DEVELOPMENT OF A SET OF TRIP GENERATION MODELS FOR TRAVEL DEMAND ESTIMATION IN THE COLOMBO METROPOLITAN REGION

DEVELOPMENT OF A SET OF TRIP GENERATION MODELS FOR TRAVEL DEMAND ESTIMATION IN THE COLOMBO METROPOLITAN REGION Ravindra Wijesundera and Amal S. Kumarage Dept. of Civil Engineering, University of Moratuwa Paper Presented at the Engineering Research Unit Symposium 2001 University of Moratuwa ABSTRACT Estimation of demand for travel is very useful in planning transport infrastructure in keeping with anticipated human settlement patterns and activities. The amount and nature of travel depend on the population size, income level and type of employment etc. Therefore, by correlating the trip generations with socio-economic parameters, it is possible to develop mathematical models to predict travel demand. The objective of this study is to develop a family of trip generation models for the Western Province in Sri Lanka, to estimate travel demand for work, education and other purposes by available motorized forms of transport. While the main focus is on estimation of bus passenger demand, another set of models is calibrated estimate aggregate demand for bus, rail, car, motor cycle and three - wheeler travel. Finally, a mode choice model is developed to estimate the variation of bus passenger modal share in terms of availability of rail and private vehicles. The calibrated traffic generation models can be used for estimating future trip generations in Western Province at Divisional Secretariat Division level. In addition, it is possible to use these models for trip generation estimates for other geographic regions in Sri Lanka after validating the models for the new study region. INTRODUCTION Planning of transport infrastructure often requires quantitative measurements of demand for transport. Typical demand data required by transport planners include total passenger and freight trips generated from different geographic areas, their mode of travel such as bus, rail and car, the route of travel and the destination. In addition, accurate forecast of future travel demand is also an essential input in planning transport facilities. As such, the need for a scientific approach to estimate the demand for transport has been identified. Since the demand for transport has a strong correlation with the socio-economic parameters such as population, type of employment, income level etc, it is possible to estimate the transport demand in terms of the socio-economic variables. Mathematical modelling is one such technique that has been used extensively for transport demand estimation. This study focuses on calibrating a family of trip generation models that identifies the influential parameters of trip generation for motorised travel. The selected models predict the trip generation for each Divisional Secretariat Division (DSD) in the Western

Province. In addition, the trip generations would be predicted by the purpose of trip. Three types of trips were chosen for this study. Theses are; 1. home based work trips - A passenger trip that has its origin at home and its destination at his/her place of work 2. home based educational trips - A passenger trip that has its origin at home and its destination at school/university or other academic institution 3. home based other trips - A passenger trip that has its origin at home and its destination at a place other than the ones mentioned above Trip generation models that estimate trips by their purpose of origin are very useful for transport planning in a predominantly urban area such as the Western Province. METHODOLOGY In the first phase of study, a family of models was calibrated to predict the trip generations for bus travel for the above-identified trip purposes. The second phase of the research was the calibration of a family of trip generation models for total passenger travel demand. It considered travel by bus, rail, car, motorcycle and 3-wheeler modes. The third and final phase of research was the calibration of a modal share model for bus passenger home based work trip generation. It estimates the bus passenger travel as a percentage of total passenger trip generation depending on the accessibility to alternative passenger transport models such as rail and private modes. Table 1 summarises the types of models calibrated in this research. Table 1: Summary of Demand Models Calibrated Type of Models Travel Mode(s) Home Based Work Type of Trip Home Based Educational Home Based Other Trip Generation Bus 3 3 3 Trip Generation Bus, Rail, Private Vehicle 3 3 3 Modal Share Bus 3 5 5 There are 34 DSD s in the Western Province. The reason for choosing the DSD as the forecast unit is that it is the smallest administrative unit for which reliable socioeconomic data is available. There are basically two types of data required for model

calibration. The first type is the travel data, which was obtained from the household interview surveys carried out in 1995 for the Colombo Urban Transport Study (CUTS-1). The second type is the socio-economic data. These were obtained from Central Bank publications and Transport Database developed by the University of Moratuwa. The socio-economic data such as population, employment distribution and vehicle ownership were not available for a single year. As such socio-economic data was converted to a single year (base year) by applying appropriate growth rates. The year 1998 was chosen as the base year. One of the problems faced during model calibration was the non-availability of trip data for all DSD s. As a result, the models were calibrated for the DSD s for which data was available and subsequently, these models were validated using the socio-economic data of the remaining DSD s. The regression technique was used for obtaining the best form of the predictor models. Accuracy of the models and independent variables were tested with significance statistics such as F statistic, adjusted R squared value and t statistic. The residual analysis points to the presence of any influential data points. Regression analysis was carried out on a personal computer using SPSS (SPSS Inc. 1999) software. CALIBRATION OF BUS PASSENGER TRIP GENERATION MODELS Selection of the Base Variable The first step of the model calibration was the selection of a base variable. Since a family of models was calibrated, it was desirable to have a single base variable appearing in all models. This makes cross comparison of the models easy. Generally the base variable is the single most dominant independent variable in the model. There were two variables from which the base variable was chosen. These were population and total households. As these two variables are highly correlated, only one variable can appear in a model. Since home based work trips are the single most significant type of trips, it was decided to take the base variable of that trip category for other trip purpose models as well. In order to select the most appropriate base variable, two home based work trip generation models for bus travel were calibrated using one base variable at a time. The regression analysis indicated that the household variable is marginally better than the population variable. As such, household variable was chosen as the base variable for the family of models. Home Based Work Trip Model The final model for home based bus work trip generation in its log-transformed form is given as equation (M1). HBWT = 0.32 { }..( M1) 1.161 2 ( HHTOT ) exp 1.58( PVHH ) 3.14( STDEN ) 6.06 10 ( EDUPC)

Where HBWT = one-way home based bus work trips per DSD per day HHTOT = total households in DSD PVHH = Private vehicle ownership per household in DSD (= registered vehicles/ number of households) STDEN= Rail station density of the DSD (= number of stations in DSD/ area of DSD in square kilometers.) EDUPC = Percentage of school and university attending population of the entire population over five years of age of a DSD There exists a positive relationship between the number of trips and households. According to the model, if the number of households in one DSD increases by 10%, and if everything else remains unchanged, then the corresponding increase in home based work trip generation would be approximately 12%. Similarly, various growth scenarios can be analyzed to determine the proportionate variations in trip generation. The private vehicle ownership variable fits into the model in positive exponential form. As the private vehicle ownership increases, the home based work trips by bus also increase. This is seemingly counterintuitive. But the reason for this is the vehicle ownership variable here is present as an income variable rather than modal share variable. When the vehicle ownership increases by 0.10 vehicles per household, the increase in home based work trip generation is 17%. Rail station density shows a negative exponential relationship with home based work trip generation by bus. If a new rail station is added to each of the DSD, then the observed decrease in bus work trip generation percentage varies between ( 1.6%) for Horana DSD and ( 16.8%) for Maharagama DSD. Home Based Educational Trip Model The log-transformed version of the home based educational trip generation model for bus is given below. HBEDT = 8 10 3 1.396 ( HHTOT ) exp{ 3.68( STDEN )}...( M 2) Where HBEDT = one-way home based educational trips per DSD per day

Here too, the relationship between total number of households and the total number of home based educational trips generated is positive. For example, if the number of households in one DSD increases by 10%, and if everything else remains unchanged, then the increase in home based educational trip generation would be approximately 14%. As expected, the effect of rail station density is negative. If a new rail station is added to each of the DSD, then the observed decrease in bus educational trip generation percentage varies between ( 1.9%) for Horana DSD and ( 19.5%) for Maharagama DSD. One of the most noticeable observations regarding the home based educational trip model is the absence of private vehicle ownership variable. This shows that income level is still not a significant factor that affects bus trips for educational purposes. Home Based Other Trip Model The best possible model that predicts home based other trip generation for bus travel in its log-transformed form is given below. Where HBOT = 6.13 10 1 0.947 2 ( HHTOT ) exp 4.85 10 ( UNEMPC) HBOT = one-way home based other trips generated per DSD by bus per DSD per day UNEMPC = percentage of unemployed population. { }...( M 3) Like in previous models, a sensitivity analysis is performed to evaluate the magnitude of the effect of change of household variable on other trips. If the number of households in one DSD increases by 10%, and if everything else remains unchanged, then the resulting increase in home based other trip generation would be approximately 9%. The absence of private vehicle ownership and rail mode variables in home based other trip model for bus is conspicuous. In general, as the income level increases, the social and recreational trips also have observed to be increasing. But here, income does not appear to be significantly affecting the "other" bus trips. Furthermore, it appears that the rail cannot compete with buses for this trip category unlike for work and educational trips. It should be noted that rail basically serves few town centres and most of the working places and schools are located there. But social and recreational trips are widely distributed in their destination and as such rail is not capable enough to serve such distributed demand.

Finally, it can be seen that accuracy and predictability of a model is dependent on the homogeneity of the trip purpose. While "work" and "educational" trips are largely predictable and can be clearly distinguished, "other" trips include a set of contrasting trips (for example, the causes leading to visiting market and visiting a relatives place are not similar). Therefore, it is difficult to identify the exact determinants of such trips unlike in "work" and "educational" trips. As such, home based other trip model does not contain strong secondary variables. TOTAL PASSENGER TRIP GENERATION MODELS Home Based Work Trip Model The following model was obtained for total home based work trip generations by bus, rail and private modes. HBWTT = 23.85 { }...( M 4) 0.697 2 ( HHTOT ) exp 2.03( PVHH ) 3.7 10 ( EDUPC) Where HBWTT is the total one-way home based work trips generated per DSD per day It can be seen that the significant variables in this model are the same as the variables in the home based bus work trip generation model. The only exception is the absence of rail station density variable. Being a modal share variable, station density cannot be considered here since rail trips are also included in the model. It is very important to emphasis that vehicle ownership variable is present in the model purely as an income variable and not as a modal share variable. When the sensitivity of household variable is analysed, a 10% increase in households would contribute to about 7% increase in total work trips. Similarly, when the household car ownership increases by 0.1 or one addition of car to 10 households, the resulting increase in total work trips is approximately 23%. Home Based Educational Trip Model The following regression model was obtained for home based educational trip generations. HBEDTT = 1.647 0.900 ( HHTOT )...( M 5) Where HBEDTT is the total daily one-way home based educational trips generated per DSD It can be seen that except for the base variable, there are no other significant secondary variables that explain the educational trip generations. When the effect of household

variable is considered, we can see that a 10% increase in the number of households in a DSD would result in approximately 9% increase in total educational trips. Home Based Other Trip Model The following model was obtained for total home based other trips. 2 HBOTT = 7.24 10 1.221 ( HHTOT )...( M 6) Where HBOTT is the total one-way home based other trips generated per DSD per day When home based bus models and total trip models are compared, one of the most conspicuous observations is the absence of secondary variables in the total demand models. The only exception is the home based work trip model. When total educational and "other" trip models are examined these remain as single variable models. This has resulted in models having a very strong constant. It can be seen that 10% increase in households would result in approximately 12% increase in total other trips. CALIBRATION OF MODAL SHARE MODEL Calibration of modal share models for motorised travel was limited to home based work trip generations only since other trip types did not show a clear relationship with the mode choice variables. This was evident in the trip generation models explained in previous sections. The modal share model for work trips determines the percentage of work trips by bus considering the availability of alternative modes, namely rail and private vehicles except three wheelers. The calibrated model is given below. Where PCBWT is the percentage of home based bus trips for work per DSD The negative coefficients indicate that station density and private vehicle ownership variables have a negative impact on bus modal share. From the above model, it can be seen that when both private vehicle ownership and station density are zero, the percentage of work trips by bus is 96.5%. Therefore, we can conclude that the remaining 3.5% of the work trips are made by three wheelers. When the effect of rail station density is analysed, consider a DSD of extent 50 km 2 and having 5 railway stations. Provision of one more station (i.e. increase of station density from 0.1 per km 2 to 0.12 per km 2 ) would result in approximately 1% reduction in bus modal share. Similarly, to analyse the effect of private vehicle ownership on mode share, consider a DSD having 1000 households and 100 cars. Increase of vehicle ownership to 200 vehicles (i.e. increase of vehicle ownership from 0.1 per household to 0.2 per household) would bring about 6.5% reduction in bus travel. VALIDATION OF MODELS { 0.43( STDEN ) 0.71( PVHH )}...( 7) PCBWT = 96.54 exp M

Validation of models is an important step in model calibration as it gives an idea about the accuracy and reliability of the calibrated models. In order to validate the models, comparison of observed and predicted data was performed. This section describes the validation of home based work trip models and mode share model. Table 2 gives the observed and predicted one-way home based work trips for selected DSD s. Table 2: One-way Home Based Work Trips for Selected DSD s Home Based Work Trip Generations Bus Trips Bus Modal Share DSD Observed Predicted - Model M1 Calculated - Model (M4* M7/100) Observed Calculated - Models (M1/M4) Predicted - Model M7 Colombo 54,533 57,455 59,999 63.2 56.7 59.2 Homagama 19,532 16,606 18,583 75.4 69.3 77.5 Kaduwela 31,002 28,474 24,143 82.5 88.8 75.3 Kesbewa 28,901 25,971 22,317 75.8 77.8 78.6 Nugegoda 16,858 17,272 19,371 70.5 64.8 72.7 Dehiwela 29,128 31,433 31,284 69.0 73.1 72.8 Gampaha 26,825 25,093 24,589 85.0 72.6 71.1 Wattala 17,011 19,111 19,144 65.6 76.8 77.0 Biyagama 14,836 16,144 16,893 88.4 74.1 77.5 Minuwangoda 16,481 18,966 19,084 68.5 73.3 73.7 Mahara 25,647 21,128 20,316 76.8 79.1 76.1 Horana 11,171 13,186 14,408 75.7 76.1 83.2 Bandaragama 14,008 15,285 15,919 79.9 78.2 81.4 Kalutara 11,098 10,160 12,888 83.7 65.9 83.5 Total 317,032 316,374 318,938 69.9 71.3 72.9 In general, it can be seen that predicted values do not significantly vary from the observed values. When the variations of the individual data points are considered, the predictions by direct bus model is found to be more accurate than the two step work trip prediction method which make use of modal share model and total work trip model. This

is understandable as when the number of steps increases, related inaccuracies due to mathematical assumptions also accumulate. SUMMARY OF MODELS Table 3 given below summarises the models calibrated under this research. The statistical parameters of the models as well as individual variables are also shown where appropriate. Table 3: Summary of Demand Models Type of Model Trip Purpose Adjusted R 2 Coefficients & (t-statistics) Constant HHTOT PVHH STDEN EDUPC UNEMPC Bus Total Modal share HBWT 0.93 HBEDT 0.74 HBOT 0.66 HBWTT 0.78 HBEDTT 0.61 HBOTT 0.77 HBWT% 0.49 0.32 (-1.04) 0.008 (-2.55) 0.61 (-0.29) 23.86 (1.88) 1.65 (0.30) 0.07 (-1.69) 96.54 (62.06) 1.16 (10.95) 1.4 (7.58) 0.95 (5.81) 0.67 (4.28) 0.90 (5.65) 1.22 (8.20) - 1.58 (3.84) - -3.13 (-8.69) -3.677 (-5.49) -0.06 (-6.70) - - - 2.03 (2.77) - - - - -0.04 (-2.77) 0.05 (1.81) - - - - - - - - -0.71 (-2.18) -0.43 (-2.83) - - - CONCLUSIONS The main conclusions drawn by this study are: When the different types of calibrated models are considered, the home based work trip generation models have the highest number of predictor variables. It was able to establish clear relationships between work trips and income which is explained by the household vehicle ownership, type of employment and influence of rail mode on work trips by bus. The main explanatory variables in the home based educational trip models are the number of households and rail station density. A significant observation is the absence of income (vehicle ownership) variable both in bus and total demand models.

Home based other trip generation models do not show a strong relationship with income (vehicle ownership) or type of employment. One of the possible reasons for this is that other trips include social, business, recreational and non-specified trip types. As a result of this aggregation of a number of trip types, it is difficult to identify the influence of income or employment on the entire trip classes as a whole. The modal share model for home based work trips by bus identifies clear relationships on the influence of rail and private vehicle. Negative coefficients indicate that home based work trips by bus reduce when the availability of rail or private vehicles increases. The testing and validation show that the model predictions are close to the observed values. As such, the reliability of the models is satisfactory. Comparison of predicted trip generations for bus mode by direct and indirect models discloses that in general, the direct models gives predictions closer to the observed trip generations. Trip generation models calibrated by stratifying mode of travel and purpose of travel provide better insight into the intrinsic characteristics of the trip generation rates of sub-models. It is possible to use these models for other geographic regions after validation. This is very useful since it avoids repetition of model calibration. REFERENCES - Colombo Metropolitan Regional Structure Plan Vol.2, (1998), Urban Development Authority, Sri Lanka. - Division of Transportation Engineering, (1997), Greater Colombo Traffic Model, University of Moratuwa, Sri Lanka. - Halcrow Fox, (1996), Colombo Urban Transport Study - Stage 1, London, UK. - Kanafani, A. (1983), Transportation Demand Analysis, McGraw Hill Series in Transportation, New York, USA. - Kumarage, A.S. (1990), Intercity Travel Demand Modelling, University of Calgary, Alberta, Canada. - SPSS Inc., (1999), Statistical Package for Social Scientists: Release 10.0, U.S.A. - Wijesundera, R. (2001), Development and Testing of a Set of Mathematical Models for Travel Demand Estimation, Unpublished M.Eng Thesis, University of Moratuwa, Sri Lanka.