A terrain‐based weighted random forests method for radar quantitative precipitation estimation

Abstract

While weather radar is widely used for quantitative precipitation estimation (QPE) in China and many other countries, the performance of radar QPE is unsatisfactory. A major reason for inaccurate radar QPE is the application of conventional Z–R relationships. In this study the entire vertical profile of reflectivity (VPR) is taken into consideration and a new relationship converting the VPR to rainfall rate is developed. The new relationship is obtained by a proposed terrain‐based weighted random forests (TWRF) method. The TWRF method regards 21 levels of constant altitude plan position indicator reflectivity (from 1 to 18 km) as features. The method consists of two parts: the first is to obtain subregions based on terrain, and the second is to refine the classical random forests method by computing feature weights based on correlation co‐efficients between features and rainfall rate. Radar QPE based on the TWRF method was tested within the 45–100 km range of the radar in Hangzhou, China, on rainfall events in 2014. The proposed method showed improved performance for all verification scores over the Z–R relationship and the classical random forests method. Use of the entire VPR and the terrain‐based study proved to be effective in this example. Experimental results indicate that the proposed TWRF method can improve the accuracy of radar QPE compared to an independent network of rain gauges.

Quantitative precipitation estimation with weather radar using a data- and information-based approach

“…However, in contrast to their approach using a simple theoretical model we utilize an informationtheoretic framework. Similar work was also done by Yang et al (2017), who developed a new relationship converting the vertical profile of reflectivity (VPR) to rain rates using a terrain-based weighted random forest method. The potential advantage of applying a probabilistic relation is that it yields joint statements of both the value of R and the related estimation uncertainty.…”

“…It is further hampered by considerable error and uncertainty arising from measuring the radar reflectivity factor Z (hereinafter referred to as reflectivity) instead of rain rate R, measuring at height instead of at the ground, and many other factors such as ground clutter, beam blockage, attenuation, second-trip echoes, anomalous beam propagation and brightband effects. For a good overview on sources of errors, see Zawadzki (1984) or Villarini and Krajewski (2010).…”

Optimization of X-Band Radar Rainfall Retrieval in the Southern Andes of Ecuador Using a Random Forest Model

“…Alternatively, ensemble methods such as Random Forest have been proven to be robust and easy to implement. This decision-tree derived technique has been efficiently used in remote sensing applications [30] and radar QPE using reflectivity of several heights [31,32]. However, its application is still undocumented on radar rainfall retrieval studies using single Plan Position Indicator (PPI) scans.…”

Calibration of X-Band Radar for Extreme Events in a Spatially Complex Precipitation Region in North Peru: Machine Learning vs. Empirical Approach

“…On the other hand, the derivation of radar QPE by using the random forest algorithm has recently proven to have high potential and promising results [30][31][32], and its application has increasingly been reported in the literature [33][34][35]. This technique has many advantages in comparison with other machine learning methods and has successfully improved the performance of radar QPE data in mountain regions.…”