Storm surge modelling with geographic information systems: estimating areas and population affected by cyclone Nargis

Authors


Abstract

A methodology has been developed to model storm surges during cyclones, and to estimate affected areas and population with Geographical Information System (GIS). Surge profiles were derived from the calibrated wind velocity distribution by using regional regressive relationships between the wind velocities and surge heights. Discretised forms of the calculated surge profiles were entered in GIS analysis to estimate affected area and population. This approach was compared with a simple and more common GIS approach that uses a constant storm surge height. The study also considers the effects of different time advisories on the estimation of affected population and areas. It was found that for the cyclone ‘Nargis’, the affected area and population could be estimated with accuracies of 60 and 20%, respectively, for advisories issued when the cyclone centre was more than 600 km from the landing point. These accuracies were 80 and 60%, respectively, when the cyclone centre was within 200 km of landfall. The proposed methodology can be helpful for early warning systems and disaster risk management due to its simplicity and speed as well as in cases where minimal information on the affected areas and population is available such as in the case of cyclone Nargis in Myanmar. Copyright © 2010 Royal Meteorological Society

1. Introduction

Coastal regions face an immense threat of property damage and loss of life from cyclones. Together with sustained winds and heavy rains, cyclone-induced surges are one of the main sources of casualties and damage. Surge water can penetrate up to 50 km from the coastline, causing salinisation of fertile farm lands and resulting in death tolls in the thousands (Russo, 1998; Conner et al., 1957). Improved forecasting of the devastating effects of cyclones provides for early warning opportunities and more timely disaster management strategies (Joseph, 1994; TCG, 2008; Madsen and Jakobsen, 2004; Katz and Murphy, 1997).

Early studies on surge estimations date back to the 1950s and used the central pressure of cyclones to predict maximum surge height (Hoover, 1957; Conner et al., 1957); Tancreto (1958) improved on these methods by stratifying the cyclones according to their wind direction to better estimate surge heights, and Chan Walker (1979) improved surge height estimations by using approach paths. Pore (1964) used multistation surge models to predict surge heights, and Jelesnianski (1972) developed a methodology for estimating maximum surge heights via nomograms that used maximum wind, radius of the maximum wind, direction of landing, bathymetry, and central pressure data. Russo (1998) attempted to formalise Jelesnianski's nomographs by defining a bathymetric correcting factor based on geographical coordinates. Within the past several years, Artificial Neural Network (ANN) models have been applied to estimate surge heights using maximum wind velocity, wind direction, and central pressure (Lee, 2006). On the basis of tidal records, Jan et al. (2006) used the distance between tidal station and typhoon centre in conjunction with the central pressure, maximum wind velocity, and maximum wind radius, to estimate surge heights. Huang et al. (2007) proposed a regressive surge model based on the Jan et al. (2006) approach, however, it is difficult to reconstruct a surge profile using this type of model because calculated surge heights are valid for only a specific location.

Cyclone advisories, which are issued by different agencies, generally provide maximum wind velocity, central pressure, radii of at least three wind velocities, and landing point/direction estimation (Vatvani et. al., 2002; Joseph, 1994). Using such parameters from cyclone advisories in spatially distributed Geographic Information System (GIS) models is crucial for building timely operational models and providing information to guide early warning and response efforts. Most GIS-based models of cyclone risk are based on storm probability and provide only a general estimation for probable cyclone damage (Puotinen, 2007; Hossain and Singh, 2003). Event-based GIS models typically assume uniform surge height and are applied for the actual landing point which is known only after the cyclone has made landfall (e.g. Gorokhovich and Doocy, 2008). Their simplicity and speed are advantageous, however, they bear significant uncertainties which are very important drawbacks for emergency managers (Zerger, 2002; Zerger et al., 2002). Even though estimations of event-based models can be improved by using Light Detection and Ranging (LIDAR) elevation data (Webster et al., 2004), the assumption of uniform storm surge height remains a principal drawback for these models. With the exception of complex hydrodynamic models that require heavy parameterisation, calibration data, and skilled personnel, non-GIS-based numerical analyses have focussed on estimation of the surge height rather than its spatial distribution (Huang et al., 2007; Jain et al., 2006; Jan et al., 2006; Kumar et al., 2003). Though hydrodynamic models are spatially distributed and physically based (Jain et al., 2006; Madsen and Jakobsen, 2004; Flather, 1994), they cannot be readily applied in emergency contexts and are more effective in later stages of post-disaster analysis and mitigation planning.

The GIS methodology proposed in this study estimates affected areas and population, using a spatially distributed surge profile during a cyclone. The method aims to increase the reliability of real-time GIS models by using cyclone wind velocity distributions generated from advisory data for estimation of the surge profile throughout the coastline. The methodology was applied for cyclone Nargis, which landed in Myanmar on 2 May 2008 and which is considered the deadliest cyclone since 1970, causing over 138 000 deaths and widespread damage and displacement (TCG, 2008). The surge profile was estimated for different time advisories and discretised to make it compatible with the Shuttle Radar Topographic Mission (SRTM) data. Affected area and populations were estimated from digital elevation and population data within surge buffers that were created using the actual landing point and surge profile derived for the latest advisory. Results were compared with the output of a simple GIS model relying on a constant surge height that was estimated by the central pressure of the cyclone. The proposed GIS methodology was also applied for surge profiles estimated from different time advisories, using both actual and advised landing points to analyse the effects of different surge profiles and landing-point locations/directions on estimates of the affected area and population.

2. Methods

2.1. Study area and data

Coastal regions of the Bay of Bengal are frequently affected by cyclones with a decadal frequency of tropical cyclones (with wind velocities of 34 knots or greater) in the Bay of Bengal, of approximately 50 (Joseph, 1994). In the case of cyclone Nargis, damage assessments and relief efforts were restricted by the military junta and survivors faced many challenges in the months following the cyclone because of limited humanitarian assistance. Simple GIS models relying on post-disaster data provided a useful tool for disaster managers to identify the most affected areas and populations.

National Hurricane Center (NHC) data archive (NHC, 2008) and cyclone advisories provided by the Indian Meteorological Department (IMD) were used to build wind velocity distributions. Table I shows winds velocities and radii provided by different time advisories during the Nargis development. Nargis's original and advisory paths are shown in Figure 1. Since historical records for Myanmar's coastline do not contain parallel wind velocity and storm surge measurements, the data for the Gulf of Mexico were used to define the regressive structure of surge models based on the wind velocities (NDBC, 2008; CFHC, 2008). Only few stations have valid data (Table II), since many stations were damaged during strong winds, or because the distance between available station and hurricane centre was too far to measure extreme wind velocities. Archiving, Validation and Interpretation of Satellite Oceanographic Data Center (AVISO) produced the sea anomalies product for the cyclone Nargis that was used to calibrate the regressive surge model (hereafter called the wind-based surge model). The SRTM 3 arc-second (90 m resolution) data provided by CGIAR-CSI was used to create a digital elevation model of the affected coastal region. Population data from the Gridded Population of the World, version 3 (GPWv3), which is based on the most recent census and has spatial resolution of 1 km, was used to estimate the affected population (SEDAC, 2008).

Figure 1.

Myanmar; Nargis: actual and advisory tracks. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Table I. Wind data provided by different time advisories
No Radii of 17.5; 32.9; 25.7 m/s winds (km)
 Geographic PositionMax WindMax GustNESESWNW
 NEm/sm/s123123123123
114.486.939464683157468316746831394683130
215.087.536444674157467417646741674674157
315.889.839463774176377419437741763774157
415.990.746575683185568320456831855683167
515.792.651645611120456120241561022044693194
615.993.7597256111194561302415613024156111213
Table II. Meteorological stations and hurricanes recorded
StationStormFromToMax wind (m/s)Min pressure (mb)
PTAT2Hurricane Bret8/18/19998/24/199965944
 Hurricane Emily7/10/20057/21/200569925
GDIL1Hurricane Lili9/21/200210/4/200265940
DPIA1Hurricane Katrina8/23/20058/30/200578902
 Hurricane Rita9/17/20059/24/200578897
 Hurricane Ivan9/2/20049/24/200474910
LONF1Hurricane Wilma10/15/200510/25/200578882
 Hurricane Katrina8/23/20058/30/200578902
 Hurricane Charley8/9/20048/15/200465941
SMKF1Hurricane Wilma10/15/200510/25/200578882
 Hurricane Jeanne9/13/20049/27/200454950
 Hurricane Katrina8/23/20058/30/200578902

2.2. Methodology and application

Simple GIS models of storm surge height generally rely on a constant surge height calculated from the central pressure of the cyclone. However, the surge height is dependent on a multitude of factors including central pressure, wind velocity, forward speed of the cyclone, bathymetry, and tidal level of the coastline (Conner et al., 1957). Central pressure has a minor direct hydrostatical effect on surge height, but there are also major indirect effects from winds. Surge height is not uniform along the coastline due to the physical structure of a cyclone and the effects of both bathymetry and topographic irregularities. To account for wind distribution and variable surge height, wind models calibrated by advisory data can be used to estimate the surge profile for any cyclone position; GIS models can subsequently be applied to estimate the affected land area and population. This methodology is illustrated in Figure 2 and includes the following sequence of procedures: (i) create a mathematical model of relationship between wind velocities and surge heights (i.e. wind based surge model), using pre-cyclone data; (ii) calculate the wind velocity distribution using advisory data; (iii) derive the surge profile from the wind velocity distribution, using the wind-based surge model; and (iv) estimate affected land area and population from digital elevation and population data using surge buffers that were created for the discrete surge profile according to the advised landing point and direction.

Figure 2.

Summary of the proposed GIS methodology

2.3. Storm surge estimation

2.3.1. Wind velocity distribution

Various models have been developed to estimate wind velocity distribution during cyclones by Xie et al. (2005), Bao et al., (2004), Houston et al. (1999), Shapiro (1983), Anthes (1982), Holland (1980) and Depperman (1947). Simple parameterised models of Holland and Anthes are widely used in recent simulations of wind fields (Hsu and Babin, 2005; Xie et al., 2005; Bao et al., 2004; Vatvani et al., 2002). In this study, both Holland (1980) and Anthes (1982) models (Equations (1) and (2), respectively) were used to obtain the wind velocity distribution. In the Holland and Anthes models, parameters B, x, Rmax and vmax should be determined empirically or estimated from the advisory data.

equation image(1)
equation image(2)

where: v(r) is the wind velocity at radius r;

  • vmax is the maximum wind velocity;

  • Rmax is the radius of maximum wind;

  • B and x are model parameters;

  • Po and Pc are ambient and central pressures;

  • ρ is the density of the air;

  • f is the Coriolis factor.

Table III summarises empirical equations given in the literature for the estimation of the maximum wind velocity during cyclones. Efficiencies of these maximum wind models were tested using the data from the global hurricane archive (NHC, 2008) and compared with the following model's least square estimations derived from this archive: vmax = a + b(PoPc)0.5 + cVf + d.f where: Vf is the forward speed of the hurricane; a, b, c and d are parameters to be determined empirically.

Table III. Relationship between observed and estimated maximum winds
  ModelsParameters usedReferenceGlobalIndian Ocean
     R2SR2S
Literature1vmax = 3.44(1010 − Pc)0.644 Atkinson & Holiday (1977)93.098.3293.318.16
 2vmax = 6.3(1013 − Pc)0.5 Athens (1982)92.814.3393.284.83
 3equation imagePo = 1010 mbKumar et al. (2003)91.2311.691.4311.84
   equation image 
 4vmax = 7(Po − Pc)0.5Po = 1010 mbNatarajan & Ramamurthy (1995)91.684.2892.514.38
 5vmax = 0.856Umax + 0.5Vf;Umax = 0.447[14.5(Po − Pc)]0.5Rmax(0.31f)Rmax = 59 km; Po = 1010 mbUSACE (1984)90.926.0393.076.49
 6vmax = [B(Po − Pc)/ρe]0.5Po = 1010 mb; ρ = 1.15 kgm−3; B = 1Holland (1980)91.687.7592.517.31
Fitted7vmax = 4.78 + 6.58(Po − Pc)0.5 + 0.142Vf − 55225fPo = 1010 mb 92.023.9 4.43
 8vmax = 2.89 + 6.52(Po − Pc)0.5Po = 1010 mb; c = 0; d = 0 91.54.02 4.42
 9vmax = 2.55 + 6.51(Po − Pc)0.5 + 0.068VfPo = 1010 mb; d = 0 91.514.02 4.38
 10vmax = 6.79(Po − Pc)0.5 + 0.283VfPo = 1010 mb; a = 0; d = 0 91.344.12 4.22
 11vmax = 7.1(Po − Pc)0.5Po = 1010 mb; a = 0; c = 0; d = 0 91.54.23 4.36

Model efficiency was evaluated by the determination coefficient R2 and standard error S between the observed and estimated maximum winds. Maximum wind velocity changes with the migration of the cyclone position (Figure 3). In order to approximately determine wind velocity distribution as closely as possible to the values provided by advisories, four different models were used and are summarised in Table IV. Ideally, the wind velocity distribution perpendicular to the cyclone path should be used to calculate the surge profile. Because data required to calculate the wind velocity distribution perpendicular to the cyclone path is not always available, wind velocity distributions calculated for the advisories directions closest to the direction perpendicular to the cyclone path were used. The wind velocity distributions calculated by the models described above for the northeastern direction of Nargis advisories 5 and 6, which were the closest advisories to the Nargis landing point, are shown in Figure 4. The fourth model was the simplest and best-fitting model and was used to estimate wind distribution of cyclone Nargis (Figure 5).

Figure 3.

Variation of the maximum wind velocities during cyclone Nargis

Figure 4.

Wind profile obtained by different models

Figure 5.

Wind velocity distributions obtained by the model 4 for advisories 5 and 6

Table IV. Wind model parameters
AdvisoryPositionModel 1Model 2Model 3Model 4
 NEVmaxRmaxVmaxRmaxBRmaxxRmax
       NESESWNWNESESWNWNESESWNWNESESWNW
  1. Model 1: Holland model with observed value of Vmax and average value of Rmax = 59 km (proposed for the Indian ocean cyclones by Kumar et al., (2003)). Model 2: Holland model with the observed value of Vmax and Rmax ≈ 21.5 km. Model 3: Holland model calibrated using wind advisories. Model 4: Anthes model calibrated using wind advisories

114.486.938593821.51.201.171.291.3317.517.517.517.50.500.480.540.5734.534.035.536.0
215.087.536593621.51.231.151.191.2317.017.017.017.00.490.470.480.4937.037.037.037.0
315.889.839593921.51.010.971.011.0615.015.015.014.50.460.460.460.4628.028.528.027.5
415.990.746594621.51.251.161.251.3420.019.520.020.00.540.500.540.5829.027.029.030.5
515.792.651595121.51.131.031.151.0721.522.521.018.50.460.460.470.4621.522.522.018.5
615.993.759595921.51.151.031.031.1221.523.523.522.00.470.460.460.4616.517.017.016.0

2.3.2. Wind-based storm surge model

Empirical methods of storm surge height estimation available from the literature (Huang et. al., 2007; Hsu and Babin, 2005; Chan and Walker, 1979) do not provide a surge profile in a form that can be used in GIS analysis, and it is difficult to build a hydrodynamic surge model for the affected coastal region before cyclone landfall. However, the regressive relationships between cyclone wind velocities and surge heights can be used to derive a surge profile prior to landfall and are advantageous because they are applicable for any point along the coastline and do not require intensive calculations. In cases where historical data are not available to build regression models, they can be developed using output from hydrodynamic models of the region prior to cyclone season; several hydrodynamic models characterising the Bay of Bengal have been developed (Jain et al., 2006; Madsen and Jakobsen, 2004; Vatvani et al., 2002; Flather, 1994)

Meteorological data of the stations in Table II were used to determine the relationships between the observed values of wind and surge. To eliminate the effects of wave-generated fluctuations on regressive models, 12-h moving average of surge records were applied. Only two stations were found to have more than one significant hurricane record; on these stations, observed wind velocities and surge heights were plotted reciprocally. Fitted regression models for reciprocal plots are shown in Table V indicating that surge height can be readily modelled by linear regressive models with high-determination coefficients obtained especially for velocities greater than 12.5 m/s. Lower wind velocities are not effective for GIS modelling since they produce surge heights less than one meter that are outside of the vertical precision of SRTM data.

Table V. Wind-based surge models
StationModel noModelR2 (%)Valid for
  1. h: surge height (m); v: wind velocity (m/s)

LONF11h = 0.0016v2 + 0.040796.29v > 0
 2h = 0.0643v − 0.573694.48v > 12.5
 3h = 0.0018v2 − 0.011394.69v > 12.5
DPIA11h = 0.0511v + 0.205691.77v > 0
 2h = 0.0432v + 0.357493.04v > 12.5
 3h = 0.0011v2 + 0.754491.38v > 12.5

Simultaneous historical in situ wind and surge measurements along the Myanmar's coastline are not available; synthetic wind velocity and surge height data from hydrodynamic models of the region were also unavailable. The sea anomalies merged altimetry product from the AVISO data centre created for the period 28 April to 2 May 2008 (NHC 2008) was used to calibrate the wind-based surge model of the Myanmar coast. The profile extracted for the northwestern–southeastern direction of Advisory 6 was used for this purpose. The calibrated wind-based surge model (Equation (3)) was assumed to be valid along the Myanmar coast, and was used to determine surge profiles for each advisory. Figure 6 shows the comparison between surge profiles calculated for each advisory and surge profiles extracted from the AVISO product.

equation image(3)

where h: surge height (m), and V: wind velocity (m/s). Before using calculated surge profiles in GIS analysis, they should be discretised to create a set of input elevation values for spatial analyses. The proposed set of discretisation points includes 0.5, 1.5, 2.5, and 3.5 m surge heights.

Figure 6.

Surge profiles calculated based on different advisories

2.4. GIS-based storm surge models

Two GIS models, one based on constant surge height and the other based on the spatially distributed surge profile, are introduced together with an evaluation of the effects of the different time advisories on the results of proposed GIS models. Since the results of the GIS model are about the affected areas and population, an initial assessment was conducted using elevation and population data of the coastal states of Myanmar to define focus areas and understand the distribution and ranges of elevation and population data (Figure 7).

Figure 7.

Demographic and areal structure of Myanmar's coastal states

2.4.1. GIS model based on constant surge height

Simple GIS models based on the constant surge height use the coastal zone area within a specific elevation range for estimation of the affected area and population. These models do not consider the location of the landing point and the extent of the cyclone effects, and so, yield very general results. The empirical model given in Equation 4 was used to calculate maximum surge height during Nargis (Conner et al., 1957). The affected area and population were estimated by a GIS model that used the calculated maximum surge value (Table VI).

equation image(4)

where: hmax—maximum surge height (m); Po—Central pressure of cyclone (mb)

Table VI. Estimates of affected area and population for known landing data
RegionClassical approachProposed approach
 Area (km2)PopulationArea (km2)Population
Ayeyarwady1,89640,3928,707376,788
Bago East4129131495
Chin State930
Kayin State433641929,064
Mon State462,15230559,670
Rakhine State6,481133,5553,22079,144
Tanintharyi6071,4943841,340
Yangon816,687893147,214
Total9,203184,93613,735673,716

2.4.2. GIS model based on spatially distributed surge profile

Because surge height during a cyclone varies as a result of cyclone wind velocity and topographic effects of the coastline, a GIS model using a spatially distributed surge profile would be more effective. The proposed GIS model uses a wind-based surge profile estimated from the cyclone advisories and assumes a discrete surge profile is valid along lines parallel to the cyclone's landing direction. The discretised surge profile created for the Nargis advisory 6 was used to make surge buffers for the actual landing point and direction, and thus estimate the affected area and population (Table VI).

2.4.3. Effect of time advisories on the results of the proposed GIS methodology

In order to find out the effects of different time advisories on the results of the proposed GIS model, sensitivity analyses were conducted for two scenarios: i) when landing point and direction are known; and ii) when landing point and direction are provided by different time advisories. The first scenario allowed for analysis of the effects of different surge profiles on the estimation of affected area and population (excluding the error due to the unknown location of the cyclone landing). The second scenario used landing point and direction estimates from different time advisories; surge profiles were calculated from these advisories using wind data. Table VII shows the affected area and populations calculated for each scenario, and Figure 8 illustrates the variation of the calculated affected area and population and their spatial overlap in relation to the distance to the landing point.

Figure 8.

Change in calculated area and population by advised landing point

Table VII. Estimates of affected area and population for different time advisories
Proposed GIS approach/known landing point
 Area (km2)Population
Region\Advisory123456123456
Ayeyarwady6,5906,3975,7268,6748,0008,707171,426170,328164,460318,770304,066376,788
Bago East222437292831380377350406407495
Chin State444433000000
Kayin State72727561581921,0781,0791,0903,8773,8779,064
Mon State6767712302213057,8827,8417,97032,53332,71459,670
Rakhine State3,6093,6623,8163,2213,1183,22076,77176,86277,66675,08375,60279,144
Tanintharyi4394434493893923841,0241,0018531,0541,3361,340
Yangon415417403905630893107,825107,57781,017144,045131,837147,214
Total11,21811,08810,58113,51312,45013,735366,388365,066333,406575,767549,839673,716
Proposed GIS approach/advised landing point
 Area (km2)Population
Region\Advisory123456123456
Ayeyarwady Divisio1,9651,9363,1414,7117,2439,10741,83041,80146,94988,127277,794417,652
Bago East Division4242423331403023023002843431,951
Chin State555443000000
Kayin State4544443543863773773753551,5215,046
Mon State393939351601762,0472,0692,0572,05331,74619,017
Rakhine State5,0134,9524,7483,7623,1973,03691,48391,96589,67884,26276,28076,542
Tanintharyi5205145134104033609049038988501,3491,293
Yangon8483831204392,2826,9256,9176,8769,006110,187418,436
Total7,7137,6168,6149,11111,51915,090143,869144,334147,132184,936499,221939,937

3. Results

3.1. Wind models for estimation of surge profile

The empirical equations and regressive models in Table III produced reasonable maximum wind velocity estimations reflected by the coefficient of determination and standard errors. For the Nargis advisory 5, maximum wind estimations based on the models in Table III are in the range of 40.5∼48.2 m/s which is lower than the actual observed value of 51 m/s but similar to the actual landing value of 44 m/s. Among the wind models in Table IV, models 1 and 2 do not fit well the advisory data at the cyclone centre and tail. Model 3 shows agreement with the wind advisories only in the middle and tail areas of the wind velocity distributions. Model 4 provides the best agreement with the wind observations from Nargis advisories (Figure 4); this agreement is also valid for other advisory directions (Figure 5). The accuracy and simple structure of model 4 make it preferable for surge profile estimations.

3.2. GIS models for estimation of affected area and population

Approximately 4150 km2 of land area and 15 000 people in Myanmar coast within the elevation range of 0–1 m are directly at risk due to sea level fluctuations, and increased further in areas and population above 3 m in elevation. Rakhine State, Ayeyarwady, Yangon, and Tanintharyi Divisions have the largest land areas that could be potentially affected even for small storm surges. The area and population under the potential surge risk in the Ayeyarwady and Yangon Divisions and Rakhine State increase remarkably for higher surge heights as compared to Tanintharyi Division. Yangon Division has a small land at risk from storm surges, but a high population density which increases rapidly at elevations above 3 m (Figure 7).

The simple GIS model based on constant surge height shows Rakhine State and Ayeyarwady Division has the largest affected area and population; Yangon Division has a small surge-affected area but a large affected population (Table VI). The proposed GIS model based on spatially distributed surge profile shows similar results; however, the affected area and population values were found higher for Ayeyarwady and Yangon regions exposed to the central part of the surge profile and lower for Rakhine State which was exposed to the tail of the surge profile (Table VI). The total affected area and population calculated using both simple and proposed GIS models (for the actual landing point) can be seen in Figure 9. Despite similarities between the results of both models, affected population estimates are not consistent across impacted areas. The spatial overlap between results from two models is around 50% for the impacted area and 20% for the affected population. Differences between the affected area and populations calculated by the two models are primarily due to: i) the spatially distributed surge profile, unlike the uniform surge height, is higher around the landing point and lower in the middle and tail parts of the cyclone; and ii) the maximum surge height in the simple GIS model was based on the central pressure of the cyclone and, therefore, only approximates the actual maximum surge height.

Figure 9.

Surge-affected regions calculated by the classical and proposed GIS models. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Analyses of the affected population and areas using the actual landing point and surge profiles obtained for different advisories shows that accuracies of the estimated affected area and population are 80 and 50%, respectively, for the advisory issued when the cyclone was 800 km from the landing point. These ratios increase almost linearly for advisories issued when the cyclone was within 600 km of the coastline. Spatial overlap of the results obtained for advisories issued when the cyclone was at the distance of 400 km from the coast and for the advisory closest to landfall (advisory 6) is 80% for the affected area and 90% for the affected population. Both ratios increased to more than 95% for advisories issued when the cyclone was 200 km from the coast (Figure 8). Note that accuracies were calculated using results of the proposed GIS model built for advisory 6.

As a result of the analysis using advised landing points and surge profiles calculated from different advisories, the following outcomes were reached: i) the affected area can be estimated with an approximate accuracy of 60% for advisories farther than 600 km from the coastline; ii) the affected population can be estimated with an accuracy of around 20% for advisories farther than 600 km from the coastline; and iii) the spatial overlapping ratios of the results obtained for an advisory as far as 600 km from the coastline, and for the advisory closest to landfall, are around 75 and 50%, respectively, for affected area and population. Accuracies of both the affected area and population estimates increase when advisories are close to the landfall area (Figure 8). Parallelism, with the maximum wind velocity increment variation by advisory position, was observed (Figure 3). In the proposed GIS model, the most reliable and accurate estimations resulted from advisories issued when the cyclone was within 200 km of the coastline. Note, that Nargis advisories could provide reliable landing-point estimates when the cyclone was within 200 km from the coast.

4. Discussion

GIS models are important tools for predicting damage from storm surges caused by cyclones. However, in the literature, there is no reliable model that can be applied universally before a cyclone makes landfall. Many available models use constant surge height estimation derived from the central pressure of the cyclone, though it is clear that surge height is not uniform and depends on a multitude of factors including central pressure, terrain heterogeneity, and cyclone characteristics that are subject to temporal and spatial changes. The GIS methodology proposed herein makes it possible to consider real-time cyclone advisories for early estimation of the surge profile, affected area, and population. Extents of the affected area and population are modelled taking into account the spatially distributed surge profile which is dependent mainly on the wind velocity and topographic irregularities along the coast. The proposed methodology can also be used to assess risk of mortality or degree of damage before cyclone landfall, making it a useful tool for emergency planners and responders.

4.1. Storm surge estimation

The study suggests that surge height can be modelled by linear and power-regressive models (i.e. wind-based surge models) using wind velocity. Segmented regression can also be used to improve the accuracy surge estimations for both low and high surge levels. However, for hazard assessment models only the highest surge levels throughout the coastline would be of concern; therefore, simple linear models are more efficient for surge modelling. Because wind and surge data are not available for many regions, and extreme wind and surge data generally are not recorded by gauging stations due to the violent conditions, wind-based surge models cannot generally be calibrated by gauged data. In such cases, hydrodynamic model simulations can be used to obtain regressive models for any location along the coastline. Use of different wind-based surge models for varying locations would make possible to consider local topographic effects on surge heights.

Wind models 1 and 2 require estimation of the maximum wind velocity during the cyclone. The maximum wind velocity during a cyclone can be modelled using central pressure considering that forward speed, maximum wind radius, and the Coriolis factor do not have a significant effect on maximum wind estimations. Use of the regional maximum wind radius, applied in wind model 1, resulted in larger wind velocities around the cyclone centre and lower wind velocities around the tail. Model 2 does not accurately reflect the wind velocity distribution in the tail of the cyclone, and model 3 does not accurately predict wind velocity distribution in the central part of the cyclone. However, model 2 is practical when advisories are absent, and model 3 can be used for the estimation of cyclone extent because it shows good agreement with advisory wind data (with the exception of maximum wind velocity). Model 4 is the most plausible and simplistic approach for defining wind velocity distribution using cyclone advisories. It should be preferred for surge profile estimation when advisories are available.

The asymmetric structure of cyclone Nargis was not captured by wind data provided by advisories (Figure 6), thus, the calculated surge profile does not fit the surge profiles extracted from the AVISO imagery for the northwest–southeast direction (Figure 6). This is especially true in the right-hand side of Nargis, however, the error is negligible within the discretisation range of the surge profile. The discretisation of the surge profile had an important effect on GIS model outputs since digital elevation data is generally provided with 1-m precision; ideally, discretisation should be done in a manner to better represent the calculated surge profiles. The surge profiles can be discretised at the locations corresponding to the integer surge height (values such as 1, 2, 3 m, etc.), or at some certain distance (such as 0, 100, 200 km, etc.) from the cyclone centre. These approaches generally resulted in higher or lower estimates for the affected area and population. This was especially true for the coastal plain, where the majority of the population lives, which flattens out a few meters above sea level. Discretisation of surge heights would be improved by using fractional values such as 0.5, 1.5, 2.5 m, etc.

4.2. Comparison between simple and proposed GIS models

The simple GIS model uses a constant surge height throughout the coastline, therefore, the extent of calculated damage depends only on terrain irregularities which results in potential for bias of affected area and population estimates in regions with wide coastal plains. This approach can be ideal for preliminary estimates of the affected area and population and can be used when there is no available advisory for informing disaster response planning. The proposed GIS approach uses advisory data on landing point and direction of the cyclone landfall as well as wind-based surge profile, provided that a reasonable advisory is available for the landing-point location, the affected area and population can be estimated in a more comprehensive way. Calculations for the affected population are sensitive to the accuracy of surge profile estimates and landing-point location since settlements and associated population densities are distributed unevenly along the coast.

Though calculated total affected areas are similar for both GIS models, the calculated affected populations differed greatly. The simple GIS model based on constant surge height produced higher affected population estimates for Rakhine State and Tanintharyi Division, and lower estimates for Aveyarwardy and Yongon Divisions. The simple GIS model does not provide precise estimations for the regions around the cyclone centre and tail; however, accuracy could be improved by using more specific maximum surge height estimations. Cyclone Nargis reports indicate that only Ayeyarwady and Yongon Divisions were exposed to significant damage (TCG, 2008). Both GIS modelling approaches found significant affected areas and population values for Rakhine State. High estimates for affected land area and population in Rakhine State from GIS models can be partially attributed to the numerous islands which increased estimates of the total affected area (which may have been excluded from government estimates that employed standard administrative boundaries which do not always include islands).

4.3. Effectiveness of the proposed GIS model for early risk assessment

GIS models based on the actual landing point made possible to assess the effects of surge profiles from advisories on estimates of affected area and population. Affected population estimates are more sensitive to accurate estimation of surge profile than to the estimates of affected area (Figure 8 and Table VII). The accuracies of the estimated affected area and population for the advisory issued when the cyclone was at a distance of 600 km (approximately 50 h before landfall) from the coastline were 80 and 50%, respectively (these ratio values are given for calculations relying on Nargis's actual landing point, direction and advisory wind data). For the advisories issued when the cyclone was closer than 600 km, more accurate surge profile estimations and precise estimations of affected area and population were obtained.

Landing-point location is an important element in the proposed GIS model. The dynamic behaviour of cyclones results in different landing-point locations and wind data at different time advisories. This yields varying surge profile estimations for different time advisories. In the case of cyclone Nargis, advisories issued when the cyclone was within 200 km (approximately 20 h before the landfall) of the coastline provided the most accurate location of the landing point. Models based on earlier advisories were, thus, less accurate. However, they might be more useful for emergency planners. The affected area and population were estimated with accuracies of 60 and 20%, respectively, for the advisory issued at 600 km from the coastline These ratios increased almost linearly for the advisories closer than 400 km and reached up to 100% for the landfall position (these ratios are the values given for calculations relying on the advised landing point, landing direction, and wind data).

The information on magnitude of the affected area and population can be used for disaster response planning such as, estimation of the scale of the required response and pre-positioning of emergency supply stocks. However, for allocation of emergency forces and supply stocks, information on geographical location of the affected area and population is more important than magnitude. For the analyses that use Nargis's actual landing point and direction data, the overlapping ratios of the calculated and observed affected area and population differed by 70 and 50%, respectively, for the advisory issued when cyclone Nargis was more than 400 km from the coastline. After this advisory, they increase reasonably (shown in the graph of overlapping ratio on the left side of Figure 8). For the analyses that use advisory landing point and direction data, 75% of the affected area and 50% of the affected population estimates were geographically accurate for the advisory issued when cyclone Nargis was 400 km from the coastline. These ratios reach 90% for the advisories issued when the cyclone was within 200 km of landfall (shown in the graph of overlapping ratio on the right side of Figure 8). Unfortunately, there is no clear break point that can serve as a threshold for determining which advisories to employ as the basis for GIS models. The possible relationship between the spatially varying maximum wind velocities and the affected area estimations (Figures 3 and 8) could potentially be used to define a threshold advisory position, though the relationship between maximum wind velocities and different time advisories would have to be characterised by reviewing available historical cyclone data from the region.

5. Conclusions

Surge heights can be modelled by linear and power-regression models using wind velocity, though more complicated numeric models would likely increase the accuracies of the estimations. A more complicated process makes simple linear regression models preferable for high wind velocities (>10–15 m/s). Estimates of affected area and population are highly sensitive to the surge profile and landing-point location. Simple GIS models that use constant surge height can produce preliminary estimations for the affected region. However, affected land area and population can be overestimated, especially in low-lying coastal plains where surge values are often inaccurate. The proposed GIS methodology which uses a spatially distributed surge profile eliminates the bias associated with simple GIS models and improves the accuracy of estimations.

In the case of cyclone Nargis, Rakhine State and Ayeyarwady Division had the largest land areas at risk from storm surge. The largest populations at risk due to storm surge were found in Rakhine State, Ayeyarwady, and Yongon Divisions. The affected areas and population were estimated with accuracies 60 and 20%, respectively, for the advisories issued at 50 h prior to landfall (when > 600 km from the coastline), and these accuracies increased to 80 and 60%, respectively, for the advisory issued at 20 h before landfall (when the cyclone centre was within 200 km of landfall). The relationship between the distance-based variation of maximum wind velocities and the distance-based variation of the affected area estimations can provide a basis for further development of GIS modelling methodologies.

Ancillary