Wulfmeyer et al, 2022).…”

Reducing parametrization errors for polar surface turbulent fluxes using machine learning

“…The algorithm chosen for the Ąnal analysis was the ar-tiĄcial neural network. Neural networks can approximate continuous functions of multiple variables (Hornik et al, 1989) and have performed well in previous studies as estimators of surface turbulent Ćuxes (Pelliccioni et al, 1999;Qin et al, 2005a;Wang et al, 2017;Safa et al, 2018;Xu et al, 2018;Leufen and Schädler, 2019;McCandless et al, 2022;Muijoz-Esparza et al, 2022;Wulfmeyer et al, 2022). Many specialized conĄgurations of network nodes or architectures have been developed for speciĄc applications, for example the convolutional networks used for image recognition (e.g.…”

“…This restriction allows a fair apples-to-apples comparison of the performance of the different parametrizations. Wulfmeyer et al (2022) use incoming and outgoing radiation as predictor variables.…”

“…Pelliccioni et al, 1999;Qin et al, 2005a,b;Wang et al, 2017;Safa et al, 2018;Xu et al, 2018;Leufen and Schädler, 2019;Wang et al, 2021). In recent studies, machine-learning parametrizations have been developed which can accurately estimate turbulent Ćuxes observed at measurement towers (McCandless et al, 2022;Wulfmeyer et al, 2022), and the properties of such parametrizations investigated in large-eddy simulations (Muijoz-Esparza et al, 2022). In this study, we apply data-driven methods to the SHEBA campaign to investigate the viability of this approach in polar conditions, and we assess its performance compared to state-of-the-art bulk formulations over a few other recent Arctic campaigns.…”