Development of a modular streamflow model to quantify runoff contributions from different land uses in tropical urban environments using Genetic Programming
Introduction
Increasing urbanization has severely altered the hydrological cycle in many places worldwide, accelerating runoff due to a decrease of pervious areas and therefore infiltration. In order to efficiently drain the increase in surface runoff, intensive drainage networks are often built to prevent flash floods during heavy storm events (Marshall and Shortle, 2005). However, as cities are dynamically expanding, the continuous increase of impervious surfaces and the accompanied excess runoff often exceeds the present channel capacity resulting in local flash floods. To reduce the impact of surface runoff, water sensitive urban infrastructure (e.g. green roofs, porous pavement, bioretention ponds, swales) retaining rainfall and enhancing infiltration rates in urban cities are being promoted (Burns et al., 2012, Chang, 2010). Water Sensitive Urban Design (WSUD) is an engineering design approach which aims to minimize hydrological and water quality impact of urban development by integrating land use planning with urban water management (Singh and Kandasamy, 2009). The implementation of such technologies requires a detailed understanding of runoff contributions from each specific land use in order to plan the location of these local source control measures. Therefore, a better understanding is needed regarding rainfall–runoff processes in urbanized areas, including an accurate assessment of contributions from different land uses towards quickflow as well as baseflow. This understanding would be essential for integrated management and sustainable development of water resources particularly in tropical megacities which are dependent on water sources that are more vulnerable to inter-annual fluctuations in precipitation.
Land use and land cover affect catchment hydrology primarily through changes in hydrological processes such as infiltration, rainfall interception, and evapotranspiration (Calder, 1993, Calder, 2005, DeFries and Eshleman, 2004, Potter, 1991, Tran and O’Neill, 2013) which may have significant effects on rainfall–runoff processes and catchment water yields (Roa-García et al., 2011). The various contributions from different land uses towards rainfall–runoff processes have attracted worldwide attention, especially in temperate urban regions (e.g. Burns et al., 2005, Diaz-Palacios-Sisternes et al., 2014, Loperfido et al., 2014, Miller et al., 2014). Comparing runoff generation from different land uses enables us to understand the rainfall–runoff response influenced by particular catchment components and processes and their contribution towards the overall catchment. This understanding contains valuable information with regards to a physical based understanding of rainfall–runoff behaviour when designing appropriate water management infrastructure in tropical megacities. However, it is interesting to note that a review of the literature shows that to date, no detailed investigation has been done to assess the impact of different land uses on rainfall–runoff processes for tropical urban cities.
To evaluate the impact of different land uses on catchment hydrology, rainfall–runoff processes need to be simulated. There are multiple Rainfall–Runoff (R–R) models available that can be applied to simulate rainfall–runoff processes; each one characterized by a different level of complexity, limitations and data requirement (Sorooshian et al., 2008). Physically-based models usually incorporate simplified forms of physical laws and are generally non-linear, time-varying and deterministic, with parameters that are representative of watershed characteristics. Although these models enhance our understanding towards the physics of hydrological processes, they require significant computational time and large amounts of data (Beven, 2012, Dye and Croke, 2003). Over the past decades, machine learning tools such as Artificial Neural Network (ANN) (e.g. Jeong and Kim, 2005, Kisi et al., 2013, Sudheer et al., 2002, Talei and Chua, 2012) and Genetic Programming (GP) (e.g. Babovic, 2005, Babovic and Keijzer, 2002, Babovic and Keijzer, 2006) have been used to develop rainfall–runoff models. GP offers advantages over other data driven techniques since it is more likely to generate a function with understandable structure. However, most data driven models are one unit models with adequate input variables that cover all system processes in one input/output structure (Abrahart and See, 1999, Bowden et al., 2005). Such models combine all the various flow components losing valuable information on their specific contributions which is needed when designing local mitigation measures (Corzo and Solomatine, 2007). In addition, covering all the rainfall–runoff processes in one unit without taking into account the different physically interpretable sub-processes may lead to low accuracy in extrapolation. Streamflow is commonly conceptualized to include baseflow and quickflow (also called direct runoff) components. The baseflow component represents the relatively steady contribution to streamflow from groundwater flow, while the quickflow represents the additional streamflow contributed by surface flows (i.e. rapid runoff) and shallow subsurface flows (delayed runoff) (Beven, 2012). One way of retaining as much information as possible is to build separate models for each of the different physically interpretable flow components leading to a modular approach. As such, a modular model for the simulation of streamflow time series consisting of separate modular units for baseflow and stormwater runoff would be suitable in quantifying both flow components in a more flexible manner. The concept of a modular model has been used in modelling tools that use a linear reservoir approach (e.g. unit hydrograph methods) by splitting streamflow into baseflow and quickflow components. However, these models may fail to represent the nonlinear dynamics in the rainfall–runoff process (Rajurkar et al., 2002).
Therefore, this paper used GP to develop a physically interpretable modular universally applicable model accounting for baseflow and quickflow. The modular model was applied to address the following research questions in a tropical urbanized system:
- •
What are the contributions of the various land uses towards quickflow?
- •
How does the baseflow contribution change among sub-catchments with different land uses?
- •
How do runoff generation processes vary among the different types of rainfall events?
- •
What are the effects of antecedent catchment conditions on runoff response?
In this paper, a description of the study site as well as monitoring network is described in Section 2. Section 3 focuses on the methodology used to develop the modular model consisting of baseflow and quickflow components using GP. The methodology with regards to the quantification of land use specific contributions to the quickflow component is presented in Section 4. The results are discussed in Section 5 and lastly, conclusions are summarized in Section 6.
Section snippets
Description of the study area
Kent Ridge Catchment, a small catchment (0.085 km2) within the National University of Singapore (NUS), located in the southern part of Singapore was chosen to setup an intensive monitoring network (Fig. 1). This catchment contains all the main land uses of Singapore and hence is representative from a hydrological point of view. Furthermore, the use of a small catchment reduces data uncertainty and inaccuracy with regards to the spatial distribution of precipitation. The overall topography of the
Development of a modular model using GP
A modular model for simulating streamflow can be defined as:where is streamflow (L3/T), is baseflow (L3/T), is quickflow (L3/T).
In Meshgi et al. (2014) a generalized empirical equation has been derived to estimate baseflow time series using GP with minimal data requirements and preservation of physical catchment information. According to Meshgi et al. (2014), differences between baseflow time series simulated by a groundwater numerical model (i.e.
Clustering analysis
In a tropical area, catchment responses to the rainfall events are expected to vary significantly from event to event due to different types of rainfall events and antecedent catchment conditions (Peng and Wang, 2012). Therefore, rainfall events were divided into clusters and sub-clusters based on types of rainfall events and antecedent catchment conditions using a statistical hierarchical clustering technique proposed by Ward (1963).
The following variables were used: total precipitation in the
Approximating quickflow time series
The following empirical equation, based on the quickflow time series filtered from the observed streamflow data using Eq. (1), with a minimal RMSE was selected:where presents the quickflow (), is the rainfall intensity (mm/min) with L being minutes of lag time, is the total rainfall depth during the event (mm), is the normalized daily
Summary and conclusion
Meteorological, physiographic, hydrologic and land use data was used to derive a physically interpretable modular model consisting of a baseflow module and a quickflow module. The structure of the derived modular model, using GP, was simple and physically interpretable. The quickflow module contained a rapid and delayed streamflow generation component which corresponds to the overland flow and shallow sub-surface flow, respectively.
The modular model was generalized to predict rapid and delayed
Acknowledgments
The authors gratefully acknowledge the support and contributions of Singapore-Delft Water Alliance (SDWA). The research presented in this work is carried out as part of the SDWA’s Multi-objective Multiple Reservoir Management research program (R-303-001-005-272). The authors also would like to thank Dr. Abhay Anand for his assistance in MATLAB coding. They are also very grateful to the reviewers whose insightful comments guided improvements on the original manuscript.
References (52)
- et al.
Input determination for neural network models in water resources applications. Part 1—background and methodology
J. Hydrol.
(2005) Effects of suburban development on runoff generation in the Croton River basin, New York, USA
J. Hydrol.
(2005)- et al.
Hydrologic shortcomings of conventional urban stormwater management and opportunities for reform
Landsc. Urban Planning
(2012) - et al.
A method for detecting and describing land use transformations: an examination of Madrid’s southern urban-rural gradient between 1990 and 2006
Cities
(2014) - et al.
Evaluation of streamflow predictions by the IHACRES rainfall–runoff model in two South African catchments
Environ. Modell. Soft.
(2003) - et al.
The effect of initial soil water content and rainfall intensity on near-surface soil hydrologic conductivity: a laboratory investigation
CATENA
(2006) - et al.
Effects of rainfall intensity, underlying surface and slope gradient on soil infiltration under simulated rainfall experiments
CATENA
(2013) - et al.
Modeling rainfall–runoff process using soft computing techniques
Comput. Geosci.
(2013) - et al.
Effects of distributed and centralized stormwater best management practices and land cover on urban stream hydrology at the catchment scale
J. Hydrol.
(2014) - et al.
An empirical method for approximating stream baseflow time series using groundwater table fluctuations
J. Hydrol.
(2014)
Assessing the impact of urbanization on storm runoff in a peri-urban catchment using historical change in impervious cover
J. Hydrol.
River flow forecasting through conceptual models part I — a discussion of principles
J. Hydrol.
Effects of land use, land cover and rainfall regimes on the surface runoff and soil loss on karst slopes in southwest China
CATENA
Effects of land cover on runoff coefficient
J. Hydrol.
Influence of lag time on event-based rainfall–runoff modeling using the data driven approach
J. Hydrol.
Detecting the effects of land use/land cover on mean annual streamflow in the Upper Mississippi River Basin, USA
J. Hydrol.
Multi-model data fusion for river flow forecasting: an evaluation of six alternative methods based on two contrasting catchments
Hydrol. Earth Syst. Sci.
Data mining in hydrology
Hydrol. Process.
Genetic programming as a model induction engine
J. Hydroinform.
Rainfall runoff modelling based on genetic programming
Nordic Hydrol.
Rainfall–Runoff Modeling Based on Genetic Programming, Encyclopedia of Hydrological Sciences
Rainfall–Runoff Modelling: The Primer
Discharge Measurement Structures, 20. International Institute for Land Reclamation and Improvement
Hydrologic Effects of Land-use Change
Blue Revolution: Integrated Land and Water Resources Management
Cited by (0)
- 1
Present address: The International Water Management Institute, East Africa & Nile Basin Office, Addis Ababa, Ethiopia.