Remote Sensing of Environment 86 (2003) 385 – 400 www.elsevier.com/locate/rse Satellite and ground-based microclimate and hydrologic analyses coupled with a regional urban growth model S. Traci Arthur-Hartranft a,*, Toby N. Carlson b, Keith C. Clarke c a Penn State University, P.O. Box 8048, Philadelphia, PA 19101, USA Department of Meteorology, Pennsylvania State University, University Park, PA 16802, USA c Department of Geography, University of California at Santa Barbara, Santa Barbara, CA 93106, USA b Received 25 March 2002; received in revised form 25 September 2002; accepted 28 December 2002 Abstract Urban development is shown to induce predictable changes in satellite-based measures of radiant surface temperature and evapotranspiration fraction—as long as certain features of the development are known. Specifically, the vegetation changes that accompany the development and the initial climatic state of the land parcel must be noted. Techniques are also developed for quantifying the effects of urbanization on the surface hydrology at a watershed scale. Streamflow and precipitation data are related graphically in order to determine a watershed’s general ratio of stormwater runoff to rainfall, along with any changes in the ratio over time. Four distinct runoff responses, separated by season and antecedent moisture conditions, are distinguishable for a particular basin, with the response during the non-summer months under typical antecedent moisture conditions the most representative of and responsive to land-use patterns. This particular runoff response can be estimated from satellite-derived land cover patterns and certain physical attributes of a basin. These satellite-based microclimate and hydrologic analyses are coupled with an existing urban growth model (SLEUTH). The SLEUTH urban growth model simulates future development scenarios for a region of interest. The resulting changes in urban land use lead to the evolution of site-specific climate and hydrology based on the scheme that is presented in this paper. This study, as well as related tools and bodies of knowledge, can be used to broaden the scientific basis behind land-use management decisions. D 2003 Elsevier Science Inc. All rights reserved. Keywords: Urban environment; LANDSAT; Development; Hydrology; Modeling 1. Introduction At the beginning of the 21st century, the modern metropolis has evolved into a vast urban field with multiple centers (Spirn, 1984). The environmental impact of the land cover change associated with this urbanization pattern is the basis of the work presented in this paper. Local land cover, i.e., grass, concrete, soil, water, etc., largely dictates the energy exchanges that occur between the earth and atmosphere and thus, is one of the primary determinants of a site’s microclimate. The idea that an architect can design a building and its immediate surroundings in order to achieve a desirable microclimate is now a traditional one. Much less commonly, * Corresponding author. Current affiliation: Lockheed Martin Remote Sensing Systems Integration, P.O. Box 8048, Philadelphia, PA 19101, USA. Tel.: +1-610-531-5592. E-mail address: [email protected] (S.T. Arthur-Hartranft). however, is the idea extended to an entire region. Individual land-use decisions become integrated into meso-climatic zones that mirror the form of urban development, but rarely is the theory that a regional planner can ‘‘control the mesoclimate of broad zones of the city’’ put into practice (Chandler, 1976). As late as 1996, despite many attempts to model the effects of urban development on the local climate, complete models for urban planning were unavailable and there was ‘‘still a striking gap between climate and design’’ (Eliasson, 1996). The first two authors, Carlson and Arthur (2000), took a step towards closing this gap by using the technique of multiple linear regression to formulate equations for predicting the effect of various urban development plans on the micro-scale surface temperature and moisture. Data for the analysis was obtained from a combination of satellite remote sensing and surface climate modeling; the validity of the parameters used (fractional vegetation cover, radiant surface temperature, evapotranspiration fraction, surface moisture 0034-4257/03/$ - see front matter D 2003 Elsevier Science Inc. All rights reserved. doi:10.1016/S0034-4257(03)00080-4 386 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 availability and percentage developed land) has been previously established (Arthur, Carlson, & Ripley, 2000; Carlson, Arthur, & Ripley, 1997; Gillies, Carlson, Cui, Kustas, & Humes, 1997; Owen, Carlson, & Gillies, 1998). This paper continues the work presented in Carlson and Arthur (2000), as well as introduces a new parameter, stormwater runoff, which specifically addresses surface hydrology. Additionally, it presents a demonstration of the potential to operationalize the research by coupling the microclimate and surface hydrology results with a cellular automaton urban growth model (SLEUTH) developed by the third author (Clarke, Hoppen, & Gaydos, 1996). The resulting regional urban planning model will enable users, for the first time, to clearly incorporate surface microclimatic and hydrologic changes into decision-making. The SLEUTH model is used to drive future development scenarios, while the start-up microclimate and surface hydrology for the region of interest are determined from a satellite-derived land cover map. As the SLEUTH model predicts new growth and generates changes in land cover, site-specific climate and hydrology evolve based on the work presented here. The SLEUTH growth prediction is thus advanced beyond a qualitative look at development patterns and towards an environmental assessment of a region’s future. Additionally, the quantitative information derivable from multi-spectral satellite imagery is finally presented in a form that is of use to policy makers. 2. Methodology 2.1. Microclimate 2.1.1. Background The urban heat island has been described as ‘‘one of the most clearly established examples of inadvertent modification of climate’’ (Roth, Oke, & Emery, 1989). When detection of heat islands is based on thermal infrared satellite and airborne data, the heat island intensity is greatest during the day and least at night—the opposite of results from studies based on air temperatures (Roth et al., 1989). The use of a daytime radiant surface temperature to monitor the impact of land-use change emphasizes urbanization’s influence on the surface energy balance, a feature which also allows for the extraction of surface moisture properties. Soil wetness is the atmospheric boundary condition second only to sea surface temperature in its impact on climate; over warm continental areas during the spring and summer, it is considered the most important factor (Dirmeyer, Dolman, & Sato, 1999). Providing moisture for evaporation, it increases the surface latent heat release and thus limits the daily maximum temperature (Dai, Trenberth, & Karl, 1999). Vegetation, with its ability to transpire, also plays a large role in the distribution of energy between the latent heat flux and the sensible heat flux. When soil moisture is not limiting, higher levels of vegeta- tion result in increased transpiration and redistribute the energy fluxes towards substantially reduced Bowen ratios and thus cooler and moister near-surface climates (Bounoua et al., 2000). These three physical parameters (radiant surface temperature, evapotranspiration/soil moisture, and vegetation) are combined in the ‘triangle method’ as presented in the works of Carlson, Capehart, and Gillies (1995), Carlson, Gillies, and Perry (1994), Gillies and Carlson (1995), Gillies et al. (1997) and Owen et al. (1998). The method conceptually relates variations in satellite-derived radiant surface temperature and fractional vegetation cover, while coupling the interpretation of their association to an inverse modeling scheme. 2.1.2. Calculations from imagery Radiant surface temperature (T) was derived from the thermal band of the Landsat Thematic Mapper (TM) sensor. At-sensor radiances were converted to apparent temperatures using an empirical form of Planck’s function. To correct to surface values, MODTRAN (an atmospheric radiative transfer model) was applied using a standard mid-latitude summer sounding, the mean terrain altitude and an estimate of atmospheric visibility (Kneizys et al., 1996). An NDVI (normalized difference vegetation index) was defined using the TM’s near infrared and visible red wavelength bands. For these wavelengths, conversion of the at-sensor reflectances to surface values was not necessary since a scaled NDVI, as used here, is insensitive to atmospheric correction (Carlson & Ripley, 1997). Scaling anchor points for both the radiant surface temperature (T) and the NDVI were determined using each image’s T/NDVI scatterplot. Fig. 1, adapted from Carlson and Arthur (2000), displays the recurring shape that is seen for such plots—as long as the image used covers a large, heterogeneous area and as long as it is taken during normal sunlit drying conditions. If the image is acquired shortly after rain or during a drought, the shape will be modified. Once the cloud and water pixels are removed from Fig. 1, the scatterplot resembles a truncated triangle. The radiant surface temperature of any given pixel is hypothesized as determined by the relative amounts of bare surface and vegetation viewed by the sensor, as well as the surface moisture conditions. Scaled values, represented by N* and T*, are defined as: N* ¼ NDVI NDVIo NDVIs NDVIo T* ¼ T Tmin Tmax Tmin ð1Þ NDVIs is set at the value where the scatterplot folds over and becomes flatter near the top of the distribution of points and where it is assumed that NDVI has reached saturation at 100% vegetation cover. NDVIo defines the base of the triangle where it is assumed bare soil is represented. Tmax is the result of extrapolating the ‘‘warm edge’’ down to a possible hottest surface, and Tmin is representative of the full S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 387 Fig. 1. Scatterplot of radiant surface temperature and NDVI adapted from Carlson and Arthur (2000). An area of approximately 90 100 km for an AVHRR scene near Philadelphia, PA is represented. canopy temperature, generally similar to that of the air. N* and T* thus each vary between 0 and 1. An N* of 0 represents bare soil and 1, full vegetation cover. A T* of 0 is typical of a surface at air temperature and 1, of the hottest surface in an image. The scaled NDVI was then converted to a physical index, fractional vegetation cover (Fr), defined as the proportion of a pixel covered by vegetation and related though Fr c N2* (Choudhury, Ahmed, Idso, Reginato, & Daughtry, 1994; Gillies & Carlson, 1995). The final transformed scatterplots of fractional vegetation cover (Fr) and scaled radiant surface temperature (T*) were then compatible with the output from a soil – vegetation – atmosphere transfer (SVAT) model (Carlson, 1986; Carlson, Dodd, Benjamin, & Copper, 1981). The underlying constraint in this model is the balance of the energy fluxes at the earth’s surface. Initialization was based on a ‘‘typical’’ morning sounding along with land surface parameters representative of the study area. An equation for predicting the fraction of net radiation (Rn) used in evaporative processes (ET/Rn—ranging from 0 to 1) as a function of Fr and T* was developed from the model output using multiple regression analysis. The physical interpretation of the T*/Fr scatterplot was used to set the domain for the output of the SVAT model simulation. Contours of constant ET/Rn were overlaid on the space defined by the T*/Fr scatterplot by interpreting the defined ‘‘warm edge’’ (see Fig. 1) as the upper limit to possible surface temperatures due to dryness. T*, Fr and ET/Rn were generated via satellite imagery in order to monitor microclimate changes over time and their possible correspondence with alterations in land use. Thus, the Landsat TM images were georeferenced to a UTM map base, each with a total root mean square error less than 1, i.e., the average positional error was less than approximately 30 m. Each scene was then re-sampled to a 25-m grid using the nearest neighbor criteria. To generate land cover maps from the TM images, a maximum likelihood supervised classification scheme was used. Six classes were clearly separable: dense development (commercial), less dense development (residential), bare soil, short vegetation, forest and water. For use with the SLEUTH urban growth model, however, classes were combined into four Anderson Level I land cover categories: urban or built-up land, agricultural land, forested land and water (Anderson, Hardy, Roach, & Witmer, 1976). An error analysis of each satellite-derived land cover map was completed by generating an error matrix (Congalton, 1991). The overall accuracy and the KHAT statistic from the error matrix were set at a threshold of 80% for acceptance. Off-diagonal errors were used to determine inter-class confusion and guide reclassification efforts. When microclimate and land use are combined, Table 1 shows that the average T* and ET/Rn values for a given land class vary little from image to image. Note, however, that for a given year, T* increases and ET/Rn decreases as the land class changes from forest to agriculture to residential to high density development—reflecting warming and drying. Images were selected for clear days in the absence of clouds, Table 1 Average ET/Rn and T* values for all of the pixels classified in each land cover class for the given year of image acquisition Land cover class 1987 1997 1987 1997 (mean ET/Rn) (mean ET/Rn) (mean T*) (mean T*) Forest Agriculture (bare/vegetated) Residential High density 0.61 0.58 0.62 0.57 0.15 0.29 0.13 0.30 0.54 0.46 0.55 0.45 0.35 0.51 0.33 0.53 The sampled area was a 61-mile2 basin in southeastern Pennsylvania. 388 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 rain and abnormal climatological conditions such as drought, helping to assure that the only substantial, nonreversible surface climate changes observed over time were due to urbanization. 2.2. Surface hydrology 2.2.1. Background If urbanization is observed to affect the local energy balance through its influence on the turbulent fluxes of latent and sensible heat, the hydrologic balance must likewise be impacted. The water balance for an urban volume can be expressed as: p þ I ¼ r þ E þ DS ð2Þ where p and I serve as the system inputs—precipitation and piped-in water, respectively—r is net runoff, E is evapotranspiration, and DS is the net water storage change. Since the latent heat term in the energy balance equation is based on the mass flux of water, E, the two balances are inextricably linked. The typical view is that urbanization leads to a decrease in both evapotranspiration and storage and an increase in runoff. Traditional stormwater management tends to focus on the latter. Detailed computer models are available that attempt to simulate the urban system of pipes, gutters, sewers and storage ponds/reservoirs in order to predict storm runoff with a focus on pre- and post-development conditions. The work presented in this paper differs from these runoff prediction models in that its goal is to develop an index that can monitor change in the hydrologic environment in a manner compatible with the output from the SLEUTH urban growth model. The method is essentially what is known as ‘‘black box’’—the output, storm runoff, is produced in response to the input, total storm precipitation, without detailed consideration of the physical processes involved (Dingman, 1994). As with Woodruff and Hewlett (1970), the premise of this work was not ‘‘to probe delicate relationships among obscure parameters that are difficult to determine, but rather to identify physical attributes of watersheds that can be easily measured on available maps, yet might readily indicate how the watershed handles its storm precipitation.’’ Surface hydrology is defined as the spatial and temporal storage and redistribution of rainfall as it falls on or enters into the soil (Engman, 1997). The various physical paths the rainfall may take collectively generate what is known as the storm hydrograph, a plot of discharge vs. time during the period when a stream or river is affected by precipitation. This plot characterizes the hydrologic response of the basin. The work presented in this paper closely follows Boughton (1986, 1987), which present a graphical technique for determining the proportion of a watershed that contributes overland flow for different storms, as well as Boyd, Bufill, and Knee (1993, 1994). Graphical analysis was used to determine if urban- ization resulted in a greater proportion of the basin contributing to runoff for a given storm depth. Changes in the contributing area were related to the general land-use types and patterns that are predictable by the SLEUTH urban growth model. The focus for this study was on the storm runoff signal rather than total yield or low flows. 2.2.2. Study area and stream data methodology Eleven watersheds and seven rain gauges were selected in southeastern Pennsylvania. Site selection was limited by the availability of daily mean discharge data from the United States Geological Survey (USGS) and 24-h precipitation data from the National Climatic Data Center (NCDC). The selected basins also had to lie within cloud-free areas of the Landsat TM scenes that were used for land cover mapping. The sites were preferably small (less than about 60 mile2), relatively urbanized, and exhibiting urban growth over the study period—although some sites with no growth were chosen as controls, and one basin with a drainage area of 324 mile2 was also included. All data was downloaded from the internet at http://waterdata.usgs.gov/nwis/sw/ and http:// www.ncdc.noaa.gov. Discharge data was recorded in cubic feet per second (cfs) and precipitation data in hundredths of inches. All data was left in the units in which it was acquired. Due to the empirical nature of the equations in this study, it is necessary that these same units be used throughout this paper, despite the convention for SI units. Each watershed was checked for changes in baseflow during the study period by examining October 1 daily discharge values over time; no significant trends were found. This date (the start of the water year) was chosen since it represents the approximate point in time when the soil moisture should be drawn down to its lowest annual value and flow is generally supported by groundwater (Black, 1991). Total storm runoff was related to total storm precipitation for approximately 20 years worth of data in each basin. The total storm runoff depth for each rain event was approximated by removing the previous day’s flow as baseflow and then applying the storm’s daily mean discharge value to a 24-h period and dividing by the basin’s area. Negative values, representing periods of drawdown in streamflow, were set to zero runoff. This method assumed that the stream’s response lasts only for the given day, so that the flow returns to pre-storm levels within 24 h. It thus had the potential to miss a substantial amount of runoff during any recession flow that may occur after the primary day of the rain event. This error should have been minimal, however, since the majority of the basins studied were relatively small. The 324-mile2 basin did display a general lag between a day’s rain event and the discharge response. The precipitation values for this site were thus moved forward in time one day in an attempt to auto-correct the data. An alternative technique to determine stormflow from USGS daily mean discharge data was presented by Woodruff and Hewlett (1970) and is recommended for any similar studies in the future. S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Additional manual processing was used in order to reduce potential sources of error. First, events that were contaminated by snowfall, as indicated by the NCDC 24-h precipitation records, were automatically removed from the analysis. A scan was then run for implausible events—that is, days when the runoff generated was greater than the rainfall. For the basins studied here, these events generally displayed one of two scenarios: a heavy snowfall had recently occurred or the rain event lasted for 2 or more consecutive days, such that the tail-end precipitation became associated with the main discharge response in the data. If the event was the first rainfall after a previous snow, it could have induced melting of the snowpack and increased the runoff above the day’s actual precipitation depth. Such events were removed since they involved other physical processes than the straightforward rainfall – runoff relationship that was being sought. If the rainfall was observed to span consecutive days, the daily rainfall amounts and relevant discharge values were cumulated into one storm, with all runoff values calculated relative to pre-storm conditions. For the remaining ‘inconsistent’ cases where neither of the above scenarios was applicable, the data could often be corrected by replacing presumed misrepresentative rainfall measurements with data from a nearby site or, in the presence of particularly strong spatial rainfall gradients, the event was removed from the dataset. Since the amount of runoff from any given storm will depend highly on the soil moisture deficiency at the beginning of rainfall, the final rainfall –runoff pairs were stratified by antecedent moisture conditions, using the antecedent precipitation index (API), given by Kohler and Linsley (1951) as: APIi ¼ Pi þ 0:92APIiI ð3Þ The index for any given day, i, is equal to 0.92 of the previous day’s value. The choice of this recession factor is not critical, although values generally range from 0.85 to 0.90 over most of the eastern and central portions of the United States (Kohler & Linsley, 1951). If rain occurs on any given day, then the amount of rain observed is added to that day’s index. Since the concern is the API at the start of the rain event, each pair of rainfall – runoff values was related to the previous day’s API. Essentially, the API acts as a surrogate for a basin’s soil moisture status, which integrates much of the land surface hydrology by controlling the infiltration and surface runoff processes (Engman, 1997). Histograms of all the rain events’ API values suggested some natural cutoff points for separating dry antecedent conditions from typical and saturated conditions. API values less than 0.5 in. were assumed to represent extremely dry conditions and those greater than 3.0 in., extremely moist or saturated conditions. All events in between were considered to be a collection of typical scenarios. 389 The annual time series of discharge and precipitation displayed a seasonal pattern, with summer events generating less runoff than non-summer events with the same API. The effect that antecedent precipitation has on subsequent storm runoff strongly depends on the extent to which the precipitation is dissipated between storms through such processes as evapotranspiration. To account for this, the API recession factor could be set to vary throughout the year, or, alternatively, a second variable—season—can be introduced into the analyses. For this study, the rainfall – runoff points were further stratified by season. Events occurring between June 21 and October 1 were defined as part of the summer regime, while those that occurred outside this time period were designated as non-summer events. These dates were selected based on an annual water budget for southeastern New York (Black, 1991). Potential evapotranspiration was noted to exceed the actual value around mid-June, and a decrease in storage occurred until near the end of September. The particular rainfall – runoff relationships obtained in this study are specific to the geographical region. For example, model simulations of soil moisture regimes in Pennsylvania identified all of the sites as lying within the region of the greatest summer deficit (Waltman et al., 1997). The rainfall – runoff pairs, stratified by both season and API, were used to generate characteristic plots of runoff vs. rainfall for each basin (Fig. 2). The plots were used to determine the fraction of the basin that contributed to streamflow during a storm event, as well as the initial rainfall losses that must have occurred before runoff began. If a basin is fully saturated from prior rain events, 100% runoff will theoretically occur from both pervious and impervious surfaces, and the slope of the runoff vs. rainfall plot will be 1 and the intercept 0. Alternatively, a basin entirely covered by impervious surfaces will produce a similar graph. With this interpretation, the slope in Fig. 2 represents the fraction of the drainage basin that contributes to runoff, and the intercept represents the depth of rainfall that must occur before runoff commences (Boyd et al., 1993). The separation of rainfall – runoff events by season, and then further by API within each season, succeeded in discriminating four rainfall – runoff lines, such that the source area for runoff was less in the summer and in general, for dry antecedent conditions. The least squares linear regression best-fit line was used within each group to estimate the contributing area. 2.3. SLEUTH urban growth model On their own, the climatological and hydrological analyses are limited in function as far as their practical role in urban planning. The ability to estimate changes in these environmental parameters is most useful if a scenario for the growth that drives these changes can be made. Since the animation of urban growth alone is not capable of providing an environmental evaluation, coupling such work 390 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Fig. 2. Visualization of the runoff vs. rainfall graphical method used to characterize a basin’s runoff response. results in an enhancement of both efforts. This paper presents one of the first attempts at using the SLEUTH urban growth model as a basis for assessing the ecological and climatic impacts of urban change and estimating the sustainable level of urbanization in a region (Clarke, Hoppen, & Gaydos, 1997). The SLEUTH urban growth model has been calibrated for and applied to datasets of the Baltimore– Washington corridor and the San Francisco Bay area (Clarke & Gaydos, 1998; Crawford-Tilley, Acevedo, Foresman, & Prince, 1996; Kirtland et al., 1994). Based on cellular automata, the model begins with a set of initial conditions and then uses local rules to describe how each cell changes in response to its neighbors. From these responses, complex behavior emerges across the whole grid of cells (Theobald & Gross, 1994). Calibration of the model uses historical maps of land use and urban extent to determine the coefficients that will guide the growth rules during a predictive run. These decision rules and their resulting behavior are based primarily on the physical factors that control development patterns. A satellite-derived land cover map of the study region provides the starting point for the predictive run, and additional GIS-based layers such as slope, reserve areas, water bodies, and road networks drive the future development scenarios. Despite the use of guiding growth coefficients, the model is not stagnant. It has the ability to self-modify, meaning that different time periods will be dominated by different growth behavior (Clarke & Gaydos, 1998). In this paper, the purpose of the use of the SLEUTH model is not to generate a thorough and validated growth run for a particular study site. The intent rather is to present a demonstration of a proof of concept for making the hydrology and climate analyses operational. The result is not necessarily the production of definitive predictions but rather the development of a useful tool for scenario planning—a tool for imagining, testing and choosing between possible futures (Clarke & Gaydos, 1998). 3. Results and discussion 3.1. Microclimate In the work of Carlson and Arthur (2000), the equations for predicting changes in scaled radiant surface temperature, DT*, and evapotranspiration fraction, D(ET/Rn), due to growth-induced land-use transitions were developed from NOAA AVHRR and Landsat TM satellite analyses of Chester County, Pennsylvania. The Landsat TM satellite data was used strictly for land cover mapping and served as the basis for determining the change in urban development that occurred within each 1-km2 AVHRR pixel. The climate parameters—temperature, vegetation and surface moisture—were all derived from the AVHRR data. For the work presented in this paper, the AVHRR imagery was no longer used, and instead, 1-km2 averages of TM-derived climate variables formed the basis of the analyses. A comparison of results from this new TM-based work with the work of Carlson and Arthur (2000), DT* and DET/ Rn multiple regression equations demonstrate improved R2 values, as well as increased significance for individual coefficients. However, the TM-based analyses do not identify the land class ‘‘near a large water body and natural reserve area’’ as distinct. This was the class that, in the work of Carlson and Arthur (2000), stood out as especially sensitive to urbanization. In general, the sensor-based data and equations in this paper and those by Carlson and Arthur (2000) cannot be interchanged. For each sensor, the data that was used to develop the equations had slightly different ranges and inter-variable relationships, so one sensor’s equation will not yield physical results when used with the other sensor’s data. Overall though, the comparison S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 391 Table 3 Initial conditions for use with the climate module in the SLEUTH urban growth model served to validate the original work. With the exception of the one distinct initial land class, the similarity in the local climate change predicted for various scenarios from these completely independent analyses was encouraging. Although based on the same techniques, one set of equations used 1-km2 averages of TM climate variables for watersheds throughout southeast Pennsylvania, while the other set of equations used AVHRR climate variables, a different study site and different TM scenes for land cover mapping. When used with the SLEUTH urban growth model, the DET/Rn and DT* predictions are no longer one-time calculations. The climate fields are updated each year that a landuse map is predicted based on any new growth that has occurred, as well as the surface climate conditions from the previous time step. Since the parameter that was predicted in the last time step should not be used to generate a new prediction in the present step, the form of the original equations given in Carlson and Arthur (2000) was altered slightly. The initial ET/Rn value (ET/Rn0) was removed from the DET/Rn equation, and likewise, the initial T* value (T*0) was removed from the DT* equation. The resulting final equations for use with the SLEUTH urban growth model are given in Table 2. The original equations used the surface moisture availability (Mo) as an indicator of surface wetness in the DT* equation. Also an output of the SVAT model, isopleths of Mo can be overlaid on the T*/Fr scatterplot and defined for each pixel in an image. When used with the SLEUTH model, however, Mo is only available for the first time step and so it was replaced with the predicted variable, ET/Rn. During equation development, satellite images were used to derive each pixel’s initial T* and ET/Rn values. Users of the SLEUTH urban growth model cannot be expected to obtain satellite images of their areas and work with both an image processing program and a SVAT model to determine these values. Thus, Table 3 can be used to initialize the climate variables from the land-use map that serves as initial input to the growth model. T* and ET/Rn are estimated by aggregating the individual pixels of the user’s land cover map into 1-km2 areas and determining the percentage of developed land within the parcel and the nature of the remaining land (i.e., wooded or agricultural). The result of the initial land class on the predicted climate change and the general behavior of the regression Undeveloped agricultural Undeveloped wooded Lightly developed, otherwise agricultural Lightly developed, otherwise wooded Moderately developed, otherwise agricultural Moderately developed, otherwise wooded Greater than 50% developed T *0 Fr0 ET/Rn0 0.26 0.19 0.29 0.26 0.35 0.33 0.43 48.2 68.7 46.0 60.0 43.8 54.2 39.4 0.59 0.60 0.57 0.58 0.54 0.55 0.50 Undeveloped regions were defined by less than 10% developed land, lightly developed by 10% to 25%, and moderately developed by 25% to 50%. The remaining land was predominantly agricultural if the 1-km2 region was less than 5% forested. If the undeveloped land was classified as at least 10% forests, the remaining land was set as predominantly wooded. equations themselves can be understood through a series of land-use change scenarios. These scenarios were applied to each initial land type based on the sequence of inputs for Ddev and DFr given in Table 4. The idea was to start with the greatest impact—equal and opposite changes in development and vegetation—and then progress to the most ‘‘friendly’’ form of development, i.e., that which actually increases the amount of vegetation. This scenario could occur if, for example, landscaped residential development occurs on bare agricultural fields. Stages in between these extremes represent development with no impact on vegetation (for example, commercial development on barren land) or development with a low level of impact on vegetation (for example, residential development that takes care to remove only enough vegetation for the actual house area). The DT* and DET/Rn equations in Table 2 were used for each of these scenarios within a descriptive initial land class; then the sequence of scenarios was repeated for the next initial class. The results are displayed in Fig. 3. The descriptive land classes for the initial conditions are given in the graphs from left to right, just as they are presented in Table 3 from top to bottom. The majority of the points lie above the zero line for the DT* graph, indicating that radiant surface temperature tends to increase with development. In contrast, the majority of the points for the DET/Rn graph lie below the zero line, indicating that with development, surface moisture tends to decrease. In addition, three patterns can be observed in the graphs. First, there is the general trend of decreasing climate impact Table 2 TM-based equations in the same format as that presented in the work of Carlson and Arthur (2000) Y = a + b(DFr) + c(Ddev) + d(T *0 ) + e(Fr0) + f (ET/Rn0) DT* DET/Rn a b c d e f R2 F-significance 0.4926 0.1656 0.0035 0.0016 0.0019 0.0010 N/A 0.2761 0.0020 0.0016 1.0742 N/A 0.79 0.75 < 0.0001 < 0.0001 Fr is fractional vegetation coverage; dev is developed land; T* is the scaled radiant surface temperature, and ET/Rn is the fraction of net radiation used in evapotranspiration. D indicates that the change in the independent variable over time is what is important to the analysis, while a subscript 0 indicates an emphasis on the variable’s initial conditions. 392 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Table 4 Inputs for growth scenarios Note how the sequence starts with development that has the greatest impact on vegetation then gradually decreases this impact to the point where vegetation is actually increased. Within each of these scenarios, there is another sequence occurring from the largest magnitude changes to the lowest (for example, 40 to 5). the more developed the 1-km2 parcel is before the growth modeling process begins. A second pattern is found within each descriptive land class; it shows the greatest increases in temperature and decreases in moisture for the first development scenarios, i.e., those that are most destructive of the vegetation. The third pattern lies within each of these general scenarios; the greatest impact is typically observed for the largest Ddev values. An exception to this third pattern is noted for the rather improbable final scenario— when the increases in vegetation are nearly equal to the increases in development. In this situation, the equations are dominated by the magnitude of the vegetation term. This means that when the vegetation increase is great enough (even though development is equally large), the parcel actually becomes cooler or wetter. 3.2. Surface hydrology Table 5 provides an initial exploration of the runoff vs. rainfall plots in terms of runoff source area. The relationships given are a result of applying the graphical method of Fig. 2 to summer events during the 1980s for one of the study basins in southeastern Pennsylvania. As stated in Section 2.2, API values less than 0.5 in. were assumed to represent extremely dry conditions and those greater than 3.0 in., extremely moist or saturated conditions. All events in between were considered a collection of typical scenarios. In Table 5, as the antecedent conditions became moister, the slope of the best-fit trend line increased, indicating that more of the basin was contributing to runoff. On the other hand, during extremely dry events (API < 0.5), it was likely that the impervious surfaces directly connected to the drain- age system dominated the runoff response. This fact can be seen in the data by the low slope, as well as the intercept at zero—i.e., those few surfaces that generated runoff ( < 1% of the basin in this case) did so at the first sign of rainfall. Based on such an interpretation, for the data in Table 5, about 10% of the basin contributed to runoff after approximately 0.02 in. of rain for typical API values, while under very wet antecedent conditions, it took approximately 0.04 in. of rain to then tap into about 22% of the basin. A unique response for each of the three API scenarios was distinguishable for this particular basin. In general, however, the events with API values above 3.0 in. did not present a clear relationship between rainfall and runoff. For the rest of this work, only two categories are presented: API < 0.5 in. for extremely dry antecedent conditions and API z 0.5 in. for typical conditions. When these API categories were also separated by season, four distinct runoff– rainfall relationships emerged for most of the study basins. Fig. 4 provides an example of these relationships using a different southeastern Pennsylvania basin during the 1980s. The linear regression best-fit trend line is included on each plot. Note the difference in scale between typical and dry conditions. The relationship shown in Fig. 4 between the runoff vs. rainfall slopes for this particular basin is typical of results in this study. The greatest runoff for a given rain event was generated in the non-summer months under typical antecedent moisture conditions (API z 0.5 in.). When antecedent moisture was very low, the runoff was reduced. Overall, there was less runoff in the summer, with the lowest amounts of the year generated during this season under dry conditions (API < 0.5 in.). These patterns are readily understandable in terms of physical processes (Pielou, 1998). During the non-summer months in southeastern Pennsylvania, the cool temperatures and scarcity of growing vegetation allow evapotranspiration to consume only so much water, often much less than that supplied by precipitation. Particularly under typical antecedent moisture conditions, more of a basin’s pervious areas have experienced a recharge in soil moisture, and the proportion of the basin contributing to storm runoff increases. During the summer months, when plants are actively growing and temperatures are warm, transpiration and evaporation are vigorous and rapid, such that actual evapotranspiration is generally held below its potential by a lack of water. High potential evapotranspiration values, soil moisture withdrawals and possible water deficits help to explain why the summer slopes in Fig. 4 are lower than the corresponding ones during the non-summer months. Using a different technique, Betson (1964) also found a seasonal runoff pattern, with high runoff occurring during the winter and low runoff during the summer. During the winter, when the amount of moisture stored in the soil profile was high, runoff occurred from a somewhat larger proportion of the watershed than during the summer, when the contributing area was smaller. Likewise, using a water balance model, Grimmond and Oke (1986) showed that in S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 393 Fig. 3. Results of the growth scenarios for DT* and DET/Rn. The equations and the initial conditions for each descriptive land type were used from Tables 2 to 3, while the DFr and Ddev values were determined by cycling through the growth scenarios in Table 4. the winter ‘‘when evapotranspiration is limited and the soil moisture store is well stocked,’’ 90% of the losses are as runoff, while in the summer ‘‘the primary output is to the atmosphere.’’ Human activities, however, do have the potential to alter this typical runoff pattern. For example, summer irrigation of suburban lawns was found to artificially stock the soil moisture in Vancouver, British Columbia, and significantly impact the city’s natural seasonal water balance (Grimmond & Oke, 1986). In general, for the runoff vs. rainfall plots over all of the southeastern Pennsylvania study basins, R2 values tended to be lower in the summer months, and a few of the regressions during that period were not even significant. The runoff – rainfall relationship for a complete basin, which includes many types of surfaces, should, in fact, not necessarily be considered constant, and a high degree of temporal variability in a watershed’s runoff response is common for most regions (Dingman, 1994). Even though season and antecedent moisture conditions were considered in this study, there were many other controlling factors that were not included and could have contributed to the data’s scatter. In particular, rainfall intensity, despite its dominating influence on runoff generation, was excluded because of the 24-h nature of the precipitation dataset. The same total precipitation is capable of generating anywhere from zero runoff to near the actual depth of precipitation, depending 394 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Table 5 Three runoff – rainfall relationships based on different antecedent moisture condition regimes API (inches) Runoff = slope(rainfall) + intercept Observations R2 F-significance < 0.5 0.5 V API < 3.0 z 3.0 y = 0.009x y = 0.098x 0.02 y = 0.217x 0.04 24 217 44 0.81 0.42 0.72 < 0.0001 < 0.0001 < 0.0001 Data is for a study basin in southeastern Pennsylvania during the 1980s. Each equation is the least-squares linear regression trend line from the relevant runoff vs. rainfall plot. All storm events occurring from June 22 through September 30 were used, but they were stratified based on their antecedent precipitation index (API) value. only on the rate at which the rain falls (Smith, 1997). This factor is most important during the summer months due to the predominance of convective rainfall events. During this period, not only does the rainfall intensity tend to vary much more from event to event, but individual storms can also exhibit greater spatial variability. Another potential source of scatter represented in Fig. 4 that was not accounted for, specifically during the non-summer months, is that of frozen soil, which can act as an impervious surface. Despite this omission, these months’ runoff vs. rainfall plots under typical antecedent moisture conditions were found to be the most representative of a basin’s land-use patterns. The plots’ slopes generally displayed an increase from the 1980s to the 1990s. This increase could be related to changes in land use, particularly the spread of developed land. Such decadal changes in the slope were less apparent under dry conditions, as well as for the summer months in general. Temporal changes in the plots’ intercepts were also not obvious, although a seasonal pattern was notable. The intercept represents the loss in rainfall that must occur before runoff commences. The particular value will vary based on such factors as how much rainfall is intercepted by leaves, branches or other objects, the amount of initial infiltration, and how much water is stored in surface depressions. For highly urbanized basins, these initial losses should be small. Many of the intercepts under dry conditions or during the summer months did not differ significantly from zero. In these situations, the impervious surfaces directly connected to the drainage system or other everpresent, low-loss source areas probably dominated the runoff response. Many of the remaining pervious areas were likely not contributing to runoff, since the rainfall was used for other purposes, such as soil moisture recharge. The intercepts increased for most of the plots representing typical non-summer conditions. This was suspected to result from the additional pervious surfaces, with greater initial rainfall losses, that were also contributing to the runoff response. The rest of this paper focuses on the non-summer, typical API runoff vs. rainfall slope as a representative index of a basin’s runoff response. The slope values obtained from runoff vs. rainfall plots based on individual observations were compared to those from plots based on yearly totals, a technique that is typically accepted as a way to reduce scatter (Bedient & Huber, 1992). The correlation coefficient, r, between the slopes from these two methods was 0.87. The slope values obtained from a decade of individual events also correlated well (0.90) with those slopes obtained using just the events during the years more closely associated with the corresponding satellite image. Using a full decade worth of events helped to reduce the influence of anomalous periods, and it is believed that the results would not have differed much if certain years were instead isolated. Additionally, the use of individual observations rather than annual sums allowed the results to be presented to users of the final SLEUTH application in more tangible, eventbased terms. For several cases throughout the year and for various study basins, storm runoff depth as estimated from the basin’s representative slope value was compared to the actual calculation of the runoff depth from a real-time hydrograph constructed from 15-min USGS data. The slope-based method was found to be most reliable for non-summer events. In the summer, the predictability of the runoff depth was much less consistent. For non-summer events under typical API conditions (the exact scenario that was selected as representative of a basin’s general runoff response), data points tended to lie very close to the line representing 1:1 correspondence with the real-time hydrograph analysis. In general, however, the slope-based method should not be applied to flood stage events or those events producing an anomalously large amount of runoff. This result was to be expected since the runoff vs. rainfall slopes were determined from the method of least-squares linear regression in the absence of a large number of influential flood events. The general trend for runoff production during the more frequent, smaller rain events does not extrapolate linearly to flood stage flows. Betson (1964) likewise concluded that different relationships were justified for large and small events, with mathematical models that relate storm rainfall to runoff frequently under-predicting large events. A typical storm list contains a large number of small-to-moderate events, such that the model is forced to fit this more common situation. In contrast, a study by Jackson, Ragan, and Fitch (1977) used only those storms with a return period greater than 2 years; as a result, their model was ‘‘optimized for large runoff events’’ and was found to overestimate the flows resulting from small rainfalls. For the purposes of this paper, smaller rainfall events are more important than flood episodes. The runoff vs. rainfall plots were used to determine an index of runoff that is responsive to changes in urbanization. Man-made impervious surfaces have the greatest potential to alter the hydrologic balance during the lower rainfall amounts of more typical storm events (Corbett, Wahl, Porter, Edwards, & Moise, 1997). Under heavier rain events, land use becomes somewhat indistinct due to saturation of pervious areas. The S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 395 Fig. 4. Runoff vs. rainfall plots for a southeastern Pennsylvania study basin during the non-summer and summer months in the 1980s. For the non-summer plots, events occurring from June 22 to September 30 were excluded. For the summer plots, ONLY events occurring from June 22 to September 30 were included. The ‘‘typical conditions’’ plot is based on events with an API value z 0.5 in., while the ‘‘dry conditions’’ plot is based on events with an API value < 0.5 in. purpose of the representative runoff vs. rainfall slope is to serve as an index that communities can use to determine what influence future land development may have on the basin’s runoff response. The task then was to determine this representative value in the absence of actual data and, instead, in the presence of growth scenarios predicted by the SLEUTH urban growth model. In a first attempt to develop a methodology, each land use was assigned a particular runoff to rainfall ratio, R. This area-weighted R-value explained only 35% of the variance observed in the slopes from the runoff vs. rainfall plots. Even with the added complexity of a buffering scheme, which took into account not only the land use but also its proximity to the waterway, the ability of this method to predict a basin’s representative runoff vs. rainfall slope value did not improve. Land-use patterns alone were not able to clearly distinguish the differences observed in the various basins’ runoff responses. The ability to predict the representative runoff response improved dramatically when not just the land-use patterns, but also the size of the basin and an index of channel slope were incorporated into a least-squares multiple linear regression equation. This combination of independent variables succeeded in explaining approximately 91% of the variance observed in the actual runoff vs. rainfall slopes. However, when this equation was applied to basins outside the immediate study area, certain rather arbitrary restrictions had to be placed on the use of the equation. Plausible landuse scenarios (i.e., the total proportions summed to 1) were seen to result in unphysical runoff to rainfall ratios (i.e., 396 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 In the equation, all land uses contribute to increasing the final runoff vs. rainfall slope, with certain land classes, such as forests, contributing less than others. In addition, the larger the basin, the lower the runoff response. Woodruff and Hewlett (1970) also found that larger basins exhibit lower runoff to rainfall ratios as calculated from the storm hydrograph. They attributed this trend to attenuation of the runoff signal as the stormwaters head downstream, as well as the fact that larger basins are typically not uniformly covered by the intense storm cells that can produce strong runoff signals on small basins. Finally, the greater the elevation change from the point 80% up the channel to the outlet, the greater the runoff response. The representativeness parameter serves to pull the final runoff vs. rainfall slope value down to a reasonable level; for those basins where the range in elevation is much larger than that in the channel, the predicted value will not be reduced as much. For example, one basin was observed to have a large runoff response, despite a low measure of channel terrain. This same basin displayed its highest terrain along the southeastern tributaries near the outlet—a fact accounted for by the final term in the equation. greater than 1 or less than 0). The equations could only be used to predict the effects of the same type of land-use change at another location where the conditions were similar (Maidment, 1993). As a result, this initial regression equation was viewed as an example of successful calibration of the method for these particular southeastern Pennsylvania basins—an example of how well the technique can be applied to a particular region. For transferability to other basins, a more general equation was developed as given in Table 6. The intercept was forced through zero, and there were six independent variables: (1) the total proportion of the basin’s land that was developed, (2) the total proportion of the land that was in agricultural use, (3) the total proportion of the land that was forested, (4) the area of the basin in square miles, (5) the channel’s elevation change in meters, and (6) the representativeness of this measure of basin terrain in meters/meters. The land-use proportions were determined using the basin’s satellite-derived land-use map, with each value ranging from 0 to 1. ‘‘Developed’’ meant that the land classification was some form of urban (commercial, residential, etc.), while agricultural use was defined as land classified as either bare soil or vegetation. In this way, the land-use variables should represent, at a minimum, approximately 98% of the basin’s surface area. Depending on how well the channel can be discerned in the satellite image, the only remaining land class (water) should not compose more than about 2% of the basin. The methodology presented here was based on an ideal 24-h storm discharge response and is not suitable for basins with large storage areas, such as lakes. The fifth independent variable, elevation change, represents the decrease in elevation that occurs from a point near the headwater source area to a point at the basin outlet. The point 80% up the main channel was defined as the headwater reference, while the stream gauge site represented the point at the basin outlet. Both elevation values should be in meters. In developing the equations, the USGS Digital Line Graph hydrography data was used to discern the stream’s main channel, and a USGS digital elevation model (DEM) was used to determine elevation values. The final independent variable, representativeness, is the elevation change value divided by the range in basin elevation, i.e., maximum minimum. All measurements in this final term should be in meters. The lower this final variable, the less the channel’s measured elevation decrease represents the actual terrain potential in the basin. 3.3. Modules for the SLEUTH urban growth model 3.3.1. Microclimate Module add-ons for the SLEUTH urban growth model were developed from the climate and hydrology results presented in this paper. Specific details and steps for implementation are thoroughly described at http://www.essc. psu.edu/SLEUTH. The web site, designed for both learning about and downloading the climate and hydrology modules, provides detailed instructions on the modules’ use, as well as color examples of their output. Some of the latter will also be given here in black and white versions as an example of implementation. Fig. 5 displays the general land-use patterns of White Clay Creek, which drains both southeast Pennsylvania and northeast Delaware before it enters the Delaware Bay through the Christiana River. Currently, the Pennsylvania portion of the watershed is predominantly rural, with the greatest development being in the form of small towns and suburban clusters. In Delaware, however, the watershed includes part of the city of Newark, and rampant suburbanization is the region’s main characteristic. Despite large levels of urban growth, forested areas in the watershed’s Table 6 Determining a basin’s representative runoff vs. rainfall slope in the absence of rainfall – runoff data Y = a + b(developed) + c(agricultural) + d(forested) + e(area) + f (elevation change) + g(representativeness) Runoff – rainfall a b c d e f g R2 F-significance 0 0.7497 0.4099 0.0223 0.0007 0.0012 0.3682 0.65 < 0.05 The table presents the results from a multiple linear regression analysis with the intercept forced through zero. The independent variables include total land use proportions, basin size and two measures relating to the basin’s terrain: ‘‘elevation change’’ and ‘‘representativeness’’ (see text). This new analysis included not only the southeastern Pennsylvania basins in the 1980s and 1990s, but also two additional basins from central Pennsylvania in the 1990s. S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 397 Fig. 5. General land use for the White Clay Creek watershed in southeastern Pennsylvania and northeastern Delaware. The left image was derived from a 1996 Landsat TM satellite image, while the right image is the SLEUTH predicted land use for the year 2025. These results are not intended as a verified study of the area. White Clay Creek was chosen merely to serve as an example of implementation. southwest region have remained largely intact. A natural preserve area managed jointly by Pennsylvania and Delaware and Delaware’s White Clay Creek State Park provides approximately 7 total miles of riparian corridor along the creek. White Clay Creek has been categorized as ‘‘one of only a few relatively intact, unspoiled and ecologically functioning river systems’’ remaining in the highly congested corridor that links Philadelphia, PA with Newark, DE (White Clay Creek Wild and Scenic River Study Task Force, 1998). The site thus serves as an excellent location for application of the ideas in this paper; however, results should be viewed merely as an example of implementation, or proof of concept, not model verification. In Fig. 5, both the 1996 satellite-derived land-use map and the SLEUTH prediction generated for the year 2025 can be seen. Using the techniques of Section 3.1, the climate module generates climate layers for each year that the SLEUTH model generates a forecasted land-use map. The scaled surface temperature (T*), fractional vegetation cover (Fr), and evapotranspiration fraction (ET/Rn) are estimated for each 1-km2 land parcel. As parcels experience urban growth over time, their climate variables are updated. The user of the module is able to establish the type of development that occurs. Choices range from a large decrease in the local vegetation to the most ‘friendly’ form of development, i.e., that which actually increases the vegetation. The module outputs each land parcel’s predicted climate values in an ascii file, as well as in grayscale images for visualization purposes. Fig. 6 displays a pair of these images for T*, corresponding to the time period of the land-use patterns presented in Fig. 5. To generate this output, the ‘‘high impact’’ option was chosen as the form of development. These images have been scaled so that, in both, white represents the maximum possible T* value. Areas of warming can be seen where both new development and ‘‘filling in’’ occurred—for example, in northern regions and in the east. The climate module creates similar output for the vegetation (Fr) and the surface moisture (ET/Rn) variables. For quantitative analyses, the images should be used in conjunction with the ascii files that contain the raw data. The SLEUTH growth model has the capability to model non-urban changes in land use. This additional information is not actively employed in the climate module since the driving force for the generation of the module’s equations was urbanization. Thus, for non-urbanizing parcels of land, any other land-use transitions are ignored in the actual calculations, even though the model’s full land-use change capability is still employed. For example, if a parcel of land is initialized as completely agricultural and then never experiences development but rather, transitions back to forest, this does not influence the climate variables used in the model. Until a parcel experiences urban growth, its climate remains in its initial state. 3.3.2. Runoff response module (basin-scale) The basic hydrology module uses the techniques presented in Section 3.2 to quantify the effects of urbanization on the surface hydrology at a watershed scale. The runoff to rainfall ratio (based on the representative runoff vs. rainfall slope value) is estimated for each year that the SLEUTH model generates a forecasted land-use map. The module output is in the form of a text file that summarizes each year’s land-use patterns and the runoff to rainfall ratio. 398 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Fig. 6. Scaled radiant surface temperature images (T*) for the same area and time period as Fig. 5. The 1996 image is based on the Landsat TM-derived T* values averaged at 1-km resolution. The image on the right is for the predicted temperature field in 2025. Note that the output from the climate module is at 1-km resolution, while the original land use maps (Fig. 5) are at 25-m resolution. Limitations for application of this module are given in the user’s guide on the web site. Warnings are output if any of the basin’s characteristics are out of range. For example, the basin may be too steep or too large, one particular land use may dominate the basin, or water may comprise more than 2% of the basin. However, even if a basin’s initial land-use patterns are suitable for the module, in future years, as the land-use patterns of the basin change, some of the land proportions may exceed their usable range. The module keeps track of the specific requirements and notes when the runoff to rainfall ratio can no longer be reliably estimated. Two forms of the module are available—one for use with basins in southeastern Pennsylvania and another for more general use. The southeastern Pennsylvania version allows the user to separate the urban growth based on its proximity to the waterway, as well as predict the runoff vs. rainfall slope under dry antecedent moisture conditions. The second, more general version of the module, does not include the option to buffer the waterway and only predicts the runoff response for typical conditions. 4. Conclusions It has been demonstrated that development can induce predictable changes in the local scaled radiant surface temperature and evapotranspiration fraction. Specifically, the changes in vegetation accompanying the development must be known, as well as the initial climatic state of the land parcel. For example, development of a landscaped subdivision on dry, bare fields will have a different climatic response than the development of a commercial shopping center on land that was previously forested. These ideas were examined using multiple regression analyses based on satellite-derived land cover and climate parameters in order to develop a quantitative climate module that can be used with the SLEUTH urban growth model. Additionally, it was shown that plots of runoff vs. rainfall for a gauged basin could be interpreted in terms of the proportion of the basin contributing to a storm event’s runoff signal. For a particular basin, four distinct runoff responses, separated by season and antecedent moisture conditions, were noted. The physical premise behind this separation of the runoff vs. rainfall plots included such factors as potential and actual evapotranspiration, infiltration capacity and the soil moisture deficit. The runoff response was characterized by the slope of the plot’s bestfit linear regression trend line. The slope for the nonsummer months under typical antecedent moisture conditions was seen to be most representative of and responsive to land-use patterns. Using a satellite-derived land cover map, this particular slope value was associated with the proportion of developed, agricultural and forested land, as well as the size of the basin and several measures of the basin’s terrain. The resulting multiple linear regression equation was used to develop a module for estimating the representative runoff response for each year of predicted land-use change from the SLEUTH urban growth model. The resulting index of runoff response can be used by communities to model increases in the loss of a basin’s precipitation input due to storm runoff from the basin outlet within 24 h of typical storms. This loss of water from the system is a reduction in potential soil moisture, which is necessary both for vegetation and a cooler, more comfortable microclimate. It has the potential to lead to lower stream flows during dry periods, increased water shortages and a greater reliance on imported sources. S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Overall, this work provides a strategy that allows communities to determine when their water and climate resources might become unacceptably impacted by urban growth. A working model has been developed but remains to be actively applied and tested in a community faced with answering realworld planning questions. For example, in the White Clay Creek basin, the natural preserve area in the west was part of the SLEUTH model’s excluded layer, which prevented the land from becoming developed. This area could be stripped of its exclusion privileges and the impacts of this planning decision explored. Additionally, the climate module can be run successively with different selections for the form of development. In this way, the role that vegetation plays in determining how development impacts the local microclimate could be observed, especially for a particular parcel of land that is about to be developed. Climate layers can also be generated for each year of the SLEUTH model output and animations created. Such possibilities as these lead to a new exploratory tool for communities—a tool which helps to combine satellite, hydrologic and climate data with GIS analyses and urban studies. Acknowledgements This work was supported primarily through the Pennsylvania State University Meteorology Department. Thanks are given to David Ripley for his invaluable technical assistance, as well as to Drs. John Wyngaard, David DeWalle and Donna Peuqeut for their important input, feedback and support. The reviewers of this paper are also acknowledged for their helpful comments. We would like to thank the U.S. Department of Agriculture, specifically Thomas J. Schmugge of the Hydrology Laboratory, Beltsville, MD (contract USDA 58-1270-8-0560), and the Pennsylvania Department of Transportation (PennDOT Grant 259704 #69), specifically Bert Kisner, for their assistance in supporting this research. Satellite imagery provided by NASA is also appreciated. References Anderson, J. R., Hardy, E. E., Roach, J. T., & Witmer, R. E. (1976). A land use and land cover classification system for use with remote sensor data. Geological Survey Professional Paper, vol. 964. Washington, DC: U.S. Government Printing Office. 28 pp. Arthur, S. T., Carlson, T. N., & Ripley, D. A. J. (2000). Land use dynamics of Chester County, Pennsylvania, from a satellite remote sensing perspective. Geocarto International, 15(1), 25 – 35. Bedient, P. B., & Huber, W. C. (1992). Hydrology and floodplain analysis (2nd ed.). Massachusetts: Addison-Wesley Publishing. 692 pp. Betson, R. P. (1964). What is watershed runoff? Journal of Geophysical Research, 69(8), 1541 – 1552. Black, P. E. (1991). Watershed hydrology. New Jersey: Prentice-Hall. 408 pp. Boughton, W. C. (1986). Hydrograph analysis as a basis for rainfall – runoff modelling. Civil Engineering Transactions, The Institution of Engineers, Australia, CE29(1), 28 – 33. 399 Boughton, W. C. (1987). Evaluating partial areas of watershed runoff. Journal of Irrigation and Drainage Engineering, 113(3), 356 – 366. Bounoua, L., Collatz, G. J., Los, S. O., Sellers, P. J., Dazlich, D. A., Tucker, C. J., & Randall, D. A. (2000). Sensitivity of climate to changes in NDVI. Journal of Climate, 13, 2277 – 2292. Boyd, M. J., Bufill, M. C., & Knee, R. M. (1993). Pervious and impervious runoff in urban catchments. Hydrological Sciences Journal, 38(6), 463 – 478. Boyd, M. J., Bufill, M. C., & Knee, R. M. (1994). Predicting pervious and impervious storm runoff from urban drainage basins. Hydrological Sciences Journal, 39(4), 321 – 332. Carlson, T. N. (1986). Regional-scale estimates of surface moisture availability and thermal inertia using remote thermal measurements. Remote Sensing Reviews, 1, 197 – 247. Carlson, T. N., & Arthur, S. T. (2000). The impact of land use/land cover changes due to urbanization on surface microclimate and hydrology: A satellite perspective. Global and Planetary Change, (25), 49 – 65. Carlson, T. N., Arthur, S. T., & Ripley, D. A. J. (1997). Monitoring urbanization and urban microclimate by satellite. Proceedings, Seventh International Symposium on Physical Measurements and Signatures in Remote Sensing, Courchevel, France, 7 – 11 April, vol. 2 ( pp. 697 – 702). Carlson, T. N., Capehart, W. J., & Gillies, R. R. (1995). A new look at the simplified method for remote sensing of daily evapotranspiration. Remote Sensing of Environment, 54, 161 – 167. Carlson, T. N., Dodd, J. K., Benjamin, S. G., & Cooper, J. N. (1981). Satellite estimation of the surface energy balance, moisture availability and thermal inertia. Journal of Applied Meteorology, 20, 67 – 87. Carlson, T. N., Gillies, R. R., & Perry, E. M. (1994). A method to make use of thermal infrared temperature and NDVI measurements to infer surface soil water content and fractional vegetation cover. Remote Sensing Reviews, 9, 161 – 173. Carlson, T. N., & Ripley, D. A. J. (1997). On the relation between NDVI, fractional vegetation cover and leaf area index. Remote Sensing of Environment, 62, 241 – 252. Chandler, T. J. (1976). Urban Climatology and its Relevance to Urban Design. World Meteorological Organization, Technical Note No. 149. 61 pp. Choudhury, B. J., Ahmed, N. U., Idso, S. B., Reginato, R. J., & Daughtry, C. S. T. (1994). Relations between evaporation coefficients and vegetation indices studied by model simulations. Remote Sensing of Environment, 50(1), 1 – 17. Clarke, K., & Gaydos, L. (1998). Loose-coupling a cellular automaton model and GIS: Long-term urban growth predictions for San Francisco and Washington/Baltimore. International Journal of Geographical Information Science, 12(7), 699 – 714. Clarke, K. C., Hoppen, S., & Gaydos, L. (1997). A self-modifying cellular automaton model of historical urbanization in the San Francisco Bay area. Environment and Planning. B, Planning and Design, 24, 247 – 261. Clarke, K. C., Hoppen, S., & Gaydos, L. J. (1996). Methods and techniques for rigorous calibration of a cellular automaton model of urban growth. Proceedings, Third International Conference/Workshop on Integrating GIS and Environmental Modeling CD-Rom, Santa Fe, NM, 21 – 26 January. Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37, 35 – 46. Corbett, C. W., Wahl, M., Porter, D. E., Edwards, D., & Moise, C. (1997). Nonpoint source runoff modeling: A comparison of a forested watershed and an urban watershed on the South Carolina coast. Journal of Experimental Marine Biology and Ecology, 213, 133 – 149. Crawford-Tilley, J. S., Acevedo, W., Foresman, T. W., & Prince, W. (1996). Developing a temporal database of urban development for the Baltimore – Washington Region. Proceedings, ASPRS/ACSM Annual Convention and Exhibition, Baltimore, MD, 22 – 24 April, vol. III ( pp. 101 – 110). 400 S.T. Arthur-Hartranft et al. / Remote Sensing of Environment 86 (2003) 385–400 Dai, A., Trenberth, K. E., & Karl, T. R. (1999). Effects of clouds, soil moisture, precipitation, and water vapor on diurnal temperature range. Journal of Climate, 12, 2451 – 2473. Dingman, S. L. (1994). Physical hydrology. New York: Macmillan. 568 pp. Dirmeyer, P. A., Dolman, A. J., & Sato, N. (1999). The pilot phase of the global soil wetness project. Bulletin of the American Meteorological Society, 80(5), 851 – 878. Eliasson, I. (1996). Intra-urban nocturnal temperature differences: A multivariate approach. Climate Research, 7, 21 – 30. Engman, E. T. (1997). Soil moisture, the hydrologic interface between surface and ground waters. Remote Sensing and Geographic Information Systems for Design and Operation of Water Resources Systems (Proceedings of Rabat Symposium S3), 242, 129 – 140. Gillies, R. R., & Carlson, T. N. (1995). Thermal remote sensing of surface soil water content with partial vegetation cover for incorporation into climate models. Journal of Applied Meteorology, 34, 745 – 756. Gillies, R. R., Carlson, T. N., Cui, J., Kustas, W. P., & Humes, K. S. (1997). A verification of the ‘triangle’ method for obtaining surface soil water content and energy fluxes from remote measurements of the normalized difference vegetation index (NDVI) and surface radiant temperature. International Journal of Remote Sensing, 18(15), 3145 – 3166. Grimmond, C. S. B., & Oke, T. R. (1986). Urban water balance: 2. Results from a suburb of Vancouver, British Columbia. Water Resources Research, 22(10), 1404 – 1412. Jackson, T. J., Ragan, R. M., & Fitch, W. N. (1977). Test of Landsat-based urban hydrologic modeling. Journal of the Water Resources Planning and Management Division, 103(WR1), 141 – 158. Kirtland, D., Gaydos, L., Clarke, K., DeCola, L., Acevedo, W., & Bell, C. (1994). An analysis of human-induced land transformation in the San Francisco Bay/Sacramento Area. World Resources Review, 6(2), 206 – 217. Kneizys, F. X., Abreu, L. W., Anderson, G. P., Chetwynd, J. H., Shettle, E. P., Berk, A., Bernstein, L. S., Robertson, D. C., Acharya, P., Rothman, L. S., Selby, J. E. A., Gallery, W. O., & Clough, S. A. (1996). The MODTRAN 2/3 report and LOWTRAN 7 model. Massachusetts: Ontar. 261 pp. Kohler, M. A., & Linsley, R. K. (1951). Predicting the runoff from storm rainfall. U.S. Weather Bureau Research Paper No. 34. Washington D.C., 9 pp. Maidment, D. R. (Ed.) (1993). Handbook of hydrology. New York: McGraw-Hill. Various pagings. Owen, T. W., Carlson, T. N., & Gillies, R. R. (1998). An assessment of satellite remotely-sensed land cover parameters in quantitatively describing the climatic effect of urbanization. International Journal of Remote Sensing, 19(9), 1663 – 1681. Pielou, E. C. (1998). Fresh water. Chicago: The University of Chicago Press. 275 pp. Roth, M., Oke, T. R., & Emery, W. J. (1989). Satellite-derived urban heat islands from three coastal cities and the utilization of such data in urban climatology. International Journal of Remote Sensing, 10(11), 1699 – 1720. Theobald, D. M., & Gross, M. D. (1994). EML A modeling environment for exploring landscape dynamics. Computers, Environment and Urban Systems, 18(3), 193 – 204. Smith, R. E. (1997, July). Discussion to ‘‘Runoff curve number: Has it reached maturity?’’ Journal of Hydrologic Engineering, 145 – 147. Spirn, A. W. (1984). The granite garden, urban nature and human design. New York: Basic Books. 334 pp. Waltman, W. J., Ciolkosz, E. J., Mausbach, M. J., Svoboda, M. D., Miller, D. A., & Kolb, P. J. (1997). Soil climate regimes of Pennsylvania (p. 873). University Park, PA: Penn State Agricultural Experiment Station. White Clay Creek Wild and Scenic River Study Task Force (1998). White Clay Creek watershed management plan. Philadelphia: U.S. Department of the Interior. 195 pp. Woodruff, J. F., & Hewlett, J. D. (1970). Predicting and mapping the average hydrologic response for the eastern United States. Water Resources Research, 6(5), 1312 – 1326.
© Copyright 2026 Paperzz