H141-0001
A Machine Learning Approach to Estimating Rainfall Rate Based on Simulated Polarimetric Radar Variables

Monday, 14 December 2020
Poster
Kyuhee Shin1,2, Joon Jin Song3, Wonbae Bang2 and Gyuwon Lee1,2, (1)Kyungpook National University, Department of Astronomy and Atmospheric Sciences, Daegu, South Korea, (2)Kyungpook National University, Center for Atmospheric REmote sensing (CARE), Daegu, South Korea, (3)Baylor University, Department of Statistical Science, Waco, United States
Abstract:
Recently, machine learning has begun to be used in remote sensing fields. One of the biggest advantages of machine learning algorithms is its applicability to complex non-linear relationship between response and predictors without any distributional assumption.

In this study, we have explored the potential use of two supervised machine learning algorithms (decision tree and random forest) in rainfall estimation using polarimetric radar variables. The machine learning algorithms are trained by radar variables simulated using observe drop size distribution from 2-Dimensional Video Disdrometer. We have considered several different configurations with different sets of response and predictors and the tuning of random forest algorithm. It is revealed that specific differential phase is the most important predictor to predict rainfall rate while differential reflectivity is the most important to explain residual from Z-R relationship. The models are evaluated by 10-fold cross validation. It is shown that the machine learning algorithms outperform traditional Z-R relationship. After tuning, the maximum value prediction of the random forest is improved. The best model is the random forest model using residual as the response with training set classified by reflectivity.

We have applied the best model in our study to Mountain Myeonbong S-band polarimetric radar data. The rainfall rates for all the pixels have adjusted with estimated residuals and validated with tipping bucket rain gauges in Automatic Weather Stations. The results show that estimated residuals have spatial and temporal variability. Adjusted rainfall rates show good agreement with rain gauges especially at high rainfall rate.

ACKNOWLEDGEMENT

This research is supported by "Development and application of Cross governmental dual-pol. radar harmonization (WRC-2013-A-1)" project of the Weather Radar Center, KMA. This study was funded by the Korea Environmental Industry & Technology Institute (KEITI) of the Korea Ministry of Environment (MOE) as "Advanced Water Management Research Program". (79615)

Previous Abstract | Next Abstract >>