Temporal Consistency Optimization for Alpine Lake Turbulent Flux Observations: A Machine Learning Approach

Abstract

Aiming to mitigate the temporal inconsistency in eddy covariance (EC) flux observations, an ultra-wide neural network structure is constructed based on the TensorFlow framework, with which the artificial neural networks (ANNs) are more capable of estimating flux intensity via in-situ micrometeorological features. The EC measurements and micrometeorology observations are conducted at the shore of an alpine lake Yamzho Yumco in southern Tibet Plateau (TP). The performance of the ANNs is evaluated via 10-fold cross-validation. As a result, the simulation bias level exhibits minuscule perturbation over different cross-validation subsamples. As an innovative attempt, the micrometeorological features are selected according to their thermodynamic or kinetic information utilization rather than statistical correlations with the flux intensity. The method providing uncertainty mitigation can be extended to other EC flux measurement experiments, especially in harsh regions like TP, where the environmental conditions do not allow more direct observations.