Object-Based Metrics for Forecast Verification of Convective Development with Geostationary Satellite Data

Abstract

Object-based metrics are adapted and applied to geostationary satellite observations with the evaluation of cloud forecasts in convective situations as the goal. Forecasts of the convection-permitting German-focused Consortium for Small-Scale Modeling (COSMO-DE) numerical model are transformed into synthetic observations using the RTTOV radiative transfer model, and contrasted with the corresponding real observations. Threshold-based segmentation techniques are applied to the fields for object identification. The statistical properties of the traditional measures cold cloud cover and average brightness temperature amplitude are contrasted to object-based metrics of spatial aggregation and object structure. Based on 59 case days from the summer half-years between 2012 and 2014, a variance decomposition technique is applied to the time series of the metrics to identify deficits in day-to-day, diurnal, and weather-regime-related variability of cold cloud characteristics in the forecasts. Furthermore, sensitivities of the considered metrics are discussed, which result from uncertainties in the satellite forward operator and from the choice of parameters in the object identification techniques.