A Distributed and Object-oriented Rainfall-runoff Simulation Model with High Spatial Resolution Impervious Surface

In this study we develop a distributed and object-oriented rainfall-runoff simulation (DORS) model with incorporation of 1-m high spatial resolution impervious surface area (ISA) data obtained from true-color digital orthophoto. This physically based model simulates hydrologic processes of precipitation interception, infiltration, evaporation and evapotranspiration, soil moisture and change of water table depth, runoff routing, ground water routing and channel routing. The model takes land-cover based objects as spatial units to reduce data volume, increase computational efficiency, strengthen representation of watersheds and utilize the data in various scales. The model is able to extract the relationship between neighboring objects such as slope, flow direction and border length and uses them for the runoff and ground water routing. We validate the DORS model within two watersheds with USGS discharge measurements in the state of Rhode Island. Total relative variation, ratio of absolute error, and Nash coefficient are -3.6%, 9.6%, and 0.998 for one watershed and 3%, 12.4% and 0.8 for another. The validation result indicates that the DORS model is capable of capturing the relationship between rainfall and runoff in the study watersheds. The comparison between results using high and low spatial resolution ISA demonstrates that incorporation of high spatial resolution ISA improves obviously the performance of hydrologic model. INTRODUCTION Increasing impervious surface area (ISA) resulted from urban and suburban development is a particularly important component of human-induced land-use and land-cover change (LULCC). ISA is a critical factor in cycling of terrestrial runoff and associated materials to and within ocean margin waters. Increasing ISA impacts watershed hydrology in terms of influencing the runoff and associated erosion and nonpoint pollutions (Arnold and Gibbons, 1996). Wegehenkel et al. (2006) found that a precise estimation of settlement areas in a catchment together with an improved estimation of the degree of actual imperviousness of these areas is a must for precise calculations of surface runoff and the flood peaks even in a rural catchment. Hydrologic modeling is an effective method to evaluate the impact of ISA on watershed hydrology. Reported studies employed hydrologic modeling to quantify the impacts of land-use and land-cover change on hydrological regimes at various scales (Dunn and Mackay, 1995; Ott and Uhlenbrook, 2004). The lumped, semi-distributed and distributed models were extensively used for their advantages. In terms of temporal and spatial change of parameters in hydrologic models, distributed models have advantages over lumped models because lumped models use aggregated and empirical parameters that lack clear physical meanings. Moreover, it is difficult to evaluate change of model parameters on hydrologic process based on lumped models (Kuchmenta et al., 1996). The distributed hydrologic models are promising because these models are based on physical, chemical and biological theories and the simulation results are more reliable. The performance of distributed model depends on their representation of watersheds. Those grid-based models represent and apply watershed heterogeneity in terms of distributed information of land use and land cover, slope, soil and rainfall (Jain et al., 2004). Lacking high spatial resolution ISA data, most of current hydrologic models used estimated ISA from land cover data or assigned values for specific land cover types (Ott and Uhlenbrook, 2004). However, as ISA is a key parameter in the runoff production and routing, using estimated percentage of ISA instead of precise ISA data may cause considerable errors in the hydrologic modeling. Another challenge is the data from various scales. Incorporation of geographic information system and remote sensing can reduce the number of calibration parameters in hydrologic models (Jain et al., 2004; Schumann et al., 2000). However, as remote sensing data are from diverse sources and have different spatial resolutions, it is difficult ASPRS 2008 Annual Conference Portland, Oregon ♦ April 28 May 2, 2008 to exploit all information from remote sensing data in grid-based models. In hydrologic modeling, there are problems of information lost in scaling using coarser spatial resolution units and challenges of huge data volume and associated computing time using finer spatial resolution units. As an appropriate spatial parameterization scheme and physically based descriptions of hydrologic processes are necessary for evaluating the effects of local physical characteristics such as land cover configuration, especially ISA feature on hydrologic response, we develop a distributed and object-oriented rainfall-runoff simulation (DORS) model with incorporation of high spatial resolution ISA to study the rainfall-runoff relationship. It takes objects, group of pixels, as the spatial units to reduce data volume, increase computational efficiency, strengthen representation of watersheds and utilize the data in different scales. The model derives the objects from land cover types with the help of topography and limitation of the objects size. This model is different from current distributed models that use empirically estimated proportions of ISA from land cover data. Instead, the DORS model utilizes ISA obtained from 1-m spatial resolution digital orthophoto data for improved hydrologic modeling. MODEL STRUCTURE We adopt and modify algorithms from the distributed hydrologic models by Chen et al.(2005) and Wigmosta et al. (1994). These models processed the regular grids as spatial units and simulated the groundwater and overland water movement. For our application of rainfall-runoff study, we modify the algorithms of hydrologic processes in these grid-based models and develop a new model based on spatial units of objects and add important processes such as infiltration. The structure of an object in horizontal and vertical view is described in Figure 1. The horizontal boundary of the simulated area is a watershed delineated from a digital elevation model (DEM). Vertically, the simulation extends from the saturated zone to the top of vegetation canopy. The model divides the study watershed into basic spatial units based on land cover and DEM data. Furthermore, the model splits the objects with large area into small ones to make the runoff and groundwater route more reasonably. The model treats each spatial unit, the object, as a unique vegetation-soil system. Figure 1. Hydrologic components in an object. The DORS model includes major components of: segmentation; parameterization; interception; infiltration; evapotranspiration; saturated flow; overland flow; and channel flow. The parameters of precipitation, solar ASPRS 2008 Annual Conference Portland, Oregon ♦ April 28 May 2, 2008 radiation, DEM, land cover, ISA, leaf area index (LAI), and soil properties parameters are major inputs to the DORS model. All input parameters are spatially transformed to object level after the segmentation process. Other accessorial parameters such as manning coefficient for runoff routing are determined based on land cover types. The major output is discharge at outlet of watershed. Other information such as spatial distribution of runoff and evapotranspiration can be retrieved at any time in the simulation. MODEL COMPONENTS Segmentation and Parameterization The first process in the DORS model is segmentation of land cover data. In this process, the model employs the constraint of region size to avoid over-sized objects. The relationship between objects such as length of boundary is important for runoff and ground flow routing. Different from grid-based model in which pixel interacts with its regular 4 or 8 neighbors, object deals with all neighbors in the DORS model. The effective length of boundary between two objects can be calculated as follows. W TW n i Flow i Pixel e ∑ = − = 1 , ) cos( θ θ (1) where e TW is total effective boundary length, n is number of neighboring pixels between two objects, Flow θ is aspect angle of flow direction, i Pixel , θ is aspect angle of the ith neighboring pixel direction, and W is pixel size. After segmentation and construction of relationship between neighboring objects, the model imports precipitation, solar radiation, DEM, ISA and LAI for each object from various sources. With the segmentation based on land cover, the obtained objects keep the most homogeneity for the parameters such as LAI inside the object. We employ ISA in 1-m spatial resolution to retrieve the ISA percentage in each object. We use vegetation index of Simple Ratio (SR) from Landsat TM in 30-meter spatial resolution to extract LAI, and assign soil type to each object according to majority of inside pixels in each object. We build lookup table between soil types and properties (Rawls et al., 1992). We use lookup table between land cover types and prosperities to derive parameters, such as manning coefficient, specific to land cover (Sande et al., 2003). Interception, Infiltration and Evapotranspiration Vegetation intercepts the precipitation when it falls. Part of the precipitation will be held in the foliage and then evaporate. Previous studies determined maximum interception storage capacities of the layers as a function of LAI of the layer and proportion of the gaps in the layer (Dickinson et al., 1991). In simulating this process, the model deducts the intercepted precipitation from total precipitation for infiltration and runoff production. The precipitation will fall on the ground if there is still precipitation after interception. Part of this effective precipitation will infiltrate into soil. As its broad application (EPA, 1998; Neitsch et al., 2001), we choose the Green & Ampt algorithm to simulate this process. The major parameters used in this algorithm include hydraulic conductivity, capillary pressure and effective soil porosity. We obtain these parameters from the built lookup table according to soil types. ISA plays important role in this process and the model assumes that there is no infiltration on ISA. In this study the model incorporates the high spatial resolution ISA in simulation of this process. We use 3 categories of soil without canopy, overstory canopy without understory, and overstory canopy with understory for evaporation and evapotranspiration simulation. The model calculates the evaporation and transpiration separately for the 3 categories. We employ the Penman-Monteith equation to estimate both evaporation and transpiration rates (Chen et al., 2005). The model assumes that water detained in the canopy evaporates at a potential rate. The model updates the water detained in foliage by amount of the evaporation and the soil moisture with deduction of the evaporation from soil and transpiration from vegetation. Runoff Runoff is the major component for channel flow, and it affects processes such as soil erosion and material movements in watershed. Runoff is the key factor that causes temporal change of stream discharge. This model employs De-Saint Venant equations of continuity and momentum to simulate flow of runoff (Julien, and Saghafian, 1991). The model uses a diffusive approximation to describe the runoff flow assuming that net forces acting along