9th International Conference on Urban Drainage Modelling Belgrade 2012 Evolution of Urban Drainage Networks in DAnCE4Water Christian Urich1 , Robert Sitzenfrei2 , Manfred Kleidorfer3 , Peter M. Bach4, David. T. McCarthy5, Ana Deletic6 , Wolfang Rauch7 1 University of Innsbruck, Austria, 6020 Innsbruck, Technikerstr. 13, christian.urich@gmail.com University of Innsbruck, Austria, 6020 Innsbruck, Technikerstr. 13, robert.sitzenfrei@uibk.ac.at 3 University of Innsbruck, Austria, 6020 Innsbruck, Technikerstr. 13, manfred.kleidorfer@uibk.ac.at 4 Centre for Water Sensitive Cities, Civil Engineering Department, Monash University, Clayton VIC 3800, Australia, peter.bach@monash.edu 5 Centre for Water Sensitive Cities, Civil Engineering Department, Monash University, Clayton VIC 3800, Australia, david.mccarthy@monash.edu 6 Centre for Water Sensitive Cities, Civil Engineering Department, Monash University, Clayton VIC 3800, Australia, ana.deletic@monash.edu 7 University of Innsbruck, Austria, 6020 Innsbruck, Technikerstr. 13, wolfgang.rauch@uibk.ac.at 2 ABSTRACT To identify possible transition strategies for urban water DAnCE4Water (Dynamic Adaptation for eNabling City Evolution for Water) models the complex coherences between societal system, urban environment and the urban water system under different scenarios like climate changes, changes in the societal needs and urban changes. DAnCE4Water uses three modules to describe the urban system. In this paper the authors focus on the linkage between the urban development module and the bio-physical module. By coupling the urban environment with the urban water infrastructure it is possible to investigate adaptation strategies under consideration of a dynamically evolving urban water system. In this paper the authors develop a model to dynamically expand the urban drainage networks, by mimicking a real world planning process. The planning process is split up into two stages; first the bio-physical module layouts the main trunks; second new populated areas are connected via a secondary pipe network to the main trunks. The approach has been tested on Scotchman's Creek a urban catchment within Melbourne, Australia. The urban dynamic has been reconstructed based on the construction age of the drainage pipe. These results have been used as input for the bio-physical module. The authors could successfully evolve the urban drainage network by using the two stage planning process. The presented method is the basis for further applications of DAnCE4Water for predicting future development of urban areas. KEYWORDS agent based modelling, dynamic networks, DAnCE4Water, urban development 1 1 INTRODUCTION Within the next 30 years Melbourne expects an increase in population from currently 4 million people up to 8 million (Australian Bureau of Statistics., 2008). To avoid increased urban sprawl the city plans to introduce multiple city centres with higher population densities. Thus, new growth corridors are needed to inhabit twice the population. This is just one example of how fast our urban environments can change. All these changes are directly linked to the urban water infrastructure. Hundreds of kilometres of drainage networks have to be built in new urban growth corridors and also the existing infrastructure has to be adapted to protected people and their property from flooding. An increased urban area (i.e. pavement of areas) also means that more pollution from diffuse sources is transported into creeks and rivers. But not only urban development is challenging the urban drainage systems also climate change with its expected increase in rainfall intensities is putting pressure on how we design, build and manage our urban water systems. Figure 1. DAnCE4Water module overview (Rauch et al., 2012) Several international projects like CORFU (Djordjević et al., 2011), TRUST (Stedman, 2012) or SWITCH (Howe and Steen 2008) are aiming to identify possible transition strategies for urban water systems. Within the PREPARED project the University of Innsbruck, Monash University and the Centre for Water Sensitive Cities in collaboration with Melbourne Water develop the strategic planning tool DAnCE4Water (Dynamic Adaptation for eNabling City Evolution for Water) (Rauch et al., 2012). DAnCE4Water models the complex coherences between societal system, urban environment and urban water system under different scenarios like climate changes, changes in the societal needs and urban changes. DAnCE4Water uses three modules to describe the urban system (see figure 1): The societal transition module (STM) (de Haan et al., 2011), the urban development module (UDM) and the biophysical module (BPM). In this paper, the authors focus on the linkage between the UDM and the BPM to describe the evolution and adaptation of urban drainage networks. By coupling the urban environment with the urban water infrastructure it is possible to investigate adaptation strategies under consideration of a dynamically evolving urban water system. 2 To do this, in this paper the authors develop a model to dynamically expand the urban drainage networks over time, by mimicking a real world planning process. The planning process is split into two stages; first the UDM layouts the main trunks; second new populated areas are connected via a secondary pipe network to the main trunks. To expand the drainage networks we enhanced the agent based modelling approach developed by Urich, et al. (2010). To validate the approach we analyse the evolution of the drainage network of Scotchman’s Creek (Melbourne, Australia). Therefore we reconstruct the urban development and the development of the urban drainage system. The reconstructed dynamic urban environment is used as input for the BPM. The outputs of the BPM are compared with the existing drainage network. 2 METHODOLOGY 2.1 Dynamics in DAnCE4Water In DAnCE4Water the urban development is a key driver for the expansion and adaptation of the urban drainage networks. Based on a master plan for the future city layout and demographic projections the UDM projects the spatial evolution of the urban environment; new areas get populated, already populated areas are consolidated and people are moving from less to more attractive places. This is modelled in an annual time step. Based on the changed urban environment the BPM has to adapt the urban drainage network in each modelling time step. The BPM is therefore split up into two parts, one part is taking care of the centralised network (focus of this paper) and the other part of the decentralised water infrastructure (shown by Bach et al., 2011). Both parts interact with each other, if for example infiltration systems are implemented within a catchment the connected impervious area is reduced. The performance of the systems is assessed using external simulation tools like the hydrodynamic model SWMM (Rossman, 2008). The BPM evolves the water infrastructure by mimicking real planning processes. For the urban drainage network the BPM distinguishes between two different stages of the planning process (1) creating a master plan for the layout of the main trunks; (2) connecting new built up areas to the main trunks by designing a secondary drainage network. For both processes the authors use an agent based modelling approach introduced in the next paragraph. 2.1.1 Catchment Master Plan For the planning process a master plan of the main water infrastructure is required. Similar as a master plan for the urban development the master plan for the urban water system has to include the main water infrastructure facilities e.g. waste water treatment plants or major wetlands (location of such facilities is usually a political decision and cannot be predicted by the UDM). To connect the main pipes of the new developed catchment to the existing network or to the environment (e.g. water ways) also connection nodes or outlets must be defined (see case study description). To generate a layout of the network the catchment is discretised with raster cells (blocks) with a certain resolution depending on the level of detail that is required. Here we chose a resolution of 500x500m. By using the agent-based model the BPM automatically connects the blocks to the main infrastructure given by the master plan. The layout of the main pipes is mainly driven by the elevation and natural water ways. To estimate the impervious fraction the zonal information from the master plan of the urban area is used. Finally, the pipe diameters are designed using the time area method. 3 2.1.2 Connecting new Built up Areas to the Main Trunks In this paper the authors use a raster grid with a resolution of 100x100m to describe the urban environment. If the UDM populates new cells, the BPM connects them to the existing sewer network. In the design process of the layout for the secondary drainage network the influence of the street network is much more important as for designing the main trunks (they are mainly influenced by the elevation map). For the design process, the current land use is considered. 2.2 Agent-Based Model for the Generation of Drainage Systems Urich et al., (2010) developed a model for the generation of virtual combined sewer systems. To generate the layout of the sewer network, the model placed agents in the centre of the catchments without connection to the sewer network. The aim of the agents is to find a path that connects the catchment to an existing main trunk or a wastewater treatment plant and to mark their successful path for subsequent generations of agents. Agents operate on cell grids and can sense their local neighbourhood (Batty, 2005). Based on simple rules – they prefer to move to cells that are lower than their current position and they are attracted by the path of previous successful agents – the agent evaluates the attractiveness of each neighbouring cell. The attractiveness is an indicator for the probability in which cell the agent moves at the next time step – “probability field of the neighbourhood” (see figure 2). If the agents are successful they mark their path with an attraction field. This stochastic process –if the agent is equally attracted to all neighbouring cell the movement of the agent can be compared to a random walk – is repeatedly executed. As input the attraction field of the previous generation is used. As the attraction field gets more intense, more and more agents successfully reach their goal. After 20 generation, the attraction field contains a possible layout of the combined sewer system. Finally pipe diameters of this network are designed with the time area method and performance (flooding) is assessed with SWMM (Rossman, 2008). Figure 2: Agent-based model according to (Urich et al., 2010) 2.2.1 Evolution of Networks To consider the dynamic evolution of networks the authors added a time depended activation of catchments as starting points of the agents. The agent-based model is executed in several iterations (time steps). In each time step new catchments (new developments) are added to the system by placing new agents in the centre of the newly activated catchments. The new agents try to find a connection to the network (junctions) from the previous time step. To ensure efficiency of the agents the junctions of the network are marked with an attraction field (Urich et al., 2011). Again, after multiple generations the UDM extracts the network out of the attraction field of successful agents and connects it to the existing network. The new network is used in the next time step as input. 4 2.2.2 Street Networks The layout of drainage networks is driven mainly by the layout of street networks. By adding an additional layer to the agent based model this boundary condition can be considered. An agent decides in which direction it moves based on the – so called – “probability field of the neighbourhood”. This means that the agent is more attracted by fields with higher values. To consider the street network in the agent’s decision the probability field is multiplied with the neighbourhood of a raster map that contains the street network. For example, a raster cell containing a street is marked with 1 others with 0.0002. 0.0002 means, if the agent is equally attracted to cells with and without street before the probability field is multiplied with the field of the street network, that in 1 out of 5000 movments the agent prefers a cell that is not below a street. This makes it very unlikely for the agent to move into a cell without a street. Although this method has the advantage that agents can still find alternative pathways in case they hit a dead end of a street. Additionally, the influence of the street network can be adjusted, so the influence for main trunks is lower. 2.3 Case Study Scotchman's Creek The presented methodologies are tested with the Scotchman’s Creek Catchment, Melbourne, Australia. Scotchman's Creek was populated in the 1960s mainly for residential purpose. For this paper, we used data from the drainage network for the reconstruction of the urban dynamics, the street networks and digital elevation map for the network generation. Figure 3 shows the master plan including the main water infrastructure of Scotchman’s Creek as used in the DAnCE4Water model. As main infrastructure, we placed a retard basin and an outlet that connects the catchment to the main water ways of Melbourne. m Figure 3 Master plan for Scotchman's Creek 2.3.1 Reconstruction of the historic urban development based on construction age of the drainage system To compare the evolution of the drainage network generated by DAnCE4Water we reconstruct the dynamics of the existing urban environment as input for the BPM. The UDM uses a grid based representation of the urban environment, in this work 100x100m. Based on the raster grid we look at the pipe's construction age within the cell and identify when the first pipe was built. This is also 5 assumed to be the year in which the area got populated. As result we get a time line of the urban development. 3 3.1 RESULTS & DISCUSSION Reconstruction of the urban development based on the drainage data Based on the construction age of the drainage networks the authors reconstruct the urban dynamics. Figure 4 shows the results of this reconstruction. At the top we can see the state of the system in 1960 (yellow trunks), the main trunk is already in place to drain the sparse populated urban environment (red). The main trunk is following the natural creek and in the new built up environments we can see that the layout is following the street networks. Figure 4. Urban environment and drainage network (1950-1960) Figure 5 shows the dynamics of the urban development between 1980 and 2000. The snapshots show that a cluster of new land was released for residential development. The same behaviour was found all over Scotchman's Creek. However, the urban development is not following a particular pattern in which the blocks of land are populated. The figures 4 and 5 also show the second stage of the design process for the pipe network – new blocks of land for residential development are connected with a secondary pipe network to the main trunk. The secondary main trunks are following the street network and the parcel boundaries. The results of the urban development are used in the next section as input. 6 Figure 5. Urban environment and drainage network (left 1950-1980; 1950-2000) 3.2 3.2.1 Dynamics in DAnCE4Water Catchment Master Plan By using the agent-based model, the authors generate the main trunk of the drainage system for Scotchman's Creek. Starting points for the agents are the centre of the blocks. Figure 6 on the left shows the results of one realisation of the stochastic agent based process. On the right traces of 350 realisations are shown. The generated main trunk is following in major parts the existing main trunk. The agent based model was able to connect the main trunk with the main retard basin in the sink and also to connect the retard basin with the catchment outlet. m Figure 6. Left: one main trunk, right: traces of 350 trunks generated within DAnCE4Water 7 3.2.2 Connecting new Built up Areas with the Main Trunks Figure 7 on the left hand side shows the second stage of the planning process in year 2011. The BPM connected all new built up areas from 1950-2011 to the main trunk by generating a secondary drainage network. The secondary drainage network is following the road network. Figure 7 (right) shows the real drainage network (pink) and the network generated by the BPM (yellow). A visual comparison of the real and generated networks shows good agreement. The major difference is that the existing drainage network is much denser. The reason for that is that for the real network also pipes on the parcel-parcel boundary are used to drain the water. The model could be enhanced by using parcel boarders in addition to the street network as input for the agent-based. Figure 8 shows the evolution of the urban drainage network. The blue pipes are built before 1960 the red pipes between 1960-1965. We can see that based on the dynamic input from the UDM that new areas get connected to the existing urban drainage system. Figure 7. Secondary drainage network generated within DAnCE4Water Figure 8. Evolution of the Drainage Network in DAnCE4Water 8 4 SUMMARY AND CONCLUSION To identify possible transition strategies for our urban water systems DAnCE4Water models the complex coherences between societal system, urban environment and the urban water system under different scenarios like climate changes, changes in the societal needs and urban changes. DAnCE4Water uses three models to describe the urban system. In this paper the authors focused on the linkage between the urban development module and the bio-physical module. By coupling the urban environment with the urban water infrastructure it is possible to investigate adaptation strategies under consideration of a dynamically evolving urban water system. Before the authors tested the presented approach on the Scotchman’s Creek case study we analysed the development of the urban environment and drainage network. Therefore, we reconstructed the urban dynamics out of the construction age of the drainage network. The results showed that Scotchman’s Creek was following the two stage planning approach. First, the main trunk was built along the existing creek connecting the loosely populated urban environment. Next, the secondary network following the road network was build. To compare the results we used the reconstructed data from the Scotchman’s Creek case study as input for the urban development as well as a drainage master plan containing a pond and an outlet that connects the catchment with the Melbourne’s water ways. An agent-based model was used in both stages of the planning process. The results showed that the presented method successfully followed the path of the existing main trunk. In the second stage planning process, we connected the new developed areas dynamically over time to the main trunk. The BPM could successfully evolve the urban drainage network based to dynamic input of the UDM. A comparison with the existing network structure revealed that the existing drainage system is denser. The reason is that for the real network also pipes on the parcel-parcel boundary are used to drain the water. This can be solved by using the parcel instead of the street network as landscape for the agent-based model. Consequently, in this paper the authors could successfully mimic a realistic planning process and the evolution of an urban drainage network. In the next project phase DAnCE4Water will be applied for Scotchman’s Creek case study. Therefore, the urban development model is set up and linked with the infrastructure module the reproduce the historical development of the urban environment and the drainage infrastructure. 5 ACKNOWLEDGEMENT This research is part of a project that is funded by the EU Framework Programme 7 PREPARED: Enabling Change. 6 REFERENCES Australian Bureau of Statistics. (2008). 3222.0 Population projections AUSTRALIA 2006 to 2101. Canberra. Bach, P., Urich, C., McCarthy, D., Sitzenfrei, R., Kleidorfer, M., Rauch, W., & Deletic, A. (2011). Characterising a city for integrated performance assessment of water infrastructure in the DAnCE4Water model. Proceedings of the 12th International Conference on Urban Drainage, Porto Alegre/Brazil, 10-15 September 2011. Porto Alegre/Brazil. Batty, M. (2005). Cities and Complexity: Understanding Cities with Cellular Automata, Agent-Based Models, and Fractals (p. 589). The MIT Press. 9 de Haan, J., Ferguson, B., Brown, R., & Deletic, A. (2011). A Workbench for Societal Transitions in Water Sensitive Cities. Proceedings of the 12th International Conference on Urban Drainage, Porto Alegre/Brazil, 10-15 September 2011. Porto Alegre/Brazil. Djordjević, S., Butler, D., Gourbesville, P., Mark, O., & Pasche, E. (2011). New policies to deal with climate change and other drivers impacting on resilience to flooding in urban areas: the CORFU approach. Environmental Science & Policy, 14(7), 864-873 Howe, C. and P. v. d. Steen (2008). SWITCH - A Systems Approach to Urban Water Management. 18th Annual International Symposium of the INCOSE 2008, 6th Biennial European Systems Engineering Conference, The Netherlands. Rauch, W., Bach, P. M., Brown, R. R., Deletic, A., Ferguson, B., de Haan, J., McCarthy, D. T., Kleidorfer, M., Tapper, N., Sitzenfrei, R., Urich, C. (2012). Modelling transition in urban drainage management. 9th International Conference on Urban Drainage Modelling. Belgrade. Rossman, L. A. (2008). Storm water management model User’s manual version 5.0. Water Supply and Water Resources Division National Risk Management Research Laboratory Cincinnati. National Risk Management Research Laboratory - U.S. Environmental Protection Agency. Stedman, L. (2012). Developing tools for tomorrow’s urban water services: project TRUST investigates leading-edge solutions. Water Utility Management International, 7(1) Feature 15. Urich, C., Bach, P., Hellbach, C., Sitzenfrei, R., Kleidorfer, M., McCarthy, D. T., Deletic, A., et al. (2011). Dynamics of cities and water infrastructure in the DAnCE4Water model. Proceedings of the 12th International Conference on Urban Drainage, Porto Alegre/Brazil, 10-15 September 2011. Porto Alegre/Brazil. Urich, C., Sitzenfrei, R., Möderl, M., & Rauch, W. (2010). An agent-based approach for generating virtual sewer systems. Water Science & Technology, 62(5), 1090–1097. 10