Ph.D. Capstone Seminar

Editor's Notes

• #2: Hello everyone. Thank you for coming to my capstone seminar. Today’s talk is titled distributed dry deposition modeling: Sensitivity analyses and GIS coupling with a component-based approach.
• #3: This is outline of today’s talk. I will talk an overview of my dissertation. Then, introduction to problems and objectives of the study. I will talk methods, results, and discussions for two experiments. Then, brief conclusions.
• #5: Title of my dissertation is coupling of e nvironmental models and geographic information systems I developed systems to couple environmental models and GIS and performed some analyses using these systems. First, I explored lumped hydrologic model loosely coupled with ArcGIS, meaning only data are shared between the model and GIS. I performed regression and resolution impact tests with it. Then, lumped and distributed dry deposition model were developed. Lumped model is just a visual basic standalone program. So it is not integrated with GIS. With this system, dry deposition estimation and sensitivity analyses were performed. Sensitivity analysis identified important model parameters that will be employed in a distributed model in the next step. Distributed dry deposition model was coupled with ArcGIS with a tight coupling approach, in which model functions are developed within a GIS. And using that system, area identification was performed. All of the systems were developed with a component technology called COM, which I will explain later. Today, I will focus on the sensitivity analyses and the development of the distributed dry deposition model.
• #6: Title of my dissertation is coupling of e nvironmental models and geographic information systems I developed systems to couple environmental models and GIS and performed some analyses using these systems. First, I explored lumped hydrologic model loosely coupled with ArcGIS, meaning only data are shared between the model and GIS. I performed regression and resolution impact tests with it. Then, lumped and distributed dry deposition model were developed. Lumped model is just a visual basic standalone program. So it is not integrated with GIS. With this system, dry deposition estimation and sensitivity analyses were performed. Sensitivity analysis identified important model parameters that will be employed in a distributed model in the next step. Distributed dry deposition model was coupled with ArcGIS with a tight coupling approach, in which model functions are developed within a GIS. And using that system, area identification was performed. All of the systems were developed with a component technology called COM, which I will explain later. Today, I will focus on the sensitivity analyses and the development of the distributed dry deposition model.
• #8: In urban area, air may be polluted due to emissions form facilites or vehicles, which is a big issue because air pollutant cause serious human health problems. In the United States, annualy 3,700 people die due to urban ari pollutants. Worldwide, 0.8 million people die annually because of urban air pollutants. The urban polulation worldwide is growing pretty fast, now 3.3 billion and it will jump up to 5 billion by 2030. Therefore, it is crucial for city managers to effectively control air pollutant.
• #9: One way to reduce air pollutant is the use of vegetation that can reduce air pollutants through a dry deposition. Dry deposition is a process in which the air pollutant is transported from atmosphere to tree surfaces with no precipitation. It also includes intake of air pollutants through leaf stomata. UFORE-D estimates dry deposition removals by urban trees. Air pollutants estimated are carbon monoxide, nitrogen dioxide, ground level ozone, sulfer dioxide, and particulate matter. UFORE-D is very useful for urban forest managers to evaluate the effects of urban forest to control air quality. But one limitation is that it is lumped parameter model, meaning pollutant removal can be estimated for the entire city. It is impossible to determine spatial pattern of the concentration or dry deposition. For better forest management and planning spatial pattern is very important. You want to plant trees where the pollutant could be effectively removed. To do this, UFORE-D needs to be implemented in a distributed approach.
• #10: One way to reduce air pollutant is the use of vegetation that can reduce air pollutants through a dry deposition. Dry deposition is a process in which the air pollutant is transported from atmosphere to tree surfaces with no precipitation. It also includes intake of air pollutants through leaf stomata. UFORE-D estimates dry deposition removals by urban trees. Air pollutants estimated are carbon monoxide, nitrogen dioxide, ground level ozone, sulfer dioxide, and particulate matter. UFORE-D is very useful for urban forest managers to evaluate the effects of urban forest to control air quality. But one limitation is that it is lumped parameter model, meaning pollutant removal can be estimated for the entire city. It is impossible to determine spatial pattern of the concentration or dry deposition. For better forest management and planning spatial pattern is very important. You want to plant trees where the pollutant could be effectively removed. To do this, UFORE-D needs to be implemented in a distributed approach.
• #11: One way to reduce air pollutant is the use of vegetation that can reduce air pollutants through a dry deposition. Dry deposition is a process in which the air pollutant is transported from atmosphere to tree surfaces with no precipitation. It also includes intake of air pollutants through leaf stomata. UFORE-D estimates dry deposition removals by urban trees. Air pollutants estimated are carbon monoxide, nitrogen dioxide, ground level ozone, sulfer dioxide, and particulate matter. UFORE-D is very useful for urban forest managers to evaluate the effects of urban forest to control air quality. But one limitation is that it is lumped parameter model, meaning pollutant removal can be estimated for the entire city. It is impossible to determine spatial pattern of the concentration or dry deposition. For better forest management and planning spatial pattern is very important. You want to plant trees where the pollutant could be effectively removed. To do this, UFORE-D needs to be implemented in a distributed approach.
• #13: Here are objectives of this study. As a first step, I am gonna identify important input parameters for UFORE-D via sensitivity analyses. Then, identified important parameters are employed in a distributed form, and develop distributed dry deposition modeling framework by coupling with GIS. And, finally, using the developed system, potential urban forest planting locations are identified.
• #14: Here are objectives of this study. As a first step, I am gonna identify important input parameters for UFORE-D via sensitivity analyses. Then, identified important parameters are employed in a distributed form, and develop distributed dry deposition modeling framework by coupling with GIS. And, finally, using the developed system, potential urban forest planting locations are identified.
• #15: Here are objectives of this study. As a first step, I am gonna identify important input parameters for UFORE-D via sensitivity analyses. Then, identified important parameters are employed in a distributed form, and develop distributed dry deposition modeling framework by coupling with GIS. And, finally, using the developed system, potential urban forest planting locations are identified.
• #17: This is how UFORE-D works. Dry deposition is described by resistances to air pollutant transport in three phases. Resistance to transport from atmosphere to the immediate surroundings of tree surfaces. This is called aerodynamic resistance, Ra. When you zoom in a leaf, there is a very thin layer around the surface where the effects of turbulence is minimul. The second phase is is resistence to transport through this layer. It is called quasi-laminar boundary layer resistance, Rb. And, when you further zoom in on a leaf, you can see these elements in a leaf, Cuticular, which is outer skin of a leaf, and mesophyll, inner part of a leaf, and stoma, a small opening to exchange gas by transpiration, aspiration, and photosynthesis. Third phase is resistance to transport through these elements.These three resistances are combined and called canopy resistance, Rc. Pollutant flux is a product of dry deposition velocity Vd and air pollutant concentration, C. Vd is a reciprocal of the sum of Ra, Rb, and Rc.
• #18: This is how UFORE-D works. Dry deposition is described by resistances to air pollutant transport in three phases. Resistance to transport from atmosphere to the immediate surroundings of tree surfaces. This is called aerodynamic resistance, Ra. When you zoom in a leaf, there is a very thin layer around the surface where the effects of turbulence is minimul. The second phase is is resistence to transport through this layer. It is called quasi-laminar boundary layer resistance, Rb. And, when you further zoom in on a leaf, you can see these elements in a leaf, Cuticular, which is outer skin of a leaf, and mesophyll, inner part of a leaf, and stoma, a small opening to exchange gas by transpiration, aspiration, and photosynthesis. Third phase is resistance to transport through these elements.These three resistances are combined and called canopy resistance, Rc. Pollutant flux is a product of dry deposition velocity Vd and air pollutant concentration, C. Vd is a reciprocal of the sum of Ra, Rb, and Rc.
• #19: This is how UFORE-D works. Dry deposition is described by resistances to air pollutant transport in three phases. Resistance to transport from atmosphere to the immediate surroundings of tree surfaces. This is called aerodynamic resistance, Ra. When you zoom in a leaf, there is a very thin layer around the surface where the effects of turbulence is minimul. The second phase is is resistence to transport through this layer. It is called quasi-laminar boundary layer resistance, Rb. And, when you further zoom in on a leaf, you can see these elements in a leaf, Cuticular, which is outer skin of a leaf, and mesophyll, inner part of a leaf, and stoma, a small opening to exchange gas by transpiration, aspiration, and photosynthesis. Third phase is resistance to transport through these elements.These three resistances are combined and called canopy resistance, Rc. Pollutant flux is a product of dry deposition velocity Vd and air pollutant concentration, C. Vd is a reciprocal of the sum of Ra, Rb, and Rc.
• #20: This is how UFORE-D works. Dry deposition is described by resistances to air pollutant transport in three phases. Resistance to transport from atmosphere to the immediate surroundings of tree surfaces. This is called aerodynamic resistance, Ra. When you zoom in a leaf, there is a very thin layer around the surface where the effects of turbulence is minimul. The second phase is is resistence to transport through this layer. It is called quasi-laminar boundary layer resistance, Rb. And, when you further zoom in on a leaf, you can see these elements in a leaf, Cuticular, which is outer skin of a leaf, and mesophyll, inner part of a leaf, and stoma, a small opening to exchange gas by transpiration, aspiration, and photosynthesis. Third phase is resistance to transport through these elements.These three resistances are combined and called canopy resistance, Rc. Pollutant flux is a product of dry deposition velocity Vd and air pollutant concentration, C. Vd is a reciprocal of the sum of Ra, Rb, and Rc.
• #21: Sensitivity analysis (SA) is the study of how the variations in input parameters impact the model output Output of UFORE-D is dry deposition velocity, Vd. And, Vd is a function of these resistances, and resistances are a function of temperaute, wind speed, LAI, photosynthetic active radiation, which is basically a visible portion of spectrum, pressure, and relative humidity. And, the mothods I used are Monte carlo with latin hypercube sampling and Morris one-at-atime
• #22: Sensitivity analysis (SA) is the study of how the variations in input parameters impact the model output Output of UFORE-D is dry deposition velocity, Vd. And, Vd is a function of these resistances, and resistances are a function of temperaute, wind speed, LAI, photosynthetic active radiation, which is basically a visible portion of spectrum, pressure, and relative humidity. And, the mothods I used are Monte carlo with latin hypercube sampling and Morris one-at-atime
• #23: Sensitivity analysis (SA) is the study of how the variations in input parameters impact the model output Output of UFORE-D is dry deposition velocity, Vd. And, Vd is a function of these resistances, and resistances are a function of temperaute, wind speed, LAI, photosynthetic active radiation, which is basically a visible portion of spectrum, pressure, and relative humidity. And, the mothods I used are Monte carlo with latin hypercube sampling and Morris one-at-atime
• #24: Sensitivity analysis (SA) is the study of how the variations in input parameters impact the model output Output of UFORE-D is dry deposition velocity, Vd. And, Vd is a function of these resistances, and resistances are a function of temperaute, wind speed, LAI, photosynthetic active radiation, which is basically a visible portion of spectrum, pressure, and relative humidity. And, the mothods I used are Monte carlo with latin hypercube sampling and Morris one-at-atime
• #25: Study area is the City of Baltimore, Maryland. Hourly meteorological data are taken from weather station at Baltimore Washington International Airport. And minimum, maximum, mean, and standard deviation of input parameters at noon in July, 2005 are used.
• #26: Monte Carlo analysis is based on multiple model runs with model parameters randomly selected according to probability density function of the parameters, and then the results are statistically analyzed. Latin Hypercube Sampling technique was employed here to generate parameter sets. Here is an example. The six input parameters for UFORE-D were assumed to have a normal distribution, and the distribution was truncated at the minimum and maximum values. Then, the range of parameter is divided into n equal probability segments. In this example n is equal to 8. So the distribution is divided into 8 segments of 1/8 probability. Midpoint of each segment is used to generate a parameter value. These values are saved in a matrix. The first column has eight possible values for the first parameter, X1. When all values are selected, values in each column are randomly shuffled, then each row is an input for one run of UFORE-D. For UFORE-D, I used n is equal 10,000. So UFORE-D was run 10,000 times.
• #27: In Morris one at a time method, four equidistant points are selected for the six input parameters of UFORE-D. The four points are minimum, maximum, and two intermediate values, and the interval is same and half delta. Then, parameter set is generated based on these values. For each parameter, X1 to X6, initial value is randomly selected from these four values. In the next run, only one parameter is randomly selected and its value is changed by + or – delta, and then again only one parameter is randomly selected and the values is changed, and it goes on until the end. Throughout the parameter sets, a given parameter can take exactly two values, and every parameter is changed exactly once during the seven runs. The elementary effect for Xi is calculated with this equation, which is ratio of changes in output and input. For UFORE-D, this procedure was repeated for 10 times, so totally 70 runs were performed. Then, mean and standard deviation of elementary effects are calculated. High mean indicates that parameter is influential to the model output. High standard deviation indicates the effect of that parameter is nonlinear, or interacted with other parameters
• #29: Here are scatterplots of Monte Carlo simulation for NO2. LAI showed near linear effect. In PAR, Pressure, Relative humidity charts, the output is pretty much scattered independent of input. But you can see small linear effect on PAR and relative humidity. Temperature is also scatterd pretty much, but it seems little bit decreased with the higher temperature. In wind speed chart, when the wind speed is very small Vd is very small. Wind speed primarily affects Ra and Rb. Simply wind is necessary to transport air pollutant. When there is no wind air pollutant is not transported in the air.
• #30: This is MOAT results. The means and the standard deviations of the elementary effect are plotted against each other. Large mean indicates a large impact of input on output, High standard deviation indicates nonlinearity or interaction effects. Temperature had the greatest mean and standard deviation, indicating its large impact on model output and the non-linear relationship between temperature and Vd for all pollutants. LAI had the second greatest effect on NO2 deposition. The effect of LAI is pretty small on O3 and SO2 depositions but the effect is linear because of low standard deviation. Wind speed has the 2 nd largest impact on O3 and SO2.
• #32: Now, this is a big picture of the distributed UFORE-D, which is coupled with ArcGIS. Temperature, LAI, and concentration in a distributed form will be inputted to the system. Other meteorological data are still lumped, meaning representative for the study area. And, these are outputed from the system in a distributed format. Temperature and LAI were identified as influential to the model output. Concentration directly impacts the flux Now, the issue is how to effectively couple UFORE-D and GIS.
• #33: The answer is component-based approarch. Usually environmental models are implemented as one big software. All functions are written in a software language and compiled, linked, tested, and deployed for use. If you modify one part, that could affect other part of codes and you need to test whole system again, and then compile, link. You need to repeat those processes. So that is not productive. With component-based approach, codes are divided into small components. This function 1 is implemented in component 1. Function 2 is component 2 and so on. Each component is independently developed, compiled, and tested. Then, these components are linked at run time to perform a complete model function. If something is wrong on this component, this can be just replaced with a new one. You don’t need to touch other components. This is more like a hardware system, like a PC. If you have a problem with a keyboard, or mouse, you can just replace with a new one while affecting no other hardware. So, the component-based technology enables plug-and-play approach. I used Microsoft’s Component Object Model (COM) standard.
• #34: Distributed UFORE-D was developed in two phases. First, original lumped UFORE-D written in SAS is completely rewriten with a component-based approach. I separated original SAS into user interface, data input/output, and core functions and made component for each. These COM components are registered in the Windows registry. And at run time, the user interface component gets a pointer to the COM components and the method can be called. As a whole, this system performs the exactly same functions as the original UFORE-D. The biggest advantage here is that these COM components can be pluged in to ArcGIS because ArcGIS supports COM standard.
• #35: Here is the next step. It’s a same thing. This menubar provides graphical user interface and is registered in Windows registry, and this can be imported to ArcGIS. These COM components are registered in the Windows registry, and referred from tools. Then tool calls a method of component.
• #36: This is a menubar. UFORE-D component is here, called from a tool. And, in addition to UFORE-D model, models to estimate input parameters in a distributed format were developed, which are Delta T regression model developed by Forest Service, LAI model to estimate LAI, and these road emiision model, Gaussian line source model, and Gaussian point source model.
• #37: This model esimates pollutant dispersion from facility stacks (point sources). This blue thing is a stack and emit air pollutant. The plume goes up first and bended by wind. And, it is assumed that time averaged concentration in y direction (crosswind direction) and z direction both follow a normal distribution. This is a receptor of the pollutant that has a coordinate, x, downwind distance, y, crosswind distance, and height. So the concentration at receptor is expressed by this equasion. To implement this 3D dispersion on a 2-dimensional raster grid, the hight of receptor is fixed at 1.5 meters, which is about human’s aspiration height. In this figure, suppose the wind is blowing like this. Source is here. Only cells located downwind will be receptors. Along the center line of the plume, the concentration tends to be higher, and decreases toward crosswind direction And, concentration decreses toward downwind direction and eventually becomes almost 0. So, I specified a buffer zone within which the concentration is larger than 0.
• #38: Emission model and Gaussian line source dispersion model estimate pollutant dispersion from vehicles. These four road types were used. First, emissions of the entire vehicle fleet for each road type were calculated with this emission model. Q is emission rate per unit length, VMT, vehicle miles of travel is a traffic volume, L is road segment length, and f is emission factor. And, this equation is applied to road segments to estimate dispersion from the line source. Road semgents are defined within a cell. For example in this cell, interstate highway is curved. But I assumed it a straight line with the same length, and it’s always perpendicular to the wind direction. Again, this equation represents 3-dimensional fields. So using the same 2 dimensional scheme as the previous slide, the dispersion is calculated for each cell.
• #40: Models were run for 160 hours in July and August in 2005. Here are some example of the results. Downtown areas have higher temperature, which represetns head island effect. Smaller LAI values were observed in the downtown Baltimore area. Suburban areas tend to have larger LAIs. This is a NO2 concentration map. It’s higher along highways and around facilities. Wind is blowing from west and NO2 dispersed to east. Spatial pattern of the aerodynamic and boundary layer resistances have the pattern as the temperature. In downtown area, the ground is warmed and convection occurs, increasing turbulence. The increase in turbulence accompanies smaller aerodynamic and boundary layer resistances. The canopy resistance reflected the spatial pattern of LAI. Canopy resistance is smaller for larger LAI. You added these three resistances up together and take an inverse, which is Vd.But it seems that the spatial pattern of Rc is dominant. Flux is a product of concentration and Vd. Along highways, you can see higher flux.
• #41: To generalize the results, average of cell values over 160 hours was calculated. As a major source of NO2 is automobiles, high NO2 concentrations were found along highways. High NO2 concentrations radiating from some facility stacks were also observed. These facilities emit more than 100 tons of NOx annually. This NO2 dispersion pattern corresponded to wind direction and speed in the analysis period. Figure 5.5 summarized wind patterns during the analysis period. Winds were mainly from the West and Northwest, and these winds drove NO2 dispersion to the east of facilities.
• #42: Maps were created for 160 hours. I took the average of NO2 concentration and deposition velocity over the 160 hours. The potential urban forest planting areas may be where the concentration tends to be high and deposition tends to be low. Here is an example. This map presents area where the concentration is larger than 90 th percentile. So this is the area where concentration is particularly high. And, this map presents area where the deposition velocity is less than 30 th percentile. And these maps were overlaid and logical AND was taken. Hotspots are found along highways and areas surrounding NO2 emitting facilities. It is expected the air quality in Baltimore can be improved by planting more trees in these areas.
• #50: The developed modeling framework is one integrated system with which all analyses can be performed. So it’s easier to use than EPA models, that are often command line based and require pre and post processors. Model performance and validation are limited. Temperature regression model was developed based on 7 weather stations in the area. But the the coefficient of determination (r2) is about 0.5. There is only one pollutant concentration monitoring station in the study area. Concentration is adjusted with the measured value. But there are uncertinties. Dry deposition is very hard to measure. There is no monitoring site in the study area. COM is not new. It was introduced in mid 1990’s. But it is still in a mainstream of the software development. It is now very well matured and widely used. MATLAB, Mathcad, Minifold GIS all support COM. In the future, model improvement and more distributed parameters are necessary for better estimating urban forest effects. Component-based approach allows efficient model updates and extensions. The system can be implemented as a client-server based through network, or web-based.
• #51: As the developed modeling framework is one integrated systemll analyses can be all performed. So it’s easier to use than EPA models, that are often command line based and you need pre and post processors. Model performance and validation are limited. Temperature regression model was developed based on 7 weather stations in the area. But the the coefficient of determination (r2) is about 0.5. In the study area, there is only one pollutant monitoring station. Concentration is adjusted with the measured value. But there are uncertinties. Dry deposition is very hard to measure. There is no monitoring site in the study area. COM is not new. It was introduced in mid 1990’s. But it is still in a mainstream of the software development. It is now very well matured and widely used. MATLAB, Mathcad, Minifold GIS all support COM. In the future, model improvement and more distributed parameters are necessary for better estimating urban forest effects. Component-based approach allows efficient model updates and extensions. The system can be implemented as a client-server based through network, or web-based.
• #52: As the developed modeling framework is one integrated systemll analyses can be all performed. So it’s easier to use than EPA models, that are often command line based and you need pre and post processors. Model performance and validation are limited. Temperature regression model was developed based on 7 weather stations in the area. But the the coefficient of determination (r2) is about 0.5. There is only one pollutant concentration monitoring station in the study area. Concentration is adjusted with the measured value. But there are uncertinties. Dry deposition is very hard to measure. There is no monitoring site in the study area. COM is not new. It was introduced in mid 1990’s. But it is still in a mainstream of the component-based software development. It is now very well matured and widely used. MATLAB, Mathcad, Minifold GIS all support COM. In the future, model improvement and more distributed parameters are necessary for better estimating urban forest effects. Component-based approach allows efficient model updates and extensions. The system can be implemented as a client-server based through network, or web-based.
• #53: As the developed modeling framework is one integrated systemll analyses can be all performed. So it’s easier to use than EPA models, that are often command line based and you need pre and post processors. Model performance and validation are limited. Temperature regression model was developed based on 7 weather stations in the area. But the the coefficient of determination (r2) is about 0.5. There is only one pollutant concentration monitoring station in the study area. Concentration is adjusted with the measured value. But there are uncertinties. Dry deposition is very hard to measure. There is no monitoring site in the study area. COM is not new. It was introduced in mid 1990’s. But it is still in a mainstream of the software development. It is now very well matured and widely used. MATLAB, Mathcad, Minifold GIS all support COM. In the future, model improvement and more distributed parameters are necessary for better estimating urban forest effects. Component-based approach allows efficient model updates and extensions. The system can be implemented as a client-server based through network, or web-based.
• #54: As the developed modeling framework is one integrated systemll analyses can be all performed. So it’s easier to use than EPA models, that are often command line based and you need pre and post processors. Model performance and validation are limited. Temperature regression model was developed based on 7 weather stations in the area. But the the coefficient of determination (r2) is about 0.5. There is only one pollutant concentration monitoring station in the study area. Concentration is adjusted with the measured value. But there are uncertinties. Dry deposition is very hard to measure. There is no monitoring site in the study area. COM is not new. It was introduced in mid 1990’s. But it is still in a mainstream of the software development. It is now very well matured and widely used. MATLAB, Mathcad, Minifold GIS all support COM. In the future, model improvement and more distributed parameters are necessary for better estimating urban forest effects. Component-based approach allows efficient model updates and extensions. The system can be implemented as a client-server based through network, or web-based.
• #55: Like Chuck said, I was basically a software engineer and an IT consultant, working in high-tech industry in Japan for 10+ years. I always wanted to do something to protect the environment. So I decided to change my career and came here in 2003. At first, I had no idea about hydrology, or meteorology, or GIS. I took classes from facalties listed here and I have learned a lot. Now I am pretty much confident that my background could be applied to make a difference in this environmental field. And, I’d like to thank Dr. Dave Nowak, director in the US Forest Service. I have been working for him for the last two years, which made today’s presentation possible.
• #57: To see linearity and threshold more clearly, another Monte-Carlo simulation was performed. In this case, only one parameter value was varied and other parameters remained the same at their mean. The effect of LAI is almost linear, but around 6 the slope get little lower. PAR showed that Vd increased as PAR increased until about 400 W/m2, then further increase doesn’t change a lot. Temperature had a non-linear effect. Vd increased as temperature increased until about 25 °C and further increases in temperature caused decreases in Vd .
• #58: The analysis requires a certain conditions of input parameters to satisfy assumptions of models. To estimate dry deposition, the rain must be 0. Otherwise it is going to be a wet deposition, like acid rain. Wind speed needs to be larger than 0. Analysis is limited for daytime and unstable atmosphere that ensures the well mixed atmosphere and the concentration is almost constant vertically. Because I estimated concentration at 1.5 meter, and value can be used as an above canopy concentration in UFORE-D 160 hours in July and August in 2005 met these conditions in Baltimore.
• #59: This chart shows the frequency distribution of cell values in averaged NO2 concentration. It’s highly skewed. The most of the cell values are within 20 to 30 microgram/m3, and only few cells have larger concentraion. Here are some visual representations. These blue areas represent cells whose concentration is larger than 50 th , 70 th , and 90 th percentiles. In 50 th map, high concentration occurred in downtown areas and highways. It appears local roads have some effects at this level. In 70 th , local road are almost disappeared. 90 th represents areas with particularly high concentrations. This is a frequency distributon of cell values in averaged NO2 deposition velocity. This is also skewed, and there is a gap between relatively small Vd and a large Vd classes. This indicates that there are some areas where the dry deposition work very well. By planting more trees, urban forest managers want to shift these frequencies to the right. These maps show area with deposition velocity smaller than 10 th , 30 th , and 50 th percentiles. The potential planting locations are where the concentrations is high and deposition velocity is low. So, overlaying high concentration and low deposition velocity maps allow to identify such locations.