Date of Award

January 2014

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Atmospheric Sciences

First Advisor

Gretchen Mullendore

Abstract

Forecast verification remains a crucial component of improving model forecasts, but still remains a challenge to perform. An objective method is developed to verify simulated reflectivity against radar reflectivity at a 1 km altitude utilizing the Method for Object-based Diagnostic Evaluation (MODE) Tool. Comparing the reflectivity field allows for an instantaneous view of what is occurring in simulations without any averaging that may occur when analyzing fields such as accumulated precipitation. The objective method is applied to high resolution 3 km and 1 km local convective WRF summertime forecasts in the Northern Plains region. The bulk verification statistics reveal that forecasts generate too many objects, over-forecast the areal coverage of convection, and over-intensify convection. No noteworthy increases in skill are found when increasing to 1 km resolution and instead lead to a significant over-forecasting of small cells.

A sensitivity study is performed to investigate the forecast biases found by varying the cloud droplet concentration, microphysical scheme, and horizontal resolution on a case day containing weakly forced convection mostly below the freezing level. Changing the cloud droplet concentration has a strong impact on the number of object and area biases. Increasing droplet counts to observed values generates a forecast that more closely resembles the observations in terms of area and object counts, but leads not enough rain generation. Changing the microphysical scheme produces the most pronounced effects on object counts and intensity, which is attributed to differences in autoconversion formulations. Coarsening the resolution from 3 km to 9 km leads to a decrease in skill, showing that 3 km simulations are more effective at convective forecasts. Increasing the resolution to 1 km results in amplifying the object count bias, and is found to not be worth the additional computational expense.

Download

Share

COinS