Hristo Chipilski, University of Reading, UK

The Search for Gaussian Moisture Variables at the Convective Scale


Moisture data assimilation has gained much attention recently, not only due to the continuously growing number of moisture-related observations, but also because of the in-creasing role of moist processes in higher-resolution numerical weather prediction (NWP) models. Unfortunately, moisture control variables (MCVs) are hard to use in a conventional assimilation scheme due to i) their high spatial and temporal variability, ii) their nonlinear interactions with other model variables and iii) the extremely non-Gaussian distribution of their background errors. The last issue is directly connected to the strict physical bounds con-straining the majority of moisture variables. To this end, previous studies have approached the aforementioned issues in two distinct ways – through physically defining a new MCV or applying an appropriate statistical transformation to traditional MCVs. Despite being largely successful, research efforts in the field of moisture data assimi-lation have mainly focused on global scale NWP models. As a result, this presentation exam-ines MCVs at the convective scale. To normalise their error distribution, a Rank-Based In-verse-Normal Transformation (RBINT) method, commonly used in the field of genetics, is introduced. Preliminary results reveal that the proposed method provides a superior Gaussian fit compared to earlier approaches and can be successfully implemented in current data as-similation algorithms.

File available here