Challenges in forecasting Hurricane Lorenzo

Linus Magnusson
Jean‐Raymond Bidlot


The 2019 Atlantic hurricane season produced two Category 5 hurricanes on the Saffir–Simpson scale: Dorian and Lorenzo. Dorian formed south‐ east of Barbados on 24 August. It became a Category 5 storm before hitting the Bahamas, where it stalled and caused extensive damage. The cyclone then swept north along the US East Coast. Lorenzo formed on 23 September south of Cape Verde. It reached its maximum intensity on 29 September as the easternmost Atlantic Category 5 storm on record. The storm later weakened as it moved northward but caused problems on the Azores on 2 October, when the harbour on the Flores Island was severely damaged by high waves. Finally, the cyclone hit Ireland on 3 October as an extratropical system.

ECMWF’s medium‐range forecasts predicted Lorenzo’s genesis much better than Dorian’s, and early track forecasts gave a good indication of the cyclone’s northward turn. However, the forecasts underestimated Lorenzo’s propagation speed and there was great uncertainty in track forecasts as the storm approached Europe. Wave forecasts provided early indications of high waves around the Azores.

Weekly mean anomaly of tropical storm strike probability. The map shows tropical storm strike probability anomalies relative to the model climatology for 23 to 29 September in the forecast from 16 September 2019, i.e. one week before the formation of Lorenzo. The anomalies are calculated by subtracting the model climatological probabilities from the real-time forecast probabilities. Blue colours thus indicate a lower probability of tropical storm activity than in the 18-year climatology and red colours indicate a higher probability than in the climatology.

Genesis and track

As shown in the first figure, the genesis of Lorenzo was predicted by ECMWF ensemble forecasts exceptionally early. While the genesis of Lorenzo was predicted with high confidence more than a week in advance, ECMWF forecasts captured the genesis of Dorian only two days in advance. Big differences in predictability were also seen for major hurricanes in 2017. For example, the genesis of Irma was much more predictable than that of Harvey or Maria.

Ensemble forecasts for Lorenzo from before 22 September favoured a more westward track than the outcome. However, as illustrated in the second figure, the cyclone track forecast for Lorenzo from 24 September 00 UTC, issued just after the cyclone was classified as a tropical storm, already predicted a northward turn over the central Atlantic and provided an early indication that the storm might hit the Azores. However, comparing the position in the forecasts valid on 2 October with the observed position, we find that all members placed Lorenzo too far south, indicating a too slow propagation in all ensemble members.

Track forecasts. The charts show tracks for tropical cyclone Lorenzo as predicted by individual ECMWF ensemble members, ECMWF’s high-resolution forecast (HRES) and the ensemble forecast control member (ENS control) starting at 00 UTC on 24 September 2019 (top) and 00 UTC on 28 September (bottom). They also show best track observations at 6-hourly intervals and the predicted and observed position and intensity at 00 UTC on 2 October.

In the ensemble forecast from 28 September, there was good agreement between all ensemble members on a path towards the western Azores. However, here too we find that in most ensemble members the propagation speed was too slow and the predicted position of Lorenzo on 2 October was too far south compared to the corresponding observation. The forecast from 28 September also missed Lorenzo’s rapid intensification the following day (not shown).

After passing over the Azores, medium‐range forecasts indicated large uncertainties in the path. In the forecast from 28 September 00 UTC, we find one group of members going towards the Bay of Biscay, another group towards Ireland and some members even further to the west over the central‐northern Atlantic. This injected a lot of uncertainty into the forecast for western Europe at this point. The cyclone eventually hit Ireland as an extratropical cyclone on 3 October with strong winds.


During Lorenzo’s passage over the Azores, waves were a major issue. The risk for extreme significant wave height south of the Azores between 1 and 3 October was picked up early in the forecast. As the bottom figure shows, as early as 25 October, i.e. a week in advance, the ensemble median was well outside the 99th percentile of the model climate distribution. Forecasts closer to the time predicted a significant wave height of around 14 metres (significant wave height roughly corresponds to the average height of the highest one third of waves). As shown in the figure, this corresponds well to altimeter satellite data available for about 04 UTC on 2 October close to the centre of the storm.

Ensemble wave forecasts. The chart shows the evolution of forecasts of maximum significant wave height for 1–3 October 2019 at 39°N, 31°W. The blue box-and-whisker symbols show ensemble forecasts (ENS) for different starting dates. The red dots indicate ECMWF’s deterministic high-resolution forecasts (HRES).

Wave height forecast for 2 October 2019
Wave height verification. The chart shows the high-resolution (HRES) significant wave height forecast starting at 00 UTC on 2 October 2019 and valid at 04 UTC overlaid with altimeter wave height data from CryoSat-2 (yellow numbers), which was over the area at about 04:17 UTC.


The forecasts for Lorenzo exemplify several challenges. The excellent prediction of the genesis for Lorenzo contrasts with the much poorer forecast for Dorian. This raises the question of how the predictability of tropical cyclone genesis depends on the wider meteorological situation. Lorenzo’s track was well predicted overall, but its predicted propagation speed was too slow compared to the observed speed, and forecasts missed a period of rapid intensification. These are issues which we have seen in several previous cases. Finally, Lorenzo injected a lot of uncertainty into forecasts for Europe during its extratropical transition. There are plans to study all these aspects of hurricane forecasting in greater detail in the coming years to enable us to further improve our medium‐range predictions of tropical cyclones and the weather in Europe.