Dust storms in the Sahara and Sahel: From Nowcasting to Climate

Mineral dust aerosol is ubiquitous in Earth’s atmosphere and interacts with many components of the Earth system. The Sahara and Sahel contain the world’s most productive dust sources and frequently experience large dust storms that directly impact human life by disrupting traffic and transport, posing a health risk, and affecting energy delivery from solar energy systems. Early warnings are crucial to reduce adverse impacts. Convective dust storms are difficult to predict with currently operational models because they misrepresent key dynamics, lack surface information, or do not implement processes that control dust emission. Observations can help to better constrain dust models and satellites provide excellent coverage of Sahara and Sahel. This thesis exploits the potential of satellite data to improve short-term predictions of dust storms and to evaluate dust models.

First, an observation-based short-term prediction (“nowcast”) for dust storms is developed, using optical flow extrapolation of false-colour satellite imagery. This baseline model provides skilful dust nowcasts under cloud-free conditions. Then, a generative machine learning model is trained that greatly improves over this baseline, is capable of reproducing complex motion, advection, growth and decay of features. The model generates skilful deterministic and probabilistic predictions of dust storms at 5 hours lead time and of convective storms at 4 hours and demonstrates the potential of machine-learning-based satellite nowcasting of dust storms, and in fact nowcasting of any geostationary image product.

Next, reanalyses ERA5, MERRA-2, and ERA-Interim are evaluated against research-grade station observations from the Sahara and Sahel to determine the most reliable reference winds for dust studies in this region. Height-adjustment of reanalysis winds to observation height reveals that previous studies likely used unrealistic boundary layer stability for this correction for stations in the Sahara. On average, among the products considered, ERA5 best reproduces absolute wind speed, wind power and their diurnal to seasonal variability, whereas MERRA-2 best reproduces diurnal cycles. ERA5 and MERRA-2 for the first time reproduce the timing of downward mixing of low-level-jet momentum but still miss peak winds from cold pools and the overall peak in wind power during the monsoon season.

Finally, an evaluation method for dust in weather and climate models is developed that tests models’ responses in simulated dust patterns to prescribed wind regimes. Various versions of the Met Office and the ECMWF dust models are evaluated at lead times of 2 days and longer. While models do not reproduce the location of observed dust maxima, they reproduce the observed response to different wind patterns. This might suggest that other factors such as surface constraints and dust transport play the more dominant role for dust patterns deviating from observations, but requires more research for validation. The analysis further demonstrates the value of climate simulations at the resolution of today’s NWP that can drastically improve dust patterns with an existing dust scheme. A further resolution increase to grid-spacings of 5 km (“km-scale”) does not necessarily improve dust patterns, suggesting emission parametrisation and surface representation as further key issues on these scales, and indeed the quality of dust-generating winds as a fraction of convection remains parametrised at 5 km grid-spacing.

A key outcome of this research is the nowcast model DustCast that has been tested in collaborative projects with African Met services. This research also provides the foundations for nowcasting any geostationary image product. It further provides references for the reliability of reanalyses and methods that can be used for systematic process-focused model evaluation.