Abstract/Summary

The skill of weather forecasts depends on the initial conditions obtained through data assimilation. Generation of accurate initial conditions for convection-permitting nu�merical weather prediction (NWP) requires assimilation of a large number of near�surface observations of high resolution. To optimally extract information from high�resolution observations, methods of data assimilation must compensate for the un�certainty due to unresolved scales. Here, we examine uncertainty due to unresolved scales from two perspectives: crowdsourced vehicle-based observations and data as�similation. To investigate the potential of crowdsourced observations for convection�permitting NWP, we examine a novel vehicle-based temperature dataset. A new quality-control procedure is developed for the vehicle-based temperature dataset. Approximately 75% of the dataset fails quality-control primarily due to missing or inaccurate metadata. The characteristics of quality-controlled data are explored through comparison with other meteorological datasets. We find that the uncer�tainty of vehicle-based observation-model comparisons is likely weather-dependent and possibly vehicle-dependent. We investigate two different methods to account for observation uncertainty due to unresolved scales in data assimilation. The standard approach includes the uncertainty due to unresolved scales in the observation error covariance matrix. The alternative approach, used by the Schmidt-Kalman filter (SKF), considers the variability of the small-scale processes to estimate the large�scale state. It is shown that the SKF is most suitable in regimes of high uncertainty due to unresolved scales and low instrument uncertainty. The SKF is extended to a novel ensemble transform formulation suitable for nonlinear models and shown to be most beneficial when the uncertainty due to unresolved scales is greater than the instrument uncertainty. We conclude that crowdsourced observations can help fill the gap in near-surface observations for convection-perimitting NWP. Our new ensemble transform SKF has the potential to account for the associated uncertainty due to unresolved scales in their assimilation.