Adjoint-based Predictability Studies of Atmospheric Rivers

The atmosphere is so remarkably complex that determining why a forecasted storm behaves as it does, why it takes a particular path, or why it intensifies, can be extremely difficult. One way to attack this challenging problem is to examine it in a linear framework. Tangent linear and related adjoint versions of numerical forecast models allow us to do that. Adjoint-based tools can provide valuable insight into how distinct atmospheric phenomena on various space and time scales, often separated by substantial distances, may evolve to interact and influence each other. Here we explore initial condition sensitivity and predictability aspects of the series of high-impact atmospheric river (AR) storms that hit the US west coast in early 2017. These storms were associated with both widespread flooding and wind damage. The copious amount of precipitation associated with these storms was enough to end the severe drought conditions over much of California. The adjoint, tangent linear, and nonlinear models for the atmospheric portion of the nonhydrostatic Coupled Ocean-Atmosphere Mesoscale Prediction System (COAMPS) are applied to the sequence of these AR storms. Because the storms resulted in impactful wind and precipitation, we consider both the sensitivity of precipitation and sensitivity of near-surface kinetic energy to changes in the initial state. In agreement with previous studies of mid-latitude cyclones, the adjoint diagnostics indicate that the intensity of both winds and rainfall in these storms is especially sensitive to perturbations in the moisture and temperature fields and to a lesser degree the wind fields. In particular, the adjoint sensitivity results underscore the importance of the low- and mid-level moisture distribution, particularly in filaments of the water vapor fields embedded in ARs. When the synoptic scale flow is complex, as was the case in early January 2017, the sensitivity fields are likewise complex, indicating that the storm evolution is very sensitive to distinct, remote phenomena that eventual combine to impact the conditions at landfall. While both winds and precipitation are most sensitive to low-level moisture, there does appear to be systematic differences in the sensitivity fields. For example, the initial sensitivity regions associated with forecast precipitation occur further upstream than the initial sensitivity regions associated with forecast winds. These systematic differences will be explored further in the presentation. The results of this study underscore the need for accurate moisture observations and data assimilation systems that can adequately assimilate these observations in order to reduce the forecast uncertainties for these high-impact AR events as much as possible, as well as probabilistic approaches to account for inevitable remaining uncertainties.