Investigating sensitivity to dust during the formation of hurricane Helene (2006) using the GEOS-5 Adjoint

Daniel Holdaway

Atlantic tropical cyclones and hurricanes often develop from easterly waves that propagate across the African continent. For disturbances such as these to become sufficiently organized and form tropical cyclones requires certain environmental conditions. These conditions include sufficient heat and moisture as well as an absence of shear. The Saharan Air Layer (SAL) is a region of hot, dry, dust laden air that propagates up from the Sahara desert out over the Atlantic Ocean. As it does so it can modify the environmental conditions that interact with the developing easterly waves. This happens through direct heating and drying but also through the presence of the dust.

Dust can interact with the meteorology via radiative and microphysical processes. A dusty SAL could result in cooler conditions below due to absorption of solar radiation and arguments have been put forward that this coincides with fewer and weaker tropical cyclones. On the other hand the absorbed solar radiation can produce a warming of the SAL, which if entrained may increase the available energy. Other studies have argued that dust has little overall impact and that there exists little correlation between presence of dust and a reduction in storms.

The complex interaction between dust and burgeoning tropical cyclones has been the subject of many studies, often relying on model integrations that either include dust or not. In this work we have developed a framework, using the global GEOS-5 adjoint model, to explicitly examine the sensitivity to dust during the life cycle of tropical cyclones. When using the adjoint model some observed feature, such as cyclogenesis, is chosen as the input. The adjoint then propagates the sensitivity of that quantity with respect to the model variables backwards in time. By including the linearized radiation and GOCART physics the results highlight the regions where changes in dust can have the most dramatic impact through radiative processes. It also immediately describes whether increases in dust will positively or negatively impact the chosen quantity. In this study the effect of dust through microphysical processes is not considered.

The figures on the left of the slide show an example for Hurricane Helene in 2006, a storm that underwent cyclogenesis in a dusty environment. It has been argued that Helene didn’t strengthen as much as it would have in a non-dusty environment. A four day forecast is integrated using GEOS-5 and designed such that the emergence of Helene from the African shore occurs around the end time. The adjoint is initialized with the mean sea level pressure in the box shown (the location of the disturbance that became Helene) and integrated backwards to the initial condition time. The upper panel shows the column integrated dust at this earlier time and the lower panel shows the sensitivity to dust at around 750hPa. Examining the sensitivity field we see both positive and negative sensitivity. Where sensitivity is negative it implies that a positive perturbation to dust would decrease sea level pressure in the box three days later (increase intensity). Where the sensitivity is positive it implies an increase in dust would decrease intensity.

Perturbations coinciding with only the positive or negative regions were generated and new GEOS-5 forecasts were integrated with initial conditions perturbed. It is found that the adjoint predicts the evolution of nonlinear perturbations well. However, it is also shown that the dust has a quite complex affect on the developing disturbance. The perturbations did not just decrease or increase the strength directly but did so by changing the speed of the propagating wave or through the general organization of the system.

We looked at a series of cases for the 2006 season and found quite different sensitivity structures for different storms. We also found that the structure is rather sensitive to the metric and size of the box. All of this suggests a complex relationship with dust. It was not clear that dust will either always decrease or increase the strength of storms but could act to do both and throughout the lifecycle. One clear result is that as the storm becomes stronger the sensitivity to dust decreases and much larger perturbations of dust are required in order to impact the development through radiative processes.

An outline of the adjoint model methodology:

  • Initialize the adjoint for a metric and region of interest, e.g. pressure or dust in the 2D boxes shown above.
  • Integrate the adjoint backwards to ask the question: "what is the metric in this box sensitive to at initial condition time?"
  • Use Lagrange multipliers to convert the adjoint sensitivity into optimal perturbations of dust (or some other quantity of interest).
  • Run a new forecast with the initial fields perturbed by these adjoint derived fields.
  • Compare the original and perturbed forecast to quantitatively examine how the dust impacts the storm dynamics.

« GMAO Science Snapshots