4 | Nature | www.nature.com
Article
Tracking tropical cyclones
Next, we used Pangu-Weather to track tropical cyclones. Given an initial
time point, we set the lead time to be multiples of 6 h (ref. 23) and initi
-
ated Pangu-Weather to forecast future weather states. We looked for the
local minimum of MSLP that satisfied certain conditions, such as the
cyclone eye. The tracking algorithm is described in the supplementary
material for this paper. We used the International Best Track Archive
for Climate Stewardship (IBTrACS) project
24,25
, which contains the best
available estimations for tropical cyclones.
We compared Pangu-Weather with ECMWF-HRES, a strong cyclone
tracking method based on high-resolution (9 km × 9 km) operational
weather forecasting. We chose 88 named tropical cyclones in 2018
that appear in both IBTrACS and ECMWF-HRES. As shown in Fig.4,
Pangu-Weather, forecast time 72 hours
50000
53000
53000
55000
55000
56000
56000
57000
57000
57500
57500
57500
Operational IFS, forecast time 72 hours
50000
53000
53000
55000
5500
0
56
000
56000
57000
57000
57500
57500
57500
ERA5 (ground truth)
50000
53000
53000
55000
55000
56000
6
000
0
0
57000
57500
5750
57500
45,000
50,000
52,000
54,000
56,000
57,000
58,000
Z500
Geopotential (m
2
s
–2
)Wind speed (m s
–1
)Temperature (K)Temperature (K)
Pangu-Weather, forecast time 72 hours Operational IFS, forecast time 72 hours ERA5 (ground truth)
220
250
270
280
290
295
300
T850
Pangu-Weather, forecast time 72 hours Operational IFS, forecast time 72 hours ERA5 (ground truth)
220
260
270
280
290
295
300
305
2-m temperature
Pangu-Weather, forecast time 72 hours Operational IFS, forecast time 72 hours ERA5 (ground truth)
5
10
15
20
25
10-m wind speed
Fig. 3 | Visualization of forecast results. The 3-day forecast of two upper-air
variables (Z500 and T850) and two surface variables (2-m temperature and
10-m wind speed). For each case, Pangu-Weather (left), the operational IFS
3
(middle) and the ERA5 ground truth
18
(right) are shown. For all cases, the input
time is 00:00 UTC on 1 September 2018.
10º N
20º N
30º N
40º N
120º E 130º E 140º E 150º E 160º E
30 September 2018 00:00 UTC 23 October 2018 12:00 UTC
a
Track forecast for Typhoon Kong-rey
Pangu-Weather forecast ECMWF-HRES forecast
Ground truth
0º
10º N
20º N
30º N
110º E 120º E 130º E 140º E 150º E
b
Track forecast for Typhoon Yutu
Forecast time (hours)
24
(788)
72
(492)
120
(214)
50
100
150
200
250
Error (km)
c
Mean direct position error
Pangu-Weather
ECMWF-HRES
Fig. 4 | Pangu-Weather is more accurate at early-stage cyclone tracking
than ECMWF-HRES. a,b, Tracking results for two strong tropical cyclones in
2018, that is, Typhoon Kong-rey (2018–25) and Yutu (2018–26). The initial time
point is shown below each panel. The time gap between neighbouring dots is
6 h. Pangu-Weather forecasts the correct path of Yutu (that is, it goes to the
Philippines) at 12:00 UTC on 23 October 2018, whereas ECMWF-HRES obtains
the same conclusion 2 days later, before which it predicts that Yutu will make
a big turn to the northeast. c, A comparison between Pangu-Weather and
ECMWF-HRES in terms of mean direct position error over 88 cyclones in 2018.
Each number in bracketsin the x-axis indicates the number of samples used to
calculate the average. For example, ‘(788)’ means that there are in total 788 initial
points from which the typhoon lasts for at least 24 hours, and the 788 direct
position errors of Pangu-Weather and ECMWF-HRES were averaged into the
final results. Panels a and b were plotted using the Matplotlib Basemap toolkit.
剩余19页未读，继续阅读
