Machine learning-based bias correction for tropical cyclone track simulation of the WRF model over the western North Pacific

The tropical cyclone (TC) tracks are usually simulated with the numerical models, which have an intrinsic error, although the performance of numerical models is continuously improving. Recently, machine learning has been suggested as a good tool to correct the intrinsic error of the model outputs. This study used an artificial neural network (ANN) to correct the error of TC tracks hindcasted by the Weather Research and Forecasting (WRF) model over the western North Pacific (WNP). TCs whose intensity was higher than tropical depression (i.e., tropical storm, severe tropical storm, and typhoon) from June to November were hindcasted, and TC positions at 72 h were set as the target of bias correction. WRF model output, best track data, and wind field of reanalysis were used as input variables of ANN. The structure of ANN was optimized for TCs during 2006-2015, and the optimized ANN was verified for TCs from 2016-2018. In the verification of ANN, TCs were classified using k-mean clustering to analyze the results of bias correction because the performance of the numerical model for the TC track varied depending on the region of WNP. The ANN corrected the error of WRF by 8.81% for four clusters where ANN was most effective. Moreover, the post-processing was applied to other clusters with less effect of ANN. Consequently, ANN with post-processing improved the accuracy of WRF by 4.34%.