The Influence of Spatial Factors on the Commuting Trip Distribution in the Netherlands

Abstract

Traffic flows are the result of movements of people and goods. They are modeled with the help of behavioral patterns that are supposed to remain relatively constant over time. In traditional transport modeling, some of these patterns are described by trip distribution functions, which represent the propensity to make trips with certain costs. The distribution functions (DF) are used to estimate a priori origin destination (OD) matrices.

Appendices

Appendix 1. Removal of False Reports

Most false reports are detected and corrected for within the MON acquisition process. Sometimes, however, errors remain unnoticed; for example, when a respondent accidentally fills in the wrong postal zone for the origin or destination. We checked for these false reports by comparing the reported travel distances d _r (in kilometers) with the Euclidean distance d _f between the centroids of the origin and destination zone. In Fig. 6.4 we show the results. Note that less than 4% of the distances were imputed in the MON acquisition process, so the distances reflect the answers of the respondents.

Fig. 6.4

Reported and Euclidean distances (in km)

Figure 6.4 shows the correlation between d _r and d _f . The average relation is shown by the solid line. We suggest that trips for which d _r < (d _f – 5) km or d _r > 2 × (d _f + 5) km (shown by the dashed lines) are unrealistic and can be regarded as false. The fraction of false reports is only 6% of the sample, and this fraction is rather constant with distance. The false reports are also not concentrated in specific areas. The identified cases were subsequently left out of the sample. Note that the figure includes internal trips, for which the origin and destination are in the same zone. The estimate for the free internal distance is described in Appendix 3.

Appendix 2. Network Versus Reported Distances

Here, we address the question whether to use reported or network distances. Network distances can be estimated from macroscopic models that assign trips to the network. The network distance d _n (between two postal zones) is estimated as the distance along the fastest route in free flow conditions. Since only commuting trips are considered, it is expected that many trips are affected by congestion. We claim, however, that the distances in free flow and congested conditions will often not be very different. To validate this claim, we compared d _n, provided by a transport model, with d _r. In Fig. 6.5 (upper panel), we have plotted the average values of the aggregates within free distance bins. We used the bins, as described in Sect. 6.3. For d _r and d _n, the averages are plotted for car trips (filled symbols) and all trips (open symbols). We excluded internal trips.

Fig. 6.5

Reported and network distances for car (filled symbols) and all trips (open symbols). The lower panel shows the logarithmic difference between the two distances

According to Fig. 6.5, both distances are in general more or less equal for large distances (the symbols lie around the dashed lines). This is also illustrated by the lower panel of the figure, in which the logarithmic difference between both distances is shown. The lower panel shows that the network distance is on average slightly larger than the reported distance for distances between 2 and 10 km, although this is mainly so for non-car trips. This can be explained by the fact that d _n is calculated for car trips, while cyclists and pedestrians can take short cuts. Note that the network distances are also slightly larger for car trips, because cars can use streets that are not included in the transport model.

Although there are some small differences for d _n > 2 km, network and reported distances are in general quite comparable. For very small distances, however, d _n becomes significantly smaller than d _r. We argue that this difference is caused by an overestimation in d _r because respondents tend to “round off” their reported distances. More generally, “accidental” errors in reported distances (due to flawed estimates from respondents) may contribute to inaccuracies.

Appendix 3. Internal Distances

To obtain internal network distances, we first estimated the relation between network and free distance. This relation is shown in Fig. 6.6, in which we have plotted the averages of aggregates (filled symbols) within free distance bins. Note that we used different bins here, since in this case we were able to extend our bins to larger distances (because we were not restricted to the survey sample size). According to the figure, the relation between d _n and d _f can be described by d _n = 2.04 × d _f ^0.90 (solid line). Note that this relation implies that the “detour” ratio d _n/d _f becomes significantly larger than 1 for very short distances. Such a trend is also reported in (Chalasani et al. 2004), albeit weaker for a high-resolution network. The network of the transport model, which was applied for the whole of the Netherlands, is not very detailed. As a result, travel distances may still be slightly over estimated for the very short distances.

Fig. 6.6

Network and Euclidean distances (in km)

With the relation between d _n and d _f , it becomes possible to estimate average internal network distances. For this, we need an estimate of the internal Euclidean distance. If the pools of production and attraction are distributed homogeneous (but not necessarily uniformly) throughout a postal zone, the average Euclidean distance for internal trips is equal to 0.5√(A/π) with A being the area of the zone. We adopted this estimate as the Euclidean distance for internal trips.