An Urban Land Use Model Using Muliple Periods Data

Abstract

There have been many empirical urban land use models made up to the present. But most of them used only one year or one period data for their model calibration. Those models were not examined in the light of time series analysis. Therefore, planners have been afraid that they may misproject in long term projection while they can get good results in a short term one. Model builders have been interested in time space land use models. But they can not have built such a model. One main reason is the lack of time series cross section data of urban activities distribution. This study shows a time space land use model. This model allocates household and employment at 18 zones in the Matsuyama metropolitan area (population 530 thousand in 1980). The model was examined with the data from 5 periods during 25 years from 1955 to 1980.
This model looks like the Lowry model. It is composed of the basic sector (manufacture, public utilities and govenment), the household sector, the nonbasic sector (construction, wholesale and retail, bank insurance and real estate, and services) and the agriculture sector. The number of those employed in industries in the nonbasic secrtor and agriculture and the number of households are estimated by linear regression equations. But the number employed in industries in the basic sector are given exogeneously.
The model calibration was carried out in two ways. First it was done by period. We got five sets of parameter values estimated for each regression equation. Then pooling data calibration was done. The results of these calibrations were good. The value of the multiple correlation coefficient is more than 0.97 in every regression case except the one of construction employment of which the least value is 0.93. The signs of most parameters estimated meet our expectation. T values of the parameter in the case of pooling data have less than 0.1% significance level except for 2 of all 14 parameters.
Trend analysis of 5 parameter values, estimated by period, showed no meaningful trend. With this result, it is difficult to estimate the future value of a parameter which is expected to be useful in long term projection from its trend analysis.
Two cases of long term projection which start in 1955 and end in 1980 with each parameter set were executed. One case used the parameter values determined from pooling data regression, and the other case used those derived from the first period regression (1955-1960). The number employed by industry and households were projected by zone in 1980. As for the number of households, the average error rate per zone was 32% for the pooling parameter case and 38% for the first period parameter case. We can not say whether this result means bad projection or not in a practical sense because there has been no other examination of long term projection until now. However, this result shows that the large value of multiple correlation coefficient which is found by regression analysis will not always guarantee the success of long term projection.