Short-Range QPF over Korean Peninsula Using Nonhydrostatic Mesoscale Model & "Future Time" Data Assimilation Based on Rainfall Nowcasting from GMS Satellite Measurements

This study investigates data assimilation impacts of near-term satellite-derived nowcasted rainfall information (i.e., 3-hourly independently forecasted rainrates) in the initialization phase of a nonhydrostatic mesoscale model used for predicting severe convective rainfall events over the Korean peninsula. Infrared (IR) window measurements from the Japanese Geostationary Meteorological Satellite (GMS) are used to specify downstream rainrates during a spinup period of the model -- but in a future time framework since the independently acquired forecasted rainrates are used as assimilation control after the pre-forecast period has expired. This is tantamount to initializing in advance of real time in the context of operational forecasting. The main scientific objective of the study is to investigate the strengths and weaknesses of this data assimilation scheme insofar as its influence on quantitative precipitation forecasting (QPF) during the summertime Korean rainy season. Although there have been various recent improvements in formulating the dynamics, thermodynamics, and microphysics of mesoscale models, as well as computer advances which allow the use of high resolution cloud-resolving grids and explicit latent heating over regional domains, spinup remains at the forefront of unresolved mesoscale modeling problems. In general, non-realistic spinup limits the skill in predicting the spatial-temporal distribution of convection and precipitation, primarily in the early hours of a forecast, stemming from the inability of standard synoptic plus sub-synoptic observations to represent the initial diabatic heating field being produced by the ambient convection and cloudiness. The long-term goal of the research is to improve short-range (12-hour) QPF over the Korean peninsula through the use of innovative data assimilation methods based on geosynchronous (GEO) and/or low earth orbit (LEO) satellite precipitation information, either IR or preferably time lapse microwave (MW) measurements, once the latter are introduced by the forthcoming Global Precipitation Measurement (GPM) Mission. As a step in this direction, a new type of data assimilation experiment is performed in conjunction with high-frequency GMS-retrieved nowcasted rainfall information introduced to a mesoscale model. The 3-hourly "future time" precipitation forecast information is assimilated through nudging the associated moisture field (and thus the latent heating) during the early stages of a forecast period. This procedure enhances details in the moisture field during model integration, and thus improves spinup performance, as long as the error statistics of the future time precipitation estimates are superior to those intrinsic to background error statistics of the model. To incorporate the nowcasted rainrates, a nudging-based rainfall data initialization scheme using accumulated mean rainfall based on the future time rainrate information is applied during the first three hours of each 12-hour forecast period. To invoke this scheme more effectively, Newtonian relaxation is applied on the model's dynamic and thermodynamic variables during the preforecast period and before invoking the future time rainrate assimilation, which is a process referred as dynamic nudging. This procedure limits large–scale error growth and allows development of large-scale balance in the model's prognostic variables by guiding the relaxation toward the ambient large-scale analysis. The following numerical experiments are then performed: (1) control (CTL) – without any data assimilation, (2) rain assimilation (RAIN) – rainrate nudging only during the initial forecast period with three hours of nowcasted rainrates, and (3) rain assimilation with dynamic nudging (DYNRAIN) – nudging of winds and temperatures to the large-scale analysis for six hours during the preforecast period, and then nudging the model-forecasted rainrates with three hours of nowcasted rainrates. The integration cycle of the experiments consists of a 12-hour preforecast period prefacing a 12-hour forecast period, thus defining a complete 24-hour model integration period. The above methods are tested on three flood-producing storm cases that took place over South Korea during the summer seasons of 1998, 1999, and 2000. For purpose of validating the GMS precipitation nowcasts, one-minute sampled raingauge measurements are used. These data were acquired from the dense operational Automatic Weather Station (AWS) network (i.e., some 530 raingauges distributed over South Korea) maintained by the Korean Meteorological Administration (KMA). These experiments help shed light on how operational precipitation forecasts made during the Korean rainy season could possibly be improved by applying a GEO satellite-based (or possibly a ground-radar network-based) data assimilation scheme in which "future time" rainrate conditions would be represented at a prediction model's initialization time. It is found that assimilation of nowcasted rainrates alone during the early hours of a forecast period, produces better precipitation forecasts for low to medium rainrates, as well as better organized vertical velocity fields, than generated in the CTL experiments. Application of the dynamic nudging procedure during the preforecast period produces even better precipitation forecasts vis-à-vis rain location and intensity, especially for medium to heavy rainrates. Thus, combined use of dynamic nudging during the preforecast period and future time rain assimilation during the forecast period produces superior forecasts relative to CTL, RAIN only, or dynamic nudging only (i.e., DYNRAIN assimilation without follow-on RAIN assimilation). Forecast skill, quantified by threat and skill scores for heavy rainrates, are improved in the DYNRAIN experiments, although the bias scores for the DYNRAIN experiments are only slightly larger than for the RAIN experiments. The impact of the assimilation scheme depends to some degree on the characteristics of the precipitation events. The 2000 case study undergoes a greater combined impact of dynamic nudging during the preforecast period and rain assimilation during the forecast period, relative to associated impacts for the 1998 and 1999 case studies. Overall, the analysis suggests that the combined nudging procedure, as embodied in the DYNRAIN scheme, would lead to measurable improvements in mesoscale model-based QPF from convective storms over the Korean Peninsula.