Development of an integrated model for the Campaspe catchment : a tool to help improve understanding of the interaction between society , policy , farming decision , ecology , hydrology and climate

Management of water resources requires understanding of the hydrology and hydrogeology, as well as the policy and human drivers and their impacts. This understanding requires relevant inputs from a wide range of disciplines, which will vary depending on the specific case study. One approach to gain understanding of the impact of climate and society on water resources is through the use of an integrated modelling process that engages stakeholders and experts in specifics of problem framing, co-design of the underpinning conceptual model, and discussion of the ensuing results. In this study, we have developed such an integrated modelling process for the Campaspe basin in northern Victoria, Australia. The numerical model built has a number of components: – Node/link based surface water hydrology module based on the IHACRES rainfall-streamflow model – Distributed groundwater model for the lower catchment (MODFLOW) – Farm decision optimisation module (to determine irrigation requirements) – Policy module (setting conditions on availability of water based on existing rules) – Ecology module (determining the impacts of available streamflow on platypus, fish and river red gum trees) The integrated model is component based and has been developed in Python, with the MODFLOW and surface water hydrology model run in external programs, controlled by the master program (in Python). The integrated model has been calibrated using historical data, with the intention of exploring the impact of various scenarios (future climate scenarios, different policy options, water management options) on the water resources. The scenarios were selected based on workshops with, and a social survey of, stakeholders in the basin regarding what would be socially acceptable and physically plausible options for changes in management. An example of such a change is the introduction of a managed aquifer recharge system to capture dam overflows, and store at least a portion of this in the aquifer, thereby increasing the groundwater resource as well as reducing the impact of existing pumping levels. Published by Copernicus Publications on behalf of the International Association of Hydrological Sciences. 2 T. Iwanaga et al.: Development of an integrated model for the Campaspe catchment


Introduction
Flooding is the natural disaster that has caused the most damages to the world.The Pearl River Delta Area (PRDA) in southern China is in the monsoon area, where heavy storms in the summer are observed very often, and flooding is the most frequent natural disaster threatening the safety of the local society.For the past decades, PRDA has experienced rapid urbanization, and caused significant changes of land use/cover (LUC).Urban land area increased sharply and has become the major or even the dominant land use type, thus speeding up the flood responses, i.e., the increased flood peak flow, and the decreased runoff routing time.This, in one side, strengthens the flood quantities, while on the other side, as more and more people move and live and work in the urban area, flood damages increased largely with urbanization.Dongguan City is one of the newly developed international metropolitans in PRDA in the past three decades, but the flood mitigation measurements were not built to catch up with the flood changes due to rapid urbanization.Purpose of this study is to evaluate the impact of urbanization (LUC change) on flood peak flow and flood runoff coefficient on the typical watershed, the Shigu creek, so to provide scientific base for smart flood measurement planning and building in Dongguan City.
Watershed hydrological model is the tool for watershed flood process simulation in evaluating the impact of LUC change on flooding.Early hydrological model used for flood simulation/prediction is the so called lumped model, such as the SCS model (USDA, 1972), and the Xinanjiang model (Zhao, 1977).Lumped model calibrates model parameters with the observed hydrological data series, for this reason it could not be used in data spare watershed, i. e., in the watersheds with rapid urbanization.Distributed hydrological model is the further development of hydrological model under the big data era.With the development of satellite remote sensing, large quantities of the watershed terrain data was prepared and could be used for hydrological model setting up, thus providing opportunity for distributed hydrological model.Many distributed hydrological models have been proposed, such as the SHE model (Abbott et al., 1986), and the VIC model (Liang et al., 1994).
In this study, the Liuxihe model (Chen et al., 2010), a physically based distributed hydrological model is employed to simulate the flood processes of Shigu creek in Dongguan City, which is a highly urbanized small watershed that is also a very typical watershed in Dongguan City. 2 Shigu creek and data

Shigu creek
Shigu creek locates in the south of Dongguan City center, it originates from the south peak of Shuilian mountain, and is the first order branch of Dongjiang River.Shigu creek is a small watershed in Dongguan City, previously it is a watershed in the suburb area of Dongguan City center, but with the rapid urbanization of Dongguan City in the past three decades, Shigu creek becomes a part of Dongguan City.The drainage area of Shigu creek is 19.29 km 2 , and the length is 11.87 km.Climate in the creek is belong to subtropical marine climate with an annual average temperature of 23.6 • , and an annual average precipitation of 1769.5 mm.Heavy storms in the summer monsoon season usually induces serious flood which is a great threat to the security of the downtown area of Dongguan City.Rapid urbanization in the creek makes flooding worse.Figure 1 is a sketch map of Shigu creek, and shows its location in China.

Terrain property data
Terrain property data used for Liuxihe model structure setting up and parameter deriving are DEM, soil type and land use/cover (LUC) type.In this study, the DEM was downloaded from the Shuttle Radar Topography Mission database website at http://srtm.csi.cgiar.org,its spatial resolution is at 30 m × 30 m, as shown in Fig. 2a.From the result it has been found that the highest elevation in the watershed is 368 m, the lowest elevation is 0 m, and the average elevation is 40.94 m.
The soil type was downloaded from FAO soil type dataset at http://www.isric.org, the spatial resolution is 1000 m × 1000 m, as shown in Fig. 2b.There are only two soil types, including Ferric acrisols and Cululic anthrosols, accounting for 83.23, 16.77 % of the total watershed area respectively.Due to the rapid urbanization, the LUC type in Shigu creek is in constant changing.Chen et al. (2017a) prepared a LUC dataset for the whole Dongguan City at 12 date from 1987 to 2015 by using the Landsat series satellite imageries, and the SVM algorithm is employed to estimate the LUCs.Spatial resolution of this dataset is at 30 m × 30 m. Figure 3 shows the LUCs of Shigu creek extracted from this dataset from 1987 to 2015 respectively.From the results of Fig. 3, it could be found that the LUC types have not changed.There are 6 LUC types, including forestry land, grass land, farm land, urban land, water body and bare land.But the rates of the LUCs have changed drastically from 1987 to 2015, particularly the urban land area increased rapidly, while the grass land decreased significantly.

Hydrological data
In the Shigu creek, there is no river discharge observation, but there are three rain gauges installed in recent years, and only one rain gauge data is available, so it is a hydrological data spare watershed.For this reason, model parameters could not be optimized, and can only be derived directly from the terrain properties.In this study, precipitation of 3 storms was collected at half hour interval, which will be used as the input to the hydrological model for flood simulation.

Introduction of Liuxihe model
Liuxihe Model (Chen, 2009;Chen et al., 2010) is a physically based distributed hydrological model proposed for watershed flood forecasting.In Liuxihe model, the studied watershed is divided into grids horizontally by using the DEM, and are treated as a uniform watershed.Every grid has its own elevation, LUC type and soil type, and is categorized as one of the 3 types, namely hill slope cell, river channel cell Evapotranspiration is calculated at cell scale, so does the runoff production.Runoff is categorized as surface runoff, subsurface runoff and underground runoff.Surface runoff produced on the hill slope first, and is routed through the hill slope to its adjacent river cell using the kinematical wave approximation method.Runoff routing in the river channel is treated as river flow routing, and calculated by using the one dimensional diffusive wave approximation method.
The parameters of Liuxihe model are categorized as climate-related parameters, topography-related parameters, soil-related parameters and LUC-related parameters.All parameters are derived directly from their related terrain properties, this makes it be used in hydrological data spare watershed.If there is hydrological data observation, Liuxihe model can optimize model parameters by using PSO algorithm automatically (Chen et al., 2016).
Standard software has been developed for Liuxihe model set up, parameter optimization and flood forecasting (Chen and Dong, 2013), which has several independent software components, including Evapotranspiration Modular, Runoff Production Modular, Runoff Routing Modular and Parameter Deriving Modular.A public computer infrastructure for implementing Liuxihe model has been built which utilizes big data, parallel computing and supercomputing technologies.
Liuxihe model has been widely used in China for small and medium sized watershed flood forecasting (Liao et al., 2012;Chen et al.,2013), reservoir inflow flood forecasting (Chen et al., 2010;Huang et al., 2017), and large scale watershed flood forecasting (Chen et al., 2017b;Li et al., 2017), and experiences have been accumulated in the past application.

Structure set up of Liuxihe model
The watershed is divided into cells with the collected DEM in this study, the model spatial resolution is set to be 30 m × 30 m.There are 21 434 cells in total, which is further categorized as 21106 hill slope cells and 328 river channel cells, no reservoir cell is included.The river channel system is divided into 2 orders, and 13 virtual sections have been prepared after setting 8 nodes.The cross-section sizes of the virtual sections have been estimated by referencing to satellite remote sensing imageries.The Liuxihe model structure is shown in Fig. 4

Parameterization of Liuxihe model
Topography-related model parameters are flow direction and slope, these two parameters are derived from the topography data and are not adjusted.In this study, flow direction and slope are derived by using D8 method (Jenson and Dominggue, 1988;Strahler, 1957) based on the DEM.The derived flow direction and slope of Shigu creek are shown in Fig. 5. Climate-related parameter is evaporation capacity, it is estimated as 5 mm day −1 for every cells based on local experiences and Liuxihe model parameterization practice.LUCrelated parameters are evaporation coefficient and roughness.According to the past experiences of Liuxihe model parameterization (Chen et al., 2010), evaporation coefficients for all cells are set to 0.7, and the roughness of forestry land, grass land, farm land, urban land, water body and bare land are proposed as 0.55, 0.18, 0.36, 0.01, 0.06 and 0.12.The soil-related parameters are soil water content at saturated condition, field condition and wilting condition, the soil layer thickness, the soil hydraulic conductivity at saturation and the soil characteristics coefficient.According to local experiences, the soil characteristics coefficient takes 2.5, and the soil layer thickness of Ferric acrisols and Cululic anthrosols are set to 850 and 1000 mm respectively.The soil water content at saturation condition and field condition, and the hydraulic conductivity at saturation condition are determined by the Soil Water Characteristics Hydraulic Properties Calculator (Arya and Paris, 1981), and the soil water content at wilting condition is estimated as 30 % of the soil water content at saturated condition.

Results and discussion
The flood events were simulated by using the model set up in this paper, the input is the precipitation collected in this paper.3 floods have been simulated, and the simulated flood hydrographs are shown in Fig. 6.Runoff coefficients and peak flows for each storms were calculated as shown in Tables 1  and 2. The maximum runoff coefficient is 0.745, the minimum runoff coefficient is 0.461, and the averaged runoff coefficient of all the 3 storms is 0.626.The urbanization has a significant influence on the flood characteristics of the Shigu creek.Compared with 1987, the average peak flow of 2015 increased by 19.47 m 3 s −1 (42 %), while the average runoff coefficient increased by 0.135 (24.9 %).With the increasing of the urbanization rate, the runoff coefficient and the peak flow have been increasing sharply.
It can be seen that the results are in line with our expectations.The impact of urbanization on flood response is significant.In the process of urbanization, the LUCs have changed greatly.Therefore, the LUC-related parameters and soil-related parameters in the model should be different during the study period.However, for lacking of observed discharge, the parameters cannot be optimized.As a result, in the course of the follow-up study, it is necessary to give comprehensive analysis of parameter optimization in urban area without observed discharge.

Conclusion
Results in this study shows that the urbanization of Shigu creek is very rapid, and this change has a significant impact on the flood characteristics of the Shigu creek.Compared with 1987, the average peak flow of 2015 increased by 19.47 m 3 s −1 (42 %), while the average runoff coefficient increased by 0.135 (24.9 %).This will provide sound input for flood planners and managers in planning and managing the flood of Shigu creek.
Data availability.In this study, the DEM was downloaded from the Shuttle Radar Topography Mission database website (http:// srtm.csi.cgiar.org),and the soil type was downloaded from FAO soil type dataset (http://www.isric.org).The LUC types were extracted from the whole Dongguan City LUC dataset that was prepared by Chen et al. (2017a).For which, the raw data is not put in an open access dataset, but available by writing to the contacting author by email at eescyb@mail.sysu.edu.cn.The hydrological data was provided by the department of water resources of Dongguan City inclusively for this study, and is not available for the public.
Competing interests.The authors declare that they have no conflict of interest.Special issue statement.This article is part of the special issue "Innovative water resources management -understanding and balancing interactions between humankind and nature".It is a result of the 8th International Water Resources Management Conference of ICWRS, Beijing, China, 13-15 June 2018.

Table 1 .
Runoff coefficients of the 3 storms.The 3 storms were collected on 7 May 2010, 26 July 2010 and 28 May 2016 respectively.In this paper, they were written as 20100507, 20100726 and 20160528 respectively.