Comparative Assessment of Various Machine Learning‐Based Bias Correction Methods for Numerical Weather Prediction Model Forecasts of Extreme Air Temperatures in Urban Areas

Abstract

Forecasts of maximum and minimum air temperatures are essential to mitigate the damage of extreme weather events such as heat waves and tropical nights. The Numerical Weather Prediction (NWP) model has been widely used for forecasting air temperature, but generally it has a systematic bias due to its coarse grid resolution and lack of parametrizations. This study used random forest (RF), support vector regression (SVR), artificial neural network (ANN) and a multi-model ensemble (MME) to correct the Local Data Assimilation and Prediction System (LDAPS; a local NWP model over Korea) model outputs of next-day maximum and minimum air temperatures (T maxtþ1 and T mintþ1 ) in Seoul, South Korea. A total of 14 LDAPS model forecast data, the daily maximum and minimum air temperatures of in-situ observations, and five auxiliary data were used as input variables. The results showed that the LDAPS model had an R 2 of 0.69, a bias of −0.85°C and an RMSE of 2.08°C for T maxtþ1 forecast, whereas the proposed models resulted in the improvement with R 2 from 0.75 to 0.78, bias from −0.16 to −0.07°C and RMSE from 1.55 to 1.66°C by hindcast validation. For forecasting T mintþ1 , the LDAPS model had an R 2 of 0.77, a bias of 0.51°C and an RMSE of 1.43°C by hindcast, while the bias correction models showed R 2 values ranging from 0.86 to 0.87, biases from −0.03 to 0.03°C, and RMSEs from 0.98 to 1.02°C. The MME model had better generalization performance than the three single machine learning models by hindcast validation and leave-one-station-out cross-validation.

Mutual Information Based Learning Rate Decay for Stochastic Gradient Descent Training of Deep Neural Networks

“…While experiments in this paper were developed and tested in an image classification application context, an experiment was conducted to understand if the proposed approach would also work in a regression context. A recently collated temperature prediction data set [ 32 ] was used to test if the proposed approach could train neural network models for regression problems. The paper uses two identification attributes (station and time), fourteen numerical weather prediction model attributes, two in situ temperature observations and five geographical attributes to forecast maximum and minimum next-day temperatures.…”

“… Results of the application of the proposed approach to a temperature prediction (regression problem) data set [ 32 ]. The proposed approach produced a competitive Mean Absolute Error (MAE) of 1.32 C in comparison to the best alternative approach (Adam) which produced an MAE of 0.97 C. Reported numbers are best outcomes of three random seed runs.…”

Exponential calibration for correlation coefficient with additive distortion measurement errors

“…where, z /2 is the quantile satisfying P(N(0, 1) ≥ z /2 ) = /2, and̂2 M is the direct-plug-in estimator of the asymptotic variance 2 . The expression of̂2 M is given as follows.…”

Temperature Prediction Based on Bidirectional Long Short-Term Memory and Convolutional Neural Network Combining Observed and Numerical Forecast Data

“…To solve this problem, a deep learning-based refinement model was proposed in [ 7 ], and the prediction model using the refined data provided better prediction accuracy than the model using data approximated using linear interpolation. The accuracy of future weather prediction can also be increased by augmenting the data using satellite information or combining a variety of types of information [ 8 , 9 ].…”

“…In this case, it should be assumed for the prediction model that all AWS data should include no errors or missed data, which is a situation that is not guaranteed in practice. Recently, there have been several studies indicating that combining numerical forecast data with observed data improved the accuracy of temperature prediction [ 9 ] and aerosol prediction [ 10 ].…”