a gis based hurricane prediction model for the gulf coast region

7/28/2019 A GIS BASED HURRICANE PREDICTION MODEL FOR THE GULF COAST REGION

1/31

A GIS BASED HURRICANE PREDICTION MODEL FOR THE GULF COAST

REGION

Venkata B. Dodla1

Sudha Yerramilli2 and Anjaneyulu Yerramilli1

College of Science, Engineering and Technology

Jackson State University

1Trent Lott Geospatial Visualization Research Center2National Center for Biodefense Communications

___________________________________________________________8th International Symposium on Recent Advances in Environmental Health Research

September 20, 2011


2/31

Objective

Prediction of hurricane tracks over the Atlantic Ocean

Basin with special reference to Gulf Coast

Why this study?Current predictions use individualized atmospheric models, which

require at least a few hours (6-hours for NCEP models) to generate aforecast, valid for succeeding 72 hours. Most of the devastation

from hurricanes is caused near the time of landfall. This makes the

prediction of landfall point to be very important. Policy makers as

well as public need a lead time as large as possible for effective

implementation of emergency management and response. The

present study aims to develop a modeling system which can provide

prediction of hurricane track within a few minutes.


3/31

1 nautical mile = 1.15077945 statute miles


4/31

Methodology

Development of a Geographical Information System based model for specific

application to hurricane track prediction. This is an analog model using the

historical track data


5/31

What is GIS? With its embedded relational database component, the system

assists in storing, mapping and analyzing geo-referenced data in an

organized structure.

With its framework, GIS facilitates the integration of various field-

specific applications into its interface.

GIS allows integrated ways to conduct research and develop new

analytical approaches to relate their information to the terrestrial

activities.

The conceptual approach of GIS provides the capability of queryingthe spatial data.

With its inbuilt analytical and geo processing tools, GIS provides

advanced methods for extending Arc GIS functionality and allows to

create models and share them as applications.


6/31

Build up of GIS-based Hurricane Prediction Model using Model Builder

MODEL BUILDER

This model has been developed using model builder

in Arc GIS 10.

The inputs provided from an external source to the

model are shown in blue color. The tool/processes are represented in yellow boxes

and the derived data is represented in green color.

To logically connect the data elements the connect

tool has been used in the model.

This hurricane prediction model has been built by

automating the workflows; by grouping various built-

in and custom tools/script.

TASKS

1. Identifying historical hurricane around active id

and remove duplications of hurricane ids

2. Extracting the complete lifecycle track data for

the extracted points from historical data and

removing track points preceding the buffer region

3. Assigning time id (6hr interval) to each point and

removing outliers.

4. Calculating mean center for the historical track.


7/31

The model gets initialized by a giving a point input data

Buffer tool is used with the linear unit parameters set as per the requirements

(1 degree or 0.5 degrees) to identify hurricane around the active point.

Using clip tool, the hurricane points falling under the specified buffer distance

are extracted from the historical hurricane point dataset.

Using dissolve tool, the hurricane point features with the same ids are

aggregated by choosing the dissolve field as the hurricane ids

The closest point is retained by opting the statistical field to choose the point

with a minimum distance from the active hurricane point (obtained from near

tool)

TASK1


8/31

To retrieve the complete track/lifecycle data for each hurricane point from thehistorical dataset, using Make Query Table, a SQL query is structured to pull the

data, when their ids match with the hurricane ids in the base dataset.

To exclude the track points that are present in the preceding time periods from

the buffer region, a query is built to pick the hurricane ids with hurr_id (time

stamps with 6hrs interval) value greater than the hurr_id value in the featureclass.

With this execution, the hurricane points in the buffer region will have a track

starting from the buffer region till the end of its lifecycle.

TASK2


9/31

In order to assign a sequential number (time_id) to each time point in hurricane(from the starting point to the end of its life period), a custom tool is created in

python script.

This tool reads the hurricane ids in the database and based assigns a time_id

number in a sequential/ascending order with respect to the time period.

Hurricane track points having a number value of 1 represent the first position

of the track that start in the buffer area and the next points with number 2

represents the recorded position of the track after 6 hours in each hurricane.

To remove any outliers, the track points that fall in the range of +/- 20 active

hurricane points with latitude and longitude are selected and the output is

saved to a new feature class.

TASK 3


10/31

The model predicts the projected path for the active hurricane by calculatingthe mean of all the hurricane tracks by grouping the time_id fields.

Using the Mean Center tool, by setting the case field parameters as time _id,

the mean is calculated.

The output from the mean center tool generates a possible projected path forthe active hurricane that can be tracked for every 6 hours.

From this projected path, a 72 hour track path is selected and the final output is

retrieved to a new feature class.

Successful run of the model visualizes the projected path for the active

hurricane on the Arc GIS workspace.

TASK 4


11/31

Prediction of Hurricane Katrina : A Case Study

Why Hurricane Katrina?

Katrina is one of the five deadliest hurricanes ever to strike

the United States, inflicting catastrophic damage andenormous loss of life along the Gulf Coast, mainly effecting

Louisiana and extending into the Florida, Mississippi,

Georgia, and Alabama. Considering the staggering nature of

its impacts, Katrina is noted to be one of the most

devastating natural disasters in United States history.


12/31

Life cycle of Hurricane Katrina (2005)

The life cycle -- 23-30 August 2005

It had three landfalls:

(i) first landfall -- Category 1 -- southeasterncoast of Florida -- 2230 UTC 25 August;

(ii) second landfall -- Category 3 -- Buras-

Triumph, Louisiana -- 1110 UTC 29 August

(iii) final landfall Category 3-- Pearlington,

Mississippi -- 1500 UTC 29 August

Disaster damage

Loss of life: 1836

total property damage: US$81 billion


13/31

Results : Part 1

Sensitivity Experiments

The GIS model has the following parameters which needs to be

chosen. They are

Buffer radius : Radial distance around the active hurricane

location for identification of the previous hurricanes that are

assumed to be analogous to the active hurricane.

Length of the data set : Time duration of the historical hurricane

track data for the Atlantic Ocean Basin 1842-2008

Duration of the season: Length of the hurricane season.

June-July-August-September-October


14/31

Design of Experiments

Buffer radius : 2 experiments

0.5 and 1.0 degrees

Length of the data set : 3 experiments

1842-2004 (complete data set)

1950-2004 (part data set since satellite observations started)

1970-2004 (part data set of last thirty years)

Duration of the season: 2 experiments

Complete - JJASO (Hurricane season for NAO- 5 months)

Part - JAS (3 months with the time of active hurricane at the center.

Since Hurricane Katrina occurred in August, July-August-Septemberwere taken.)

Total number of experiments: 12 (2 * 3 * 2)


15/31

Sensitivity experiments were carried out for only one time point of Hurricane Katrina, 18Z of 27

August 2005. The choice is arbitrary and the sole reason is to make a prediction for 48 hours.

Hurricane Katrina track positions at different times in intervals

of 6-hours. Black color circle shows the chosen time point

(1800 UTC of 27 August 2005 ) for sensitivity experiments.


16/31

A total of 12 sensitivity experiments were performed. Step by

step results are shown here only for the best experiment.

The best results were obtained from the experiment with

Buffer radius: 1.0 degree

Length of data: 1950-2004

Season: July-August-September (3 months)


17/31

All hurricane locations within 1 degree buffer radius of the active hurricane point, left figure shows

all points and right figure is after deleting duplicate points


18/31

Hurricane tracks for the extracted locations within the buffer region, left shows complete

tracks and right picture shows tracks succeeding the buffer point.


19/31

GIS model predicted Hurricane Katrina track starting from 1800 UTC 27 August 2005


20/31

Name Buffer Length of

history data

Seasonal

time

E1 0.5 1842-2004 JAS

E2 1.0 1842-2004 JAS

E3 0.5 1842-2004 JJASO

E4 1.0 1842-2004 JJASO

E5 0.5 1950-2004 JAS

E6 1.0 1950-2004 JAS

E7 0.5 1950-2004 JJASO

E8 1.0 1950-2004 JJASO

E9 0.5 1970-2004 JAS

E10 1.0 1970-2004 JAS

E11 0.5 1970-2004 JJASOE12 1.0 1970-2004 JJASO


21/31

Data

duration

1842-

2004

1842-

2004

1842-

2004

1842-

2004

1950-

2004

1950-

2004

1950-

2004

1950-

2004

1970-

2004

1970-

2004

1970-

2004

1970-

2004

Buffer

radius

(degrees)0.5 1 0.5 1 0.5 1 0.5 1 0.5 1 0.5 1

Season

period JAS JAS JJASO JJASO JAS JAS JJASO JJASO JAS JAS JJASO JJASO

0 7 4 6 7 14 11 4 10 14 15 4 14

6 8 23 33 29 58 15 44 19 58 22 42 30

12 18 51 86 74 118 34 127 57 118 29 110 60

18 57 92 157 137 187 72 228 117 187 64 180 113

24 93 131 226 196 254 102 334 172 254 93 254 161

30 113 165 268 245 137 417 202 142 294 175

46 143 179 310 270 144 486 223 164 349 192

42 174 167 334 265 105 526 216 130 406 170

48 284 228 390 320 65 577 239 109 492 182


22/31

With the choice of the best options for buffer radius=1.0

degrees; historical data=1950-2004; and season= July-August-September, Hurricane Katrina track predictions were obtained

starting at different times from 1800 UTC 26August2005 up to

1800 UTC 28August2005 at 6-hour interval.

A total of 9 experiments were performed. The mean

predicted tracks are shown for all the 9 experiments.


23/31

Predicted tracks of

Hurricane Katrina

from different timepoints along with

corresponding track

observations.


24/31

Track errors (miles)

Date/

Time (UTC)26/18 27/00 27/06 27/12 27/18 28/00 28/06 28/12 28/18

26/18 7

27/00 51 14

27/06 110 72 14

27/12 159 115 56 14

27/18 224 160 89 40 11

28/00 265 219 126 61 15 5

28/06 303 269 129 77 34 25 16

28/12 348 312 149 112 72 59 38 1

28/18 232 370 160 143 102 82 67 37 16

29/00 225 239 148 170 137 94 87 61 25

29/06 227 232 101 138 144 100 102 74 48

29/12 318 203 88 79 105 95 100 106 76


25/31

Track prediction errors (miles)

Start ofprediction/

prediction hour 26/18 27/00 27/06 27/12 27/18 28/00 28/06 28/12 28/18 average

0 7 14 14 14 11 5 16 1 16 11

6 51 72 56 40 15 25 38 37 25 40

12 110 115 89 61 34 59 67 61 48 72

18 159 160 126 77 72 82 87 74 76 101

24 224 219 129 112 102 94 102 106 149 137

30 265 269 149 143 137 100 100 167 166

36 303 312 160 170 144 95 131 188

32 348 370 148 138 105 87 199

48 232 239 101 79 65 143

54 225 232 88 110 164

60

227 203 74 168


26/31

No. of

point

Date/

Time (UTC)

Latitude Longitude Landfall distance

error (miles)

Landfall time

error (hours)

1 26/18 -82.6 24.9 266 -3

2 27/00 -83.3 24.6 248 -3

3 27/06 -84 24.4 74 2

4 27/12 -84.7 24.4 101 -2

5 27/18 -85.3 24.5 107 0

6 28/00 -85.9 24.8 86 -6

7 28/06 -86.7 25.2 128 -8

8 28/12 -87.7 25.7 149 -38

9 28/18 -88.6 26.3 15 +9


27/31


28/31

Summary and Conclusions

1. A hurricane prediction model was developed using GIS tools and

applications.

2. Sensitivity experiments were conducted to identify the best

options for buffer radius, historical data set and seasonal length.

The results indicate that the best options are

Buffer radius: 1.0 degree

Length of data: 1950-2004

Season: July-August-September

with the errors to be in the range of 10-150 miles for 48-hour

prediction.

3. GIS model based Hurricane Katrina track prediction at differentstages have shown the average errors to be 137, 143 and 168

miles at 24, 48 and 60 hours.


29/31

Summary and Conclusions

4. The landfall distance errors are 266 and 101 miles and landfall

time errors are -3 and -2 hours with a lead time of 66 and 48hours respectively.

5. The GIS model results are comparable to official hurricane model

predictions from National Hurricane Center.

6. There are possibilities to improve the present GIS model, by

adding further constraints to identify the analogous hurricane

tracks.


30/31

Acknowledgments

The authors thank

Felix A. Okojie, Vice President for Research Development

Support and Federal Relations.

Shelton Swanier, Assistant Dean. Operations and StrategicInitiatives.

Dr. David Bandi, Director, National Center for Biodefense

Communications


31/31

Venkata B. [email protected] -979-0204

Jackson State University

a gis based hurricane prediction model for the gulf coast region

Documents