APPLICATION QUANTILE REGRESSION METHOD TO FORECAST HEAT WAVES IN HANOI WITH 5-DAY AND 10-DAY FORECASTS

Abstract

The paper presents the results of applying the quantile regression method to forecast heat waves in Hanoi. The study uses the maximum temperature (Tx) in the months of May, June, July, August and September in 20 years (1999-2018) observed at the Hanoi station and the 5-day and 10-day drought forecast data of the sub-seasonal to seasonal forecast model (S2S). The above data series is used to build the forecast equations on different quantiles (q10, q25, q50, q75 and q90) and the linear regression equation. The results of the equation testing show that the linear regression equation and the equations at quantiles 10, 25, 50, 75 and 90 are statistically significant. The study evaluates the quality of the equations based on the data series of heat waves in 3 years 2019-2021 with a total of 45 waves corresponding to 179 days (samples) based on the statistical indexes ME, MAE, RMSE and relative error. The results show that the 5-day and 10-day forecasts, the 75th and 90th percentile regression equations improve by over 20% compared to the error of the S2S model and improve the error by about 6% compared to the linear regression method. In addition, the Tx forecast error values ​​of the 75th and 90th percentile models for 5-day and 10-day periods are both below 2^oC. Thus, the 75th and 90th percentile regression models can be applied to forecast Tx temperature operations for the Hanoi area.