GEOG 584 Final

How does a local regression model differ from a regression model?

*Local regression model* uses each known point to derive a local model. Parameters: - *Spatial nonstationarity* - relationship between variables varies over space. - *Stationarity* - relationship between variables remains the same over space. - Example: Geographically weighted regression (GWR) analysis *Regression model* relates a dependent variable to a number of independent variables - *3 types*: linear, local, and logistic

Define location-allocation analysis.

*Location-allocation* - the process of determining the most efficient spatial configuration of supply and demand. - supply may be facilities (warehouses) - demand may be consumers (people)

What does loose coupling mean in the context of linking a GIS to another software package?

*Loose coupling* - input/output is not fully integrated. Example: - GIS preprocess data for the model. - GIS data is input to model (may be on another computer) - Model computes the results, that can be returned to GIS for display.

In many instances, you can build a GIS model that is either vector-based or raster-based. What general guidelines should you use in deciding which model to build?

*Main consideration*: discrete boundaries vs. continuously varying data. *Vector-based* - uses points, lines, and polygons - useful when data has discrete boundaries (borders) *Raster-based* - representation of a surface divided into grid of cells - useful when data continuously varies (satellite images)

Cost distance measure operations are based on the node-link cell representation. Explain the representation by using a diagram.

*Node* represents the center of a cell *Link* - connects node to adjacent cell (lateral or diagonal) *Distances* - calculated along links (lateral=1.0, diagonal=1.4142)

What are Nugget, Range, Sill, and Partial Sill?

*Nugget*: the semivariance at distance 0, representing measured error, microscale variation, or both. *Range*: distance where semivariance levels off (spatially correlated portion) *Sill*: semivariance where leveling takes place *Partial Sill*: difference between sill and nugget

Explain the main difference between ordinary kriging and universal kriging.

*Ordinary kriging* - assumes the absence of drift or trend, and focuses on the spatial correlation between sampled points. *Universal kriging* - assumes the presence drift (or trend) in the spatial variation z values, in addition to ordinary kriging.

Describe the three phases of the address geocoding process.

*Preprocessing*: - Parsing (breaking down an address into components) - Address standardization (codes together similar names spelled differently) *Address Matching* - matches address against the reference database *Plotting* - plots address as a point feature - (Linear) Interpolation - Location of Address

The root mean square (RMS) error is commonly used for selecting an optimal interpolation method. What does the RMS error measure?

*RMS* - quantifies the differences between the known and estimated values at sample points. *Standardized RMS* - requires the variance of the estimated values; therefore, it is only available for kriging

What are considered restrictions in network analysis?

*Restrictions* - refer to routing requirements on a network - one-way or closed streets. - can be represented using a binary code (1 = true, 0 = false).

Provide an example of a binary model from your discipline.

*Siting analysis* - determines if a unit area (polygon) meets a set of selection criteria. Example: mineral detection

What is a source raster for path analysis?

*Source raster* defines the source cell(s) from which the least-cost path from each cell is calculated. - source cells has a value (other cells = no data) - source cells are start/end point (origin or destination)

How does a static model differ from a dynamic model?

*Static* - models spatial data at a specific time (time is constant). *Dynamic* - models changes in spatial data over time (time is variable). - Ex: groundwater pollution

How does the surface distance differ from the regular (planimetric) cost distance?

*Surface distance* - the actual surface distance based on DEM. - measures the actual distance across cells. - surface distance changes with slope.

What are Thiessen polygons?

*Thiessen polygons* • polygons that define the closest area to each point • each polygon contains only one known point. • any location within the polygon is closer to its known point than to any other.

What is turn impedance and turn table?

*Turn impedance* time it takes to complete a turn. - significant in a congested street network. *Turn table* assigns the turn impedance value to each turn in the network (stop signs)

What is the validation? Explain how one can use the validation technique for comparing different interpolation methods.

*Validation* compares interpolation methods. It does this by dividing known points into two samples: - One sample to develop models for each method. - Another sample for testing accuracy - RMS statistic can be used to compare methods. Validation is not feasible if the number of known points is too small to be split into two samples.

Explain linear interpolation as a process for plotting an address along a street segment.

- Locates the street segment in the reference database. - Interpolates where the address falls within the address range.

What are applications of the least-cost path analysis?

- Pipelines - site analysis to find the least accumulative cost path - Modeling wildlife corridors

What factors can cause low hit rates in address geocoding?

- street name misspelled - address number incorrect - abbreviation not recognized - zip codes incorrect/missing - outdated reference database - incorrect reference database

What is a traveling salesman problem?

A routing problem that adds two constraints to the shortest path analysis: - The salesman must visit each of the select stops only once. - The salesman may start from any stop, but must return to the original stop (Tabu Search - best algorithm!)

Describe the basic steps involved in the modeling process.

*Modeling* -Define the goals of the model -Define the model -Implement and calibrate the model -Validate the model

What is the vector network model?

*vector network model* - system of linear features with attributes for flow. - topology-based - Lines cannot have gaps - Land lines have directions *Example*: road network with link impedance, turns, one-way streets..

Binning is a process for creating a usable semivariogram from empirical data. Describe how binning is performed.

Binning - sorts pairs of sample points (semivariance data) by distance and direction in each bin. *Two steps*: Binning by distance: it groups pairs of sample points into lag classes. Binning by direction: it groups pairs of sample points by direction.

Why is a binary model often considered an extension of data query?

Both Data query and Binary Model can request data from an attribute table through SQL.A Binary model does this with logic expressions 1 (true), 0 (false).

List and describe attributes in the TIGER/Line files that are important for geocoding.

FEDIRP - direction that precedes street name (NW) FENAME - street name FETYPE - street name type (St, Rd, Ln) FRADDL - beginning address # (left side of street) TOADDL - ending address # (left side of street) FRADDR - beginning address # (right side of street) TOADDR - ending address # right side of street) ZIPL - zip code (left side of a street) ZIPR - zip code (right side of street)

Name common mathematical models for fitting semivariograms.

Gaussian Linear Spherical Circular Exponential

What is photo geocoding?

Geotags photos to location information (Flickr)

How does the inverse distance weighted (IDW) interpolation work? Should be able to solve a simple problem

IDW interpolation is an exact method It estimates an unknown value from a set of known sample points that have been distance weighted. 1. Choose Points - Measured Points: known location and known value - Predicted points: known location (value unknown) 2. Define kernel function (k) weight 3. Set search radius - Fixed: specifies a constant neighborhood distance - Variable: specifies # of measured points Step 4: solve problem (formula) (See sample problem)

Understand least accumulative cost algorithm for path analysis.

Least accumulative cost path is an iterative process based on Dijkstra's algorithm. 1. activate cells adjacent to source cell (and compute costs) 2. select the cell with the lowest cost-distance (assign it output) 3. activate cells adjacent to next lowest (select cell with lowest cost) 4. iterate through steps 1-3 until each cell has its lowest accumulative cost

What is intersection matching?

Matches address data with street intersections on a map (corner matching) *Input*- two streets should be listed (E Sherman Ave & N 4th St) *Common problems*: - intersection may not exist - new streets not in reference database - winding street crossing another street more than once

What is ZIP code geocoding?

Reference database contains centroids of ZIP codes (not a street level geocoding)

Describe how the semivariance can be used to quantify the spatial dependence in a data set.

Semivariance increases as the distance between points increases. *If spatial dependence exists*: - known points close together should have small semivariances - known points farther apart should have larger semivariances *If spatial dependence has directional differences*, then semivariance may change faster in one direction.

What is the semivariance?

Semivariance is simply half the variance of the differences between all points spaced at a constant distance. - Kriging uses semivariance to calculate weights (to minimize error).

What is link impedance?

The cost of traversing a link (ex: 60 seconds)

What is spatial interpolation?

The process of using points with known values to estimate values at other points to create surface data for analysis and modeling.

Many index models use the weighted linear combination method to calculate the index value. Explain the steps one follows in using the weighted linear combination method.

To build an index model, the weighted linear combination method involves evaluation at three levels. 1. Criterion weights 2. Data standardization 3. Data aggregation *Criterion Weights* - They establish paired comparisons of a criterion. - expert-derived, or software (Expert Choice) *Data Standardization* - common method: linear transformation - May be standardized based on expertise and knowledge *Data Aggregation* - computed mathematically.

The power k in IDW interpolation determines the rate of change in values from the sample points. Can you think of a spatial phenomenon that should be interpolated with a k value of 2 or higher?

When the range of geographical dependence is short, and you want a lot more weight (or emphasis) on the nearest points. Example: cities near a delta.

Where does spatial interpolation work?

Where values are spatially dependent (or spatially auto-correlated) Where nearby locations tend to have similar Z values. Example: crime levels, precipitation

A semivariogram must be fitted with a mathematical model before it can be used in kriging. Why?

so it can be used to estimate the semivariance at any given distance.

What kinds of inputs are required for spatial interpolation?

(1) Control points and distribution (2) Type of spatial interpolation

Given a network with link impedance values, be able to solve the shortest path problem using Dijkstra's algorithm. Be able to describe step by step processes.

(See practice sheet)

Describe the two required inputs for address geocoding.

*Address table*: street addresses *Reference database*: street map and attributes from each street segment (street, address ranges, and zip codes)

The result of an allocation analysis is typically presented as service areas. Why?

*Allocation* models resource distribution through a spatial network. - It measures the efficiency of public facilities (fire stations) in terms of their service areas. - service areas are determined by accessible streets within a specified impedance, such as 5-minute travel time, or 3-mile distance limit.

What are model calibration and validation?

*Calibration* An iterative process that repeatedly compares the model output to the observed data, adjusts the parameters, and re-runs the model (fine tuning the model to some dataset) - Example: Deterministic model (sensitivity analysis) *Validation* Assesses the model's ability to predict under different conditions than those used in the calibration phase (goodness of fit) - Example: Validation technique (splitting data into subsets)

What is reverse geocoding?

*Converts* point locations (XY coordinates) into descriptive address. *Searches* for nearest address/intersection based on the specified search distance. *Returns no match* if there's no associated feature in the reference database.

Define a cost raster

*Cost Raster* - defines the cost or impedance to move through each cell. - may be sum or difference costs (construction cost, operational cost). - costs may be weighted

What is the cross validation?

*Cross Validation* - compares interpolation methods by repeating the following procedure for each method: - removes a known point from the data set - use the remaining points to estimate the value at the removed point - calculates the error of the estimation.

What are applications of dynamic segmentation? Examples?

*Data management* - used by transportation departments to manage speed limits. *Data display* - links routes to events (accident reporting). *Data query* - attribute/spatial data can be queried (planning and modeling). *Data analysis*- route and event overlays (demographics).

Describe the difference between a deductive model and an inductive model.

*Deductive model* represents conclusions from scientific theories or physical law. *Inductive model* represents conclusions from empirical data and observations.

What kinds of variables can be used in a logistic regression model?

*Dependent variables* - categorical *Independent variables* - categorical, numeric, or both

Describe the difference between a descriptive model and a prescriptive model.

*Descriptive model* - models existing conditions (vegetation map) *Prescriptive model* - predicts conditions (potential vegetation map)

Describe the difference between a deterministic model and a stochastic model.

*Deterministic model* - assumes no randomness (variables to have unique values) - Example: actual road related landslides *Stochastic model* (probabilistic) - assumes randomness (variables have probability distributions) - can have errors (kriging) - Example: probability of road related landslides

What is Dijkstra's algorithm?

*Dijkstra's algorithm* - common Iterative process to solve a single-source, shortest-path problem. Each step, the algorithm chooses the shortest path from a list of candidate paths and places the node of the shortest path in the solution list.

What is dynamic segmentation? What are events, routes, and linear referencing?

*Dynamic segmentation* - process of computing the location of events along a route. *Events* - linearly referenced data that occur along a route - *Point events*: point features (stop signs) - *Line events*: linear features (speed limits, pavement conditions) *Route* - linear feature with a linear measurement system stored with its geometry (street) -stores "measure" (distance) values along their length. *Linear referencing* - measuring system for linear features (river mile) - records location - allows multiple attributes - attributes can be displayed, queried, and analyzed.

How does an exact interpolation method differ from an inexact interpolation method?

*Exact* - predicts a value at the point location that is the SAME as its known value. *Inexact* - predicts a value at the point location that DIFFERS from its known value.

Explain the difference between a global method and a local method.

*Global* - uses EVERY known point to estimate an unknown value. *Local* - uses a SAMPLE of known points to estimate an unknown value.

Describe why the density of sampling points is important.

*High density* - captures local variation for large scale (small-areas) *Low density* - loses sensitivity of local variation and captures only regional variation for small scale (large-areas).

How does an index model differ from a binary model?

*Index model* - calculates the index value for each unit area (produces a ranked map) whereas, the *Binary Model* produces a simple yes (1) or no (0) for each unit area.

Describe how cell densities are derived using the kernel density estimation method.

*Kernel Density Estimation* (probability density function) - measures cell densities per unit area unit from point data. - expected density depend on the distribution of distances (kernel function k) between known and unknown points.

What is Kriging?

*Kriging* -assesses prediction quality with an estimated prediction error (stochastic). -assumes the spatial variation may consist of three components: 1. Spatially correlated 2. Drift or trend 3. Random error

Provide an example of an index model from your discipline.

*Landslides* -Identify locations that are at low, medium, or high risk. *Vector-based index model* 1. Criterion weights are set (low, medium, high) and maps are overlayed. 2. Data is standardized (increasing with risk level) 3. Data is aggregated (index values are calculate for each polygon)

What are four types of output grids resulting from cost distance measure operations? Describe each briefly.

*Least Accumulative Cost Grid* - calculates the least accumulative cost distance for each cell to the nearest source over a cost surface. *Directional Grid* shows the least cost path from each cell to source (uses directional codes 1-8) *Allocation Grid* shows the assignment of each cell to a source on the basis of cost distance measure. *Shortest Path Grid* shows the least cost path from each destination cell to a source.