[p2] Observational issues


Forecast Verification by Pattern Recognition with Integrated Precipitation Areas

T. Awazu (RIKEN), S. Otsuka (RIKEN), and T. Miyoshi (RIKEN)


The human eye recognizes many patches of rainfall areas as a single precipitation system. However, precipitation forecast verification methods usually recognize individual patches. To mimic the human recognition, it is desirable to combine the individual patches into a single precipitation system. Hence, we propose a verification method using the integrated rainfall areas. In addition, the proposed method evaluates the location error and shape of the rainfall areas. Grid-based verification methods such as the Threat Score (TS) and Root Mean Squared Error (RMSE) are commonly used but have difficulties to evaluate the location error and the shape. Thus, the proposed method evaluates the integrated rainfall areas using the shape features, precipitation rate and the distance between the areas of the observation and the forecast. This study used the forecast data of space-time extrapolation in Global Satellite Mapping of Precipitation (Otsuka et al. 2015). The evaluation of the forecast data was compared between the proposed method and the traditional methods. The results showed that with longer lead time the TS and RMSE change only slightly, while the proposed score gets worse linearly with time. This is more consistent with human recognition.