Newsletter No. 153 banner

Predictions of tropical cyclones Harvey and Irma

Linus Magnusson
Ivan Tsonevsky
Fernando Prates


At the end of August and the beginning of September 2017, two major hurricanes, Harvey and Irma, were the first in a series to hit the Caribbean and the US. ECMWF forecasts predicted their paths fairly well. In the case of Harvey this helped to predict large amounts of rainfall over Texas. As is common in tropical cyclone (TC) forecasts, the intensity of the hurricanes was less well predicted.


On 26 August (25 August local time) TC Harvey made landfall in Texas. Subsequently the cyclone became quasi-stationary and produced more or less continuous rainfall for three to five days. The rainfall totals reached more than 1,000 mm in the worst-affected areas around Houston, where unprecedented flooding occurred.

The cyclone formed from a tropical disturbance east of the West Indies on 18 August and propagated westward as a fairly weak system. On 22 August it made its first landfall on the Yucatán Peninsula (Mexico) and was downgraded to a tropical depression. After reaching the Gulf of Mexico, the system regained its status as a tropical cyclone. Over the next few days, the cyclone rapidly intensified and became a Category 4 hurricane before making landfall in Texas. After landfall, Harvey became quasi-stationary while gradually weakening to a tropical storm. On 29 August it moved out over the Gulf of Mexico again and a day later it made landfall for the third time, in Louisiana.

Strike probability map for TC Harvey
Strike probability map for TC Harvey. The chart shows the probability that Harvey will pass within a 120 km radius during the next 240 hours, according to the forecast from 18 August. The solid line is the HRES, the dotted line the ENS mean, and the crosses show the path as subsequently observed.

Central pressure forecasts for TC Harvey
Central pressure forecasts for TC Harvey. The chart shows the evolution of central pressure for TC Harvey during the phase of rapid intensification according to ’best track’ data (black) and HRES with different starting times (red).

As early as 18 August, the ensemble forecast was confident about the propagation of Harvey towards southern Mexico. It also indicated that the system might enter the Gulf of Mexico. That risk temporarily decreased on 19 and 20 August when the cyclone was very weak in the central Caribbean Sea. After 21 August, the ensemble was confident that the system would turn into a tropical storm over the Gulf of Mexico but there was considerable uncertainty about where it might make landfall. Between 21 and 22 August, the risk of the cyclone becoming quasi-stationary over Texas increased, and with that came a risk of extreme rainfall. The high-resolution forecast (HRES), in particular, highlighted the risk of extreme precipitation. However, in the short range the predicted area of the worst rainfall was shifted to the southwest compared to the outcome. On 24 and 25 August the cyclone rapidly intensified. This was not captured well by the forecasts before the start of the intensification. However, the total accumulation of rainfall predicted by HRES over Texas was still in the same range as the observed amount.

TC Harvey precipitation forecasts
TC Harvey precipitation forecasts. The charts show accumulated precipitation in the period 26 to 28 August from the US radar network NEXRAD (top left), HRES starting from 26 August (top right) and HRES starting from 23 August (bottom left). The bottom-right plot shows the HRES (red) and ENS (blue) predicted accumulated precipitation in the period 26 to 28 August for Houston for different forecast starting dates. The box-and-whisker symbols mark the 1st, 10th, 25th, 75th, 90th and 99th percentile. Contours show mean sea level presssure as predicted for 27 August 12 UTC.


TC Irma hit several countries along its path in the Caribbean. On 5 and 6 September, the Category 5 cyclone made its first landfall on some of the Leeward Islands. The first to be hit was Barbuda, followed by Saint Barthélemy, Saint-Martin/Sint Maarten and Anguilla. All these islands were crossed by the eye of the cyclone and wind gusts up to 70 m/s were reported on Barbuda. The cyclone later hit the Virgin Islands and the Turks and Caicos Islands. It also affected Puerto Rico and Hispaniola. On 8 September the cyclone hit the Bahamas. It made landfall on Cuba on 8 to 9 September as a Category 5 cyclone. Finally the cyclone made landfall on the southern tip of Florida on 10 September. ECMWF Member States with territories in the area have given positive feedback on the Centre's forecasts. Here we will focus on ECMWF’s predictions for the Leeward Islands and Florida.

The cyclone formed on 30 August west of Cape Verde in the tropical Atlantic. The cyclogenesis was predicted about a week beforehand. The ensemble from 31 August showed a high risk of Irma passing the Leeward Islands six to seven days later. The ensemble was confident that the group of islands would be hit, but there was some uncertainty about the exact track. However, there were large forecast errors in the intensity and wind speed prediction.

The landfall in Florida was much more unpredictable than the landfall on the Leeward Islands. Early forecasts indicated a northward turn at some point, but the exact timing of this made a huge difference for the location where Irma would hit the US coast. Five days before the landfall, in the ensemble starting from 00 UTC on 5 September, possible landfall locations ranged from Louisiana in the west to North Carolina in the east. Later the range narrowed, but even three days before landfall, in the forecast from 00 UTC on 7 September, the tracks ranged from west of the Florida peninsula to east of the peninsula. These scenarios meant there was considerable uncertainty over where the impact of the cyclone would be strongest. On the day of the landfall, there were uncertainties in the final details concerning the strength of the cyclone and the exact landfall location. This created large uncertainty in the storm surge prediction. In the end, the cyclone hit Key West and made landfall just west of the tip of Florida. Meanwhile a storm surge caused some flooding in Miami on the east coast.

Strike probability for TC Irma
Strike probability for TC Irma. The left-hand chart shows the probability that Irma will pass within a 120 km radius during the next 240 hours, according to the forecast from 31 August. The solid line is the HRES, the dotted line the ENS mean, and the crosses show the path as subsequently observed. The right-hand chart shows the predicted longitude for crossing 25°N (latitude of Miami) in successive ENS (blue), HRES (red) and control forecasts (green). The size of each dot is scaled to the predicted strength of the cyclone at the crossing time. The black dot indicates the observed location and the dashed lines show the width of Florida.


Different tropical cyclones pose different types of risks. The primary risk is the winds that damage or destroy buildings, as happened with TC Irma on the Leeward Islands. Another risk is storm surges along coasts, aggravated by waves, as for TC Irma in Florida. A slow-moving cyclone (as in the case of TC Harvey) or a cyclone that hits steep coastlines can also create extreme rainfall accumulations over land leading to potentially devastating flooding. In the case of Harvey the near-stationary nature of the cyclone was the key point to predict.

For both Harvey and Irma, ECMWF forecasts struggled to correctly predict the intensity of the cyclones. This can at least partly be explained by the relatively small scale of tropical cyclones compared to the model resolution. Other phenomena that are difficult to represent and thus limit predictability include eyewall replacements, which temporarily weaken TCs; rapid intensification; intrusion of dry air; and land interaction. These elements are the subject of intense research among tropical cyclone scientists. The processes involved are not yet fully understood and even limited-area models find it difficult to capture them.