Enhancing high-resolution storm cell tracking

A multi-threshold TITAN algorithm in synergy with optical flow

technique
Carlos Muñoz Lopez*, Li-Pen Wang and Patrick Willems
KU Leuven, BE (*Carlos.MunozLopez@bwk.kuleuven.be)

1. Introduction 3. Results & Discussion

TITAN (Thunderstorm Identification, Tracking, Analysis, and The performance of the enhanced TITAN tracking algorithm was
Nowcasting) is an object-based storm tracking algorithm, which was assessed by measuring how well an one time-step deterministic
precipitation forecast (making use of the velocity field obtained from
developed for convective rain cell tracking (Dixon and Wiener, 1993).
the tracking process) captures the real radar observations. The ROC
However, due to the single-threshold setting, TITAN was (Receiver Operating Curve) curve analysis was used to evaluate the
not able to well identify adjacent storm Single-Threshold Technique
forecasting result, which allows the investigation of the performance
cell clusters. In addition, its tracking at a range of different rainfall intensities.
algorithm was developed based upon an
overlapping technique, which has proven Relative Operating Characteristic (ROC )curve Hit Rate: False Alarm Rate:
�
to be less effective for capturing fast
Thresholds �
(mm/hr) �= �=
�+� �+�

fast-moving storms and for high spatial-resolution rainfall data.

5
7.5
10
15
20
27 (True Positive) (False Positive)

39 (False Negative) (True False)

2. Methodology
48.5
55
65
75

To improve the applicability of the original TITAN to high-

resolution rainfall data processing, two treatments have been 1
ROC Curve 100 Km x 100 Km

incorporated in this work.

0.9

0.8 1
0.9
0.7

1. Multi-threshold identification. The performance of the proposed 0.6

0.8
0.7
Multi-Threshold Technique

H(Hit Rate)
treatments are evaluated at the areas
0.6
0.5

H(Hit Rate)
0.5 52 dBZ ~ 75 mm/h

Based upon the hierarchical threshold 0.4 0.4

48 dBZ ~ 40 mm/h

centred at Ghent catchment but with 0.3 42 dBZ ~ 15 mm/h

0.3

segmentation (HTS) method (Peak and

0.2
0.2

different extents (i.e. 20, 50 and 100

0.1
0

Tag, 1993), a multi-threshold technique

0.1 0 0.05 0.1 0.15
F(False Hit Rate)

km). The results suggest that both 0

was developed to better identify and

0 0.2 0.4 0.6 0.8 1
F(False Hit Rate)
t re at m e nt s ca n l e a d to i m p rove d ROC Curve 50 Km x 50 km
isolate rainfall cells at small scales. 1

performance, and the combined use of 0.9

Single Threshold Identification dBZ

these two treatments can result in the 0.8

0.7 0.9
1

best performance, which is in particular

0.8
0.6
0.7
H(Hit Rate)

0.5 0.6

evident for moderate and high rainfall

H(Hit Rate)

0.5
0.4 0.4

intensities.
0.3
0.3
0.2
0.2 0.1
0
0.1 0 0.05 0.1 0.15
F(False Hit Rate)
0
0 0.2 0.4 0.6 0.8 1
C-Band (~500 m/5 min) Resolution dBZ
F(False Hit Rate)
ROC Curve 20 Km x 20 Km
1
0 50 Km
0.9

0.8
1
0.7
Multi-Threshold Identification dBZ
0.9
0.8
0.6
H(Hit Rate)

0.7
0.5 0.6
H(Hit Rate)

0.5
0.4
0.4
0.3 0.3
0.2
0.2 0.1

0.1 0
0 50 Km 0 0.1 0.2 0.3 0.4
F(False Hit Rate)
0
0 0.2 0.4 0.6 0.8 1
F(False Hit Rate)

2. Optical Flow integration.

The optical flow technique (a field-
0 50 Km
based storm cell tracker) was employed
to provide an initial estimate of the
storm cell movement. On the basis of
this, a combinatorial optimisation
method based upon overlapping and
dynamic constraints techniques , was
developed in order to match storm cells
at two successive time steps.
4. Conclusions and future work
Results shows that the enhanced TITAN algorithm can better:
ü Handle high-resolution and high intensity storm details.
Final ü Avoid faulty tracking between decaying and new born storm cells.
ü Deal with mergers and splits.
Image It also provides a great basis for future applications, where high-
resolution rainfall estimates are critical; e.g., stochastic spatial-
temporal rainfall modelling and nowcasting for urban hydrology.