Application of the LEPS technique for Quantitative Precipitation Forecasting (QPF) in Southern Italy: a preliminary study (original) (raw)

Statistical properties and validation of Quantitative Precipitation Forecast

2010

Observed precipitation fields show a high variability both in space and time and the amount of rainfall could vary a lot within a short distance.(Zepeda-Arce et al.,2000). The increasing of horizontal resolution in NWP models seems to enable them to reproduce this variability, even if frequent errors in time and space positioning make difficult a grid-point based employment of models QPF. In order to asses the ability of the models in reproducing the variability of the precipitation fields, we investigated the statistical properties of the observed and forecasted rain values falling within a predefined geographical area and in a specific time period (also called boxes). In particular we studied the distribution function (pdf )and evaluated some summarizing quantities, such as the mean, the maximum value and quantiles for each of the selected box. Results for different size of the chosen areas and period of time are used to validate the QPF of the COSMO suites that run operationally ...

Further Evaluation of Probabilistic Convective Precipitation Forecasts Using the QPF–PoP Neighborhood Relationship

Weather and Forecasting, 2017

A neighborhood postprocessing approach that relates quantitative precipitation forecasts (QPF) to probability of precipitation (PoP) forecasts applied to a single model run was found by Schaffer et al. to be as good as traditional ensemble-based approaches using 10 members in 30-h forecasts of convective precipitation. The present study evaluates if PoP forecasts derived from additional variations of the approach can improve PoP forecasts further compared with previous methods. Ensemble forecasts from the Center for Analysis and Prediction of Storms (CAPS) are used for neighborhood tests comparing a single model run and a traditional ensemble. In the first test, PoP forecasts for different combinations of training and testing datasets using a single model member with 4-km grid spacing are compared against those obtained with a 10-member traditional ensemble. Overall, forecasts for the neighborhood approach with just one member are only slightly less accurate to those using a more tr...