Abstract
The assimilation of water vapor mass mixing ratio derived from total lightning data from the Geostationary Lightning Mapper (GLM) within a three-dimensional variational (3DVAR) system is evaluated for the analysis and short-term forecast (≤6 h) of a high-impact convective event over the northern Great Plains in the United States. Building on recent work, the lightning data assimilation (LDA) method adjusts water vapor mass mixing ratio within a fixed layer depth above the lifted condensation level by assuming nearly water-saturated conditions at observed lightning locations. In this algorithm, the total water vapor mass added by the LDA is balanced by an equal removal outside observed lightning locations. Additional refinements were also devised to partially alleviate the seasonal and geographical dependence of the original scheme. To gauge the added value of lightning, radar data (radial velocity and reflectivity) were also assimilated with or without lightning. Although the method was evaluated in quasi–real time for several high-impact weather events throughout 2018, this work will focus on one specific, illustrative severe weather case wherein the control simulation—which did not assimilate any data—was eventually able to initiate and forecast the majority of the observed storms. Given a relatively reasonable forecast in the control experiment, the GLM and radar assimilation experiments were still able to improve the short-term forecast of accumulated rainfall and composite radar reflectivity further, as measured by neighborhood-based metrics. These results held whether the simulations made use of one single 3DVAR analysis or high-frequency (10 min) successive cycling over a 1-h period.
Short-term forecast of Yangtze River water level based on Long Short-Term Memory neural network
