134 | 4 Maps, Data Entry, Editing, and Output | Building a GIS Database | not introduction ‘Most spatial date, sources may bs catego- | : 5 se rized as either hardeopy or digital, Hardcopy | ! Spatial data entry and editing are fre- forms are any drawn, written, or printed doc- quent activities for many’ GIS users. A large ments, including hance-dtawn maps, mane | : number of coordinates is necded to represent iy menstned survey data, legal records, and | features in a GIS, and each coordinate value heasured Sury . i ‘d to the GIS database. Thi coordinate lists with associated tabular data. | rns be aaeerdealy don: wen with awe, ‘Mosthistorical spatial data were recofded on pated techniques, and spetal data entry and 408 Figure4-1) and although uo all maps a ques. P dare suitable for conversion to digital formats, editing take significant time for most orgeni- Chany maps are, Much data were creaiod zations. from hardcopy sources in the early years of GIS via digitizing, the process of collecting Figure 4-: Maps have served to store geographic knowledge for at least the past 4000 years. “This early map ‘of northem Europe shows approximate shapes and relative locetions.132 GIS Fundamentals Figare 4-2: An example of commonly produged digital maps. digital coordinates, Digitizing is a common data entry method today, although primarily from satellite and aerial images Digital maps, an electronic, graphic Gepiction of spatial data, ate by Zar the most ‘common map form today (Figure 4-2). Mil- lions of electronic maps are generated each hon, composed on demand in response to ‘web queries, on automobile navigations sys~ media ges tems, and for commerce and advertising These maps are flexible, easily customized, inexpensively distributed, and often dynamic. Most maps, whether digital or hardcopy, contain severat components (Figure 4-3), A data area ot pane occupies the largest part of fhe map, and contains most of the depicted spatial data. A neatline is ofien inciuded to provide a frame around all map elements, ‘and insets may contain additional map ele- ments. Scalebars, legends. titles, and other graphic elements such as a north arrow are often included, All maps have a map scale, defined as the ratio of the distance on the map to corresponding distance on the ground, (Figure 4-3). Maps often depictgoordinate lines (Fig- ure 4-4), When the lines represent constant Jatitude and longitude, e set of coordinate ines is called a graticule (Figure 4-4a), These lines may appear curved, depending, on the map scale, the map coordinate sys- ‘tem, and the location of the area on the Earth's surface. Maps may also depict a grid consisting of lines of constant coordinates. Grid lines are typically drawn in both the x reatline title 44 [> Recreational Lands of the Contiguous US. ‘date: pane -seale bam legend! Figure 4-3: An example of a map and its componenss. rae POweeps wee Chapter 4: Maps and Data Entry 133 sano ogre ean —t ie QUE TE MOE TE a sa t} S¢onem sons 00009 ‘ono sens stun Pseoucce sano 70009 sincone 0002 igure 4: Maps often dei lines representing a} a grticule of eons atime a Songinude or) and ¢) heve been magnified more than a to better show the effects of generalization, Each map has 2 ifferent ievel of map general:zation. Generatiza- tions increase with smuller-seale maps, and include mnissions of smaller lakes, successively greater road width exegeerations, and increasingly general- ized shorelines as one moves fom mepé a throug ried to the present. These errors are disap- eating as newer data are collected with digital methods, but will be encountered and should be derstood, Large-scale, high-quality maps gemer- ally cover smell areas. This is because of the ‘trade-off between scale and area coverage, and because of Limits on the practical size of a map. Carlometrie maps larger than a meter in any dimension have proven to be imprac- tical for most organizations. Maps above this size are expensive and difficult to print, store, or view. ‘Thns, hmman ergonomics set 2 practical limit on the physical size of a map. The fixed maximnm map dimension when coupled with afined map scale defines the area coverage of the bardeopy map. Chapter 4: Maps and Data Eniry 139 Larger scale maps-generally cover smatier areas, A 1:100,000-scale map that is 18 inches (47 centimeters) on a side spans approximately:28-miles (47 kilometers). A 1:000-scale map.that is:18 inches ona side represents 9:miles (15:kilometers):on. the Barth's:surfuce.: Because spatial date'ina GIS often span several large-scale maps, these map boundaries may occur in a spatial database. Problems often-arise when adja- ‘cent. maps.are‘enteréd into. a:spatial.datubase because features do:not align or have mis- matched:attibutes:across:map boundaries. Differences in the time of data collection. for adjacent map sheets may lead to incon- sistencies across-map borders, Landscape change through time-is.a major source-of dif440 GIS Fundamentals ferences across map boundaries. For exam- ple, the U.S. Geological Survey has produced 1:24,000-scale map sheets for all of the lower 48 United States of America. "The original mapping took place over sev- oral decades, and there were inevitable time Jags benween mapping some adjacent areas, As much as two decades passed between mapping or updating adjacent map sheets. ‘Thus, many features, such as roads, canals, ‘or municipal boundaries, are discontinuous or inconsistent across map sheets Different interpreters may also cause differences across map boundaries. Large- area mapping projects typically employ sev- * erat interpreters, each working on different map sheets for a region, All professional, iarge-urea mapping efforts should have pro- tocols specifying the scale, sources, equip- ment, methods, classification, keys, and cross-cortelation to ensure consistent map- ping across map sheet boundaries, In spite of these efforts, however, some differences due to human interpretation occur, Feature place- ment, category assignment, and generaiiza- tion vary among interpreters. These problems are compounded when extensive checking and guidelines are not enforced across map sheet boundaries, especially when adjacent areas are mapped at different times or by two different organizations. Finally, ditfexences in cooramate regis- ‘ration can lead to spatial mismatch across ‘map sheets. Registration, discussed later in this chapter, is the conversion of digitizer or other coordinate data to an earth-surface coordinate system. These registrations con- tain unavoidable errors that translate into spatial uncertainty, There may be mis- matches when data from two separate regis- trations are joined along the edge of a map. Spatial data stored in a-GIS are not bound by the same constraints that fimit the physical dimensions ofherdsopy maps. Dig- ital storage enables the production of seam- Jess digital maps of large areas. However, the inconsistencies that exist on hardcopy anaps may be transferred to the digital data. Inconsistencies at map sheet édges need to ‘be identified and resolved:-when maps are converted to digital formats, Digitizing: Coordinate Capture Digitizing is the process by which coot- dinates from a map, image, or other sources are converted into a digital format in a GIS. Points, lines, and areas on maps or images represent real-world entities or phenomeng, and these must be recorded in digital forms before they can be used:in a GIS. The coor dinate vatues sheit define the lovations and shapes of entities must be captured, that is, yecorded as numbers and structured in the spatial database, There is a wealth of spatial data in existing maps and photographs, and new imagery. and maps add to this source of information.on.a neatly:contiauous basis, Manual digitization is human-guided coordinate capture from a-map or image source. The operator guides an electronic device over a mup or image and signals the ‘capture of important coordinates, often by pressing a bution on the digitizing device. Iroportant point, line,-or-area'features are traced on the source materials, and the coor dinates are recorded in GIS-compatible for- mats. Valuable data on historical maps may be converted to.digital fortis through the use of manual digitizing. On-screen digitizing and hardeopy digitizing are:the two most common forms of manual digitization. On-screen Digitizing On-screen digetizing, also known as heatis-up digitizing, invotves manually-dligi- tizing on a computer screen, using adigital image as a backdrop. Digitizing software allows the operator to trace the points, lines, ‘or polygons that are identified onthe scanned map (Figure 4-11); Digitizing soft-ware allows the innman operator to specify the type of feature to be recorded, the extent and magnification of the image on screen, the mode of digitizing, and other options to control how data are input. The opetator typ- ically guides a cursor over points to be recorded using a mouse, and depresses « but- ton or sequence of buttons to collect the point coordinates, On-screen digitizing can be used for recording information from scanned acrial photographs, digital photo- graphs, satellite images, or other images On-soreen digitizing offers advantages over hardcopy and scat-digitizing, methods that are described in the following sections. Many data sources are inherently digital, for example, image data collected from aerial photographs and airborne or satellite scan- ners. These data may be magnified on screen to any desired scale, Converting the image to a paper or other hardcopy form would likely introduce error through the slight deforma- tion of the paper or printing media, reduce Chapter 4: Maps and Data Entry 144 flexibility when digitizing, and add the cost of printing ‘On-screen digitizing is ofien more accu- | rate than manual digitizing because manual map digitization is often iimited by the vyisual acuity and pointing ability of the oper- ator. The pointing imprecision of the opera- tor and digitizing systems translates to. fixed ground distance when manually digi- tizing 2 bardeopy map. For example, con- sider an operator that can reliably digitize a location to the nearest 0.4 millimeters (0.01 inch) on 4 1:20,000-scale map. Also assume the best hardcopy digitizing table available is being used, and we know the observed error is larger than the error in the map. The 6.4 millimetey etror in precision translates to approximately 8 meters of error on the Earth's surféce. The precision cannot'be appreciably‘improved when using a digitiz- ing table, becanse a majority of the impreci- sion is due to operator abilities. In contrast, once the map is scanned, the image may be displayed on a computer screen at any map ‘Figure 4-11: An example of on-screen digitizing. Images or maps ére displayed ori a computer screen and feature data digitized manually, Buildings, roads, or-eny-other features that may be distinguished-onrtie image may be digitized.442 GIS Fundamentals seaie, The operator may zoom to a 1:5,000- seaie or greater on-screen, and digitizing ‘accuracy and precision improved. While other factors affect the accuracy of the derived spatial data (for exemple map plot- ting or production errors, or scamnex accu- racy), on-screen digitizing may be used to limit operator-induced positional error when digitizing. On-screen digitizing also semoves or reduces the need for digitizing ‘ables or map scanners, the specialized equipment used for capturing coordinates from maps. Hardcopy Map Digitization Hardcopy digitizing is human-guided coordinate capture from a paper, plastic, ot ofner hardcopy map. An operator securely attaches @ map to a digitizing surface and traces lines or poiats with an electrically sen- sitized puck (Figure 4-12). The most com- ‘mon dipitizers are based on a wire grid erabedded in or under 2 table, Depressing 2 ‘button specifies the puck location relative to the digitizer coordinate system. Digitizing tabies can be quite accurate, with a resolu- tion of between 0.25 and 0.025 millimeters (0.01 and 0.001 inches). Whils once a major method for captur- ing spatial data, haxdcopy mop digitizing is diminishing in importance as most paper documents have been converted to digital forms. The tables are large, somewhat expensive, and now little-used. However, because data from hardcopy sources ere likely to persist for many decades, and there are still many specialized documents to con- vert, you should be familiar with the process. Not all maps are appropriate as a source of information for GIS. The type of map, how it was produced, ad the intended pur- pose must be considered when interpreting the information on maps. Only cartomettic maps should be directly digitized, and even though cartometric, a map may not be suit- able, Consider the dot-density map Figure 4<12:Manval digitizing on.a digitizing tabie,described in Figure 4-2, Population is depicted by points, but the points are plotted with random offsets or using some method that does not reflect the exact location of the population within each polygon, Before the information in the dot density map is entered into a GIS, the map should be interpreted correctiy. The number of dots in a polygon should be counted, this number snultiplied by the population per dot, and the population value assigned to the entire polygon. Maps may be unsuitable for digitizing due to the media. Most hardcopy maps are on paper because itis ubiquitous, inexpen- sive, and easily printed, Creases, folds, and ‘wrinkles can Jead 10 non-uniform deforma tion of paper maps. Characteristics of Manual Digitiz- ing Manual digitizing, whether from @ digi- tal image on screen or from a hardeopy source, is common because it provides suffi- ciently accurate data for many, if not most, applications. Manual digitizing may be at Jeast the accuracy of most maps or images, so the equipment, if properly used, does not add substantial error. Manual digitizing also requires low equipment investment, often just the software for image display and coor- dinate capture. The human ability to inter- pret images or hardcopy maps in poor condition is a unique and important benefit of manual digitizing. Humans are usually better than machines at interpreting the ‘information contained on faded, stained, or poor quality maps and images. Finally, man- ‘ual digitizing is often best because short ‘taining periods are required, data quality may be frequently evaluated, and digitizing equipment is commonly available. For these reasons manual digitization is likely to remain an important data entry method for some time to come. ‘There are « number of characteristics of manual digitization that may negatively affect the positional quality of spatial data. As noted earlier, map or image scale and ‘olution impacts the spatial accuracy of digi- Chapter 4: Maps and Data Entry 143 tized data. This scale may be the production scale for hardcopy maps, or the display scale for digital images or scanned maps. Table 4 1 illustrates the effects of map scale on:data quality. Errors of one millimeter (0.039 inches) o1 # 1:24,000-svale map correspond to 24 meters (79 feet} on the surface of the Earth, This same one millimeter error on a 1;1,000,000-scale map corresponds to 1000 meters (3281 feet) on the Earth's surface. Thus, small errors in map production or interpretation may cause significant posi- tional errors when scaled to distances on the Earth, and these errors are greater for smaller-soale maps. Errors due to human pointing ability are reduced for on-screen digitizing, because the operator can zoom in to larger scales'as needed. However, thi does not over¢ome errors inherent in original images or scahned documents. Both device precision and map scaies should be considered when selecting a digi- tizing tablet. Map scale and repeatability both set an upper limit on the positional quality of digitized data, The anost precise digitizers may be required when attempting to meet e stringent error standard while digi- tizing small-scale maps, ‘Table 4-1: ‘The sutiace error causedt-by a one millimeter (0.039 inch)-mep error will change’as.map soale-changes. Note the larger error at smaller map scales, Mop Seale | Error | Error (m) (Ft) 124,900 24 19 1:50,000 50 164 1:62,500 63 205 1109,000 | 100 328 1250000 | 250 820 1,000,000 | 1,000 | 3,281144 GIS Fundamentals 16 tte pints fit probity eisebution te frequency a al _ 04-02. 00 02 64 error in x direction (mn) Figure 4-13: Digitizing error, defined by repeat digitizing. Points repestedly digitized cluster around the true location, and Zollow a norraal probability distribution. (Som Bolstad et al, 1999), The abilities and attitnde of the person digitizing (the operator) may also affect the geometric quality of manuelly digitized date. Operators vary in their visual acuity, steadi- ness of hand, attention to detail, and ability to concentrate. Some operators will more accurately capture the coordinate informa- tion contained in maps. The abilities of any single operator will also vary through time, due to fatigue or difficulty maintaining focus on a repetitive task. Operators should take frequent breaks from digitizing, and compar- sons among operators and quality and con- sistency checks should be integrated into any manual digitization process to ensure accu- tate and consistent data collection. ‘The combined errors from both opera- tors and equipment have been well-charac- terized and may be quite small. One test using a high-precision digitizing table revealed digitizing errors averaging approxi- mately 0.067 millimeters (Figure 4-13), Enrors followed a random normal distribu- tion, and varied significantly among-opera~ tors, These average errors translated to an approximately 1.6 sieter error when scaled from the 1:24,000 map to a ground-equiva- lent distance. This average error is Jess than the acceptable production error for the map, and is suitable for many spatial analyses. The Digitizing Process Manual digitizing involves displaying a digital image on screen or placing a map on a digitizing surface, and tracing the location. of feature boundaries. Coordinate data are sampled by manually positioning the puck or cursor over each target point and collecting coordinate locations, This position/coliect step is repeated for every point to be cap- tured, and in this manner the locations and shapes of alt required map features are defined. Features that are viewed as points are represented by digitizing a single loce- tion. Lines are represented by digitizing an ordered set of points, and polygons by digi- tizing e connected set of lines, Lines have a starting point, ofien called a starting node, a set of vertices defining ‘the line shape, and an ending node (Figure 4-14). Hence, lines may be viewed as a series of straight line seg- ments connecting vertices and nodes. Digitizing may be in point mode, where the operator must depress a button or other- wise signal to the computer to sample each point, or in stream mode, where points are node. \ vertex NX, Figure 4-14: Nodes define the starting and ending points of lines. Vertices define line shape.automatically sampled at a fixed time or dis- tance frequency, perhaps once each metex. Strearn mode helps when large numbers of lines ate digitized, because vertices may be sampled more quickly and the operator may hecome leas fatigued. ‘The stream sampling rate must be speci- fied with care to avoid over- or under-sam- pied lines. Too short a cotlection interval results in redundant points not needed to accurately represent line or polygon shape. ‘Too jong a collection interval may result in the Joss of important spatial detail. In addi- tion, when using time-triggered stream digi- tizing, the operator must remember to continuously move the digitizing puck; if the operator rests the digitizing puck for a period. fonger than the sampling interval there will be multiple points clustered together. These will redundantly represent a portion of the line and may resuit in overlapping segments, Pausing for an extended period of time often creates a “rat’s nest” of lines that must later be removed. Minimum distance digitizing is a variant of stream mode digitizing that avoids some of the problems inherent with tine-sampled streaming. In minimum distance digitizing a new point is not recorded unless it is more than some minimum threshold distance from the previously digitized point. The operator Figure 4-15: Common digitiz- ing errors. undershoot danglin Chapter 4; Maps and Data-Entry 145 may pause without cresting 2 rat's nest of line segments. The threshold must be-chosen carefully - neither too large, missing useful detail, nor too small, in effect reverting back to stream digitizing. Digitizing Errors, Node and Line Snapping Positional errors are inevitable when data are manually digitized. These errors may be “small” relative io the intended vse of the data, for example the positional errors may be less than 2 meters when only $ meter accuracy is required, However, these rela- tively small errors may still prevent the gen+ eration of correct networks or polygons. For example, 2 dafa layer representing a river system may yot be correct because major tributaries may not connect. Polygon fea- tures may not be correeily defined because their boundaries may not completely close. ‘These small errors must be removed or avoided during digitizing. Figure 4-15 shows some common digitizing errors. Undershoots 2nd overshoots are com- mon errors that occur when digitizing, ‘Undershoots are nodes that do not quite reach the fine or another node, and over- shoots are lines that cross over existing nodes or lines (Figure 4-15), Undershoots overshoot missing label point line crossin: ~Sn0 node446 GIS Fundamentals cause unconnecied networks and unclosed polygons, Overshoots typicaily do not cause problems when defining polygons, but they ‘may cause djficulties when defining and analyzing line networks Node snapping and line snapping are used to reduce undershoots and overshoots while digitizing. Snapping is 2 process of automatically setting nearby points to bave the same coordinates, Snapping relies on a snap tolerance ot snap distance. This dis- tance mey be interpreted as a minimum dis- tance between features. Nodes or vertices cioser than this distance are moved to occupy the same location (Figure 4-16). ‘Node snapping prevents a new node from being placed within the snap distance of an already existing node; instead, the new node is joined or “snapped” to the existing node. Remember that nodes are used to define the ending points of a line. By snapping two nodes together, we ensure a connection between digitized lines. Line snapping may ulso be specified. Line snapping inserts a node at a line eross- ing and clips the end when a small overshoot is digitized. Line snapping forces a node to connect to a nearby line while digitizing, but only wher the undershoot or overshoot is Jess than the snapping distance. Line snap- ping requires the calculation of an interseo- tion point on an already existing line. The snap process places a new node at the inter- section point, and connects the digitized line to the existing line at the intersection point. This splits the existing line into twa new lines. When used properly, tine and node snapping réduce the number of uadershoots and overshoots. Closed polygons or inter- secting lines are easier to digitize accurately and efficiently when node and line snapping are in force. ‘The snap distance must be carefully selecied for snapping t be effective. If the snap distance is too silort, then snapping has little impact. Consider a system where the operator may digitize with better than 5 meter accuracy only 10% of the'time. This means 90% of the digitized points will be more than 5 meters from the intended loca- tion. If the snap tolerance is set to the equiv- alent of 0.1 meters, then very few nodes will before snapping: after snapping: Figur 4-16: Urdrshoos, verso, aud snapping Sapping may jin nodes, or may pase anode nic anear segment. Snapping does not occur if the nodes and/or lines are separated by more than thebe within the snap tolerance, and snapping has little effect. Another problem comes from setting the snap tolerance too large. If the snap tolerance in our previous example is set to 10 meters, and we want the data accu- rate to the nearest § meters, then we may jose significant spatial information that is contained in the hardcopy map. Lines less than 10!meters apart cannot be digitized as separate objects. Many features may not be represented in tie digital data layer. The snap distance should be smaller than the desired positional socuracy, such that signif icant detail contained in the digitized map is recorded. It is also important that the snap distance is not beiow the capabilities of the system used for digitizing. Careful selection of the snap distance may reduce digitizing exrors and significantly reduce time required for later editing. Reshaping: Line Smoothing and Thinning Digitizing software may provide toois to smooth, densify, or thin points while enter- ing data, One common technique uses spline functions to smoothly interpolate curves between digitized points avd thereby both smooth and densify the set of vertices used to represent a line. A spline is set of polyno~ ial functions that join smoothly (Figure 4- 17), Polynomial functions are fit to sticces- sive sets of points along the vertices in a line; for example, a function may be fit to points 1 through 5, and a separate polyno- mial function fit to points 5 through 11 (Fig- ure 4-17), Constraints force these fanctions to connect smoothly, usually by requiring the first and second derivatives of the func- tions to be continuous at the intersection point. This means the lines have the same slope at the intersection point, and the slope is changing at the same rate for both lines at the intersection point, Onee the spline fune- tions are calculated they may be used to add vertices, For example, several new vertices may be automatically placed on the line between digitized vertices 8 and 9, leading to the “smooth” curve shown in Figure 4-17. Chapter 4: Maps and Data Entry 147, Data may also be digitized with too many vertices. High densities may occur when data are manually digitized in stream mode, and the operator moves slowly rele- dive to the time interval. High vertex dens)- tics may also bc found when data are derived from spkine or smoothing fimetions that specify too high a point density, Finally, automated seanning and then raster-io-veo- tor conversion may result in coordinate pairs spaced at absurdly high densities. Many of these coordinate data are redundant and may be removed without sacrificing spatial acev- racy. Too many vertices may be a problem in that they slow processing, although this has become less inmportant as computing. power has increased. Point thinning algoritiums inave been developed to reduce the number of points whilg maintaining the line shape. Many poitt thinning methods use a per- pendicular “weed” distance, measured from a spanning line, to identify reduadant points (Figure 4-18, top). The Lang method exem- plifies this approach. A spanning line con- neots two non-adjacent vertices in a line. A one polynomial equation 1s fit to points one through Five, y = f(x) Seoiatnes points ste, join point ¢ oy Fis eps of 9 emer ep anes polyromiai equation second is fit to points $i through eleven, y = o(%) Figure 4-17: Spine interpolation to smooth. digitized lines.148 GIS Fundamentals pre-determined number of vertices is spanned initially, The initial spanning mum- ber has been set to 4 in Figure 4-18, meaning four points will be considered at each start ing point, Areas closer than the weed dis- tance are chown in gray in the figure. A straight line is drawn between 2 starting point and an endpoint that is the 4th point down the line (Figure 4-180). Any interme- diate points that are closer than the weed dis- tence are marked for removal. In Figure + 18a, no points are within the weed distance, therefore none are marked. The endpoint is ‘then moved to the next closest remaining ‘point (Figure 4-18b), and all intermediate. points tested for removal. Again, aay points cioser than the weed distance are marked for removal. Note that in Figure 4-18b, one point is within the weed distance, and is removed. Once all points in the initial span- ing distance are checked, the last rerpaining, endpoint becomes the new starting pot, ‘end a new spamming line drawn to connect’4 points (Figure 4-18c, d). act ‘The process may be repeated for succes- sive sets of points in a line segment until all vertices have been evaluated (Figure 4-18e toh). All close vertices are viewed as'not recording a significant change in the line shape, and hence are expendahle. Increasing ihe weed distance thins more vertices, and at some upper weed distance teo many vettices may be removed. A balance must be struck between the removal of redundant vertices and the loss of shape-defining points, ust- ally through a careful set of test cases with successively larger weed ‘distances. There are many variants on this basic concept, Some look only at three immedi- ately adjacent points, testing the middle point against the line spanned by its two neighboring points. Others constrain or expand the search baged on the complexity of the line. Rather than always looking at four points, as in our example above, more points are scrutinized when the line is not complex (nearly straight), and fewer when the Line is complex (many changes in direc- tion). f) g. ti hy Figure 4-18:The Lang algorithm is a common iine-thinning method. In the Lang method vertices are removed, or thinned, when they are within a weed distauce to a spanning line (adapted drom Weibel, 1997).chapter 4: Maps and peta Entry 448 scan Digitizing arrays are ypically sed to measuTe ihe / . seeciance 50 taal NE Several rows of Optical searing © another method For, oes MEN ‘ye scanned tancousl: A . converting ‘earacopy documents into digits! motos ten moves {be optical trait. 19 the iacent ines and “process is sepented. coats rigs S19). scanners nave Ae me © at sete Liga. Most S25 vers pass 2 Sensi jernent 0 annets flathed ea jiaminated B05 This devi sures both ners in that they employ 2 rolaline indet. ‘the prec Jocat point being 82 ‘A map 15 f% yee sul vf this cylin and the serength of the fight reflect xd OF der, and the cylinder set 80 aia Unie fom tat pout. digs ON -yelocity. iar velocity jptensities are sensed and ‘converted 10 RUE otating oy lin to control than THe bers rort-ime motion OFS "pod acaiet, 8° many of tne Cath pgh-precision. Scanner ‘a threshold is often applied to deter pe gensed poss 18 PAE cof a feature vse oo scam es Foran ac consist ot 4 Wereetion of senectes i to conse BED C1 mgt be set uch CA piess than, 10% OFTHE mess ight striking ghe map t ‘erurned 10 the seme Scannery york best when very cleat ar, te sensed POHL ‘considered past oF ave available even the most expensi— ine, Lf 10% oF f che enerey 18 sonnets y report 2 senior wamiber of fie Rensor, he point 38 CO" ‘spurious fines OF PI? , 4 Tepered part of the WHE spac folded, oF wrk spre ged: THESE 3 ines. ThE then Pr raster spurious features NS Ipsecurently o “yalues at© removed ¥i8 editing, tus neEAtDE yecorded where points of a st On the spec’ gavantege of over mate and null recorded it vat digitizing ing also works best the intervenunS when yyaiiable aS ‘map se? . . with one them: ic Feature each map: Most Sa asi fe paditing takes less gine when 1085 guriace contain writing O° ofher strongly : oe ihe ee js placed. AR or oe couse colossare prefered, auch a : core ana sevnring Ne ap vo tbe bed: OF oo ae ona TT Foacksground tate ‘ene bed seETS rr toni is passed hen uk we OT . Finally ee nome Pap, nit et sans te singe most % Rrantageous when 41a ove este tak oat 8 e ut g mart gsapiie eemens is found on | ‘the maps. scan digitization usualy vequires some rpg, oF ine BABS, PA ‘data are to be converted 10-8 formal yines are fies. ingle pix) (Figue 4-20). On€ eral pixels 3% ‘pe selected 10 8] fai ‘on of tie Sine: ag true For points. ‘a pine 388 oof the Tne is oypically 0, ofa finedefined she pixel nearest ool pet150 GIS Fundamentals Before thinning After thinning Figure 4 ~ Editing Geographic Data Spatial data may be edited, or changed, for several reasons. Errors and inconsisten- cigs are inevitably introduced during spatial data entry. Undersboots, overshoots, missing or exira lines, missing or extra points or Jabels are all errors that must be corrected. Spatial data ean change over time. Parcels are subdivided, roads extended or moved, forests grow or are cut, and these changes may be entered in the spatial database through editing. New technologies may be developed that provide more aceurate posi- tional information, and even though existing data may be consistent and current, the more accurate data may be more useful, leading to data editing, Identifying exrors is the first step in edit- ing, Errors may be identified by printing a map of the digitized data and verifying that each point, line, and area feature is present and correctly located. Plots are often printed both at a similar scale and at a significantly larger scaie than the original sowrce materi- als. The large-scate plots are often paneled with some overlap among panels, Plots at scale are helpful for identifying missing fea tures, and large-scale plots aid in identifying undershoots, overshoots, and smal! omis- sions or additions. Operators typically auno- tate these plots as they are checked systematically for each feature, ): Skeletonizing, a form of line thinning thet is oftea applied after soan-digitzing. Software help operators identify poten- tial errors. Line features typically begin and end with a node, and nddes may be classified as connecting or dangling. A connecting node joins two or more lines, while a dan- gling node is attached to only one line. Some dangiing nodes may be intentional, for example, 4 cul-de-sac in a street network, while others will be the result of under- or overshoots, Dangling nodes that are plotted with unique symbols can be quickly evalu- aied, and if appropriate, corrected. Attribute consistency may also be used io identify errors. Operators note areas in which contradictory theme types occur in different data layers. The two layers are either graphically-or cartographieally over- lain, Contradictory co-ocourrences are iden- tified, such as water in one layer and upland areas in a second, These contradictions are ‘then either resolved manually, or automati- cally via some’pre-défined precedence hier- archy. Manly GIS software packages provide a comprehensive set of editing tools (Figure 4- 21), Editing typically inchudes the ability to select, split, update, and:add features. Selec- ‘tion may-bé based:on:geomettic.attributes, or with:a.cursor guiided-by:the eperator::Selec- tions may be-made individually, by geo- gtaphic:extent (select.all features in-a box, circle, or within a certain distance of the pointer) or by. geometric attributes (e.g., select all nodes that connect to-only-onei , a & Bs Chapter 4: Maps and Date Entry 454 Create New Feaiure Create Link Lit Adjustinent [Auto Complete Polygon Select Features Using a Line Select Faaiures Using an Area Extend/Tsim Features Miror Feetutes Mody Feature Reshape Feature Cut Poiycon Features Kigure 4-21: GIS software provide for a exible and complete. set of editing tools. These tools provide for the rapid, precice, controlled creation and modification of coordinates and attributes of spatial dats (courtesy ESRD. line), Once a feature is selected, various operations may be available, including eras- ing all or part of the feature, changing the coordinate values defining the feature, and inthe ease of lines, splitting or adding to the feature, A line may be split into parts, either to isolate a segment for future deletion, or to modify only a portion of the line. Coordi- nates are typically altered by interactively selecting and dragging points, nodes o: ver tices to their best shape and location, Points or line segments are added as needed. Groups of features in an area may be adjusted through interactive rubbersheeting, Rubbersheeting involves fitting a local equa- tion to adjust the coordinates of features. Polynomial equations are often used due to their flexibility and ease of application. Anchor points are selected, again on the graphies screen, and other points are selected by dragging interactively on the soreen to match point locations. Ail fines-and points except the-anchor points are interac- tively adjusted. One common application of rubbersheeting involves adjusting Finework representing cultural features, such as a toad network, when higher geometric-accuracy photo or satellite image data are available. ‘The linework is overlain on an image back- drop and subsequently adjusted. Ail edits should be made with due atten- tion to the magnitude of positional change introduced during editing. On-screen edit- ing to-eliminate:undershoots should-only-be performed when the “trué” locations of fea- tures may be identified accnrately,-and-the new features can be confidently placed in the correct location. Automatic removal of “short” undershoots may'be performed with- out introducing additional spatial error-in152 GIS Fundamentals ‘most instances. A short distance for an undershoor is subjectively defined, but typi- cally it is below the error inherent in the source map, or at least a distance that is insignificant when, considering the intended use of the spatial date. Features Common to Several Layers ‘One common problem in digitizing derives from representation of features that occur on different maps or images. These features serely kave identical locations on each map or image, and often occur in dif- ferent locations when digitized into their respective data layers (Figure 4-22). Fot sample, water boundaries on soil survey ‘maps rarcly correspond exactly te water boundaries found on USGS topographic maps. Features may appear differently on dif- ferent maps fo: many reasons. Pethaps the maps were made for different pinposes or at different times, Features may differ because the maps were from different source materi- als, for example, one map may have been border from state map border from county map Figure 4-22; Common features may be spa- tielly incoasistent in different spatial data lay er, based on ground surveys while another was based on aerial photographs. Digitizing.can also compound fhe problem due to differ- ences in digitizing methods or operators. There are several ways to remove this “common feature” inconsistency, One involves se-drafting the data from conflict- ing sources onto one base nap. Inconsisten- cies are removed at the drafting stage. For exampie, vegetation and roads data may show vegetation type boundaries at road eciges that are inconsistent with the road jocations. Both of these data layers may be drafted onto the same base, and the common boundaries fixed by a single line. This line is digitized once, and used to specify the locs- tion of both the toad and vegetation bound- ary when digitizing. Re-drafting, although labor intensive and tigre consuming, forces 2 resolution of inconsistent boundary loca- fions. Re-drafting also allows several maps to be combined into a single data layer, A second, often preferable method involves establishing a “master” boundary which is the highest accuracy composite of the availabie date sets. A digital copy or overlay operation establishes the common features as a base in all the-data layers, and this base may be used as each new layer is produced. For exampie, water boundaries might be extracted from:the soil survey and ‘USGS. quad maps andthese:data combined ina third data layer, The:third-data layer would be edited to produce a composite, high-quality water layer. The composite water layer would then be copied back into both the soils-and USGS.qnad layers. This second approach, while resulting in visually consistent spatial data layers, isan many instances only a'cosmetic improvement of the data. If there are large discrepancies (large” is defined relative to the required spatial data. accuracy), then the source of the discrepancies should be identified.and the ‘most accurate data used, ornew, higher accuracy data collected from:the field or original sources.Chapter 4: Maps and Data Entry 153 Coordinate Transformation Coordinate wansformation is a common operation in the development of spatial data for GIS. A coordinate transformation brings spatia! data into an Earth-based map coordi- ‘nate sysiem so that each data layer aligns with every other data layer. This alignment ensures features fall in their proper relative position when digital data from different lay- ers are combined. Within the limits of data acouracy, a good transformation helps avoid inconsistent spatial relationships such as farm fields on freeways, roads under water, or cities in the middle of swamps, except where these truly exist. Coordinate transfor- mation is also referred to as registration, ‘pecause it “registers” the layers to a map coordinate system. Coordinate transformation is most com- monly used to convert newly digitized data from the digitizer/scanner coordinate system toa standard map eoordinate system (Figure 4.23). The input coordinate system is usu- ally based on the digitizer or scamner- assigned values. An image may be scanned and coordinates recorded as a cursor is moved across the image surface. These coo Source : coordinate system = dinates are usually recorded in pixel, inch, or centimeter units relative to an origin located near the lower left comer of the image. The absoiute values of the coordinates depend on where the image happened to be placed on the table prior to scanning, but the relative position of digitized points does not change. Before these newly digitized data may be used with other data, these “inch-space” or “digitizer” coordinates must be transformed into an Earth-based map coordinate system, Control Points A set of contro! points is used to trans- form the digifized data from the digitizer or photo coordinate system to a map-projected. coordinate system. Control points are differ- ent from other digitized features. When we digitize most points, lines, or areas, we do not know the map projection coordinates for these features. We simply collect the digi- tizer x and y coordinates that are established with reference to some arbitrary origin on the digitizing tablet or photo. Control points differ from other digitized points in that we Target coondinate-system ‘Figure 4-23: Como points in a coordinate treusformation, Control points are used to guide the trans- formation of a source, input-set of coordinases to a tasget, ourput set of coordinates, Thore:are five-con- ‘to! points in this example, Corresponding postions are shower in both coardinate systems.454 GIS Fundamentals know both the map projection coordinates and the digitizer coordinates for these points. ‘These two sets of coordinates for each control point, one far the map projection and one for the digitizer system, are used to esti- mate the coefficients for transformation equations, usually fazough a statistical, least- squares process, The transformation eque- tions are then used to convert coordinates from the digitizer system to the map projec- tion system, The transformation may be estimated in the initial digitizing steps, and applied as the coordinates are digitized from the map or image. This “onethe-fiy” transformation aliows data to be output and analyzed with reference to map-projected. coordinates. A previously registered data layer or image ‘may be displayed on soreen just prior to dig- itizing a new map. Control points may then be entered, the new map attached to the digi= tizing table, and the map registered. The new data may then be displayed on top of the pre~ viously registered data. This allows a quick check on the location of the newly digitized objects against corresponding objects in the study area, In contrast to on-the-fly transformations, data can also be recorded in digitizer coordi- nates and the transformation applied later. All data are digitized, including the coutrol point locations, The digitizer coordinates of the control point may then be matched to corresponding map projection coordinates, and transformation equations estimated. ‘These transformation equations are then applied to convert all digitized data to map projection coordinates, Control points should meet or exceed several criteria, First, control points should be from a source that provides the highest feasible coordinate accuracy. Second, con- trol point accuracy should be at least as good as the desired overall positional accuracy requited for the spatial data. Third, control points should be as evenly distributed as possible throughout the data area. A suffi- Cient number of control points should be col- ected, The minimum number of points depends on the mathematical form of the transformation, but additional contro! points above the minimum number are usually col- lected: this usually improves the quality and aceuracy of the statistically-fit transforma- tion functions. The x, y (horizontal), and sometimes z (vertical or elevation) coordinates of control points are known to a high degree of accu- racy and precision, Because high precision and accuracy are subjectively defined, there are many methods to determine control point locations, Sub-centimeter accuracy may be required for control points used in property boundary layers, while accuracies of a few meters may be acceptable for large-area veg- ctation mapping. Common sources of con- trol point coordinates are’traditional transit and distance surveys, gibbal positioning sys- tem measurements, existing cartometric. ‘quality maps, or existing digital data layers ‘on which suitable features may be identified. The Affine Transformation The affine coordinate transformation employs linear equations to calculate map coordinates. Map projection coordinates are often referred to as eastings (E) and north- ings (N), and are relaied to the x and y digi- tizer coordinates by the equations: E = Tptaxtagy ay N = Tytbpctbsy (4.2) Equations 4.3 and 4.2 allow us to move from the arbitrary digitizer coordinate sys- tem to the project map coordinate system, ‘We know the x and y coordinates for every digitized point, line vertex, or polygon ver- tex. We may calculate the E-arid N cootdi- nates by applying the above equations to every digitized point. ‘Te and Ty are translation changes between the coordinate systems, anid can be thought of as shifts in the origins from onecoordinate system to the next. The a; and b; parameters incorporate the change in scales and rotation angle between one coordinate system and the next. The affine is the most commonly applied coordinate transforma- tion because it provides for these three main effects of translation, rotation, and scaling, and because it often introduces less error than higher-order polynomial transforma- tions, ‘The affine system of equations has six parameters (o be estimated, Te, Ty, 04,42, bs, and ba. Each contral point provides E, N, x, and y coordinates, and allows us to write two. equations. For example, we may have a con- trol point consisting of a precisely surveyed center of a toad intersection. This point has digitizer coordinates of x=103.0 centimeters and y = -100.1 centimeters, and correspond- ing Earth-based map projection coordinates of E = 500,083.4 and N = 4,903,683.5. We snay then write two equations based on this control point: 500,083.4-Teay(103.0)+a9(-100.2) (4.3) 4,903,683 5=Tytby(103.0)+ba(-100.1) (4.4) ‘We cannot find a unique solution to these equations, beoause there are six unknowns (Te, Tr, 61, a2, b;, bg) and only: two equations. We need as many equations as unknowns to solve a linear system of equations, Each control point gives us two equations, so we need a minimum of three control points to estimate the parameters of an affine transformation, Statistical estima- tion requires a total of four contro! points. As with all statistical estimates, more control points are betfer than fewer, but we will reach a point of diminishing retums after some aumber of points, typically somewhere between 18 and 30 control points. ‘The affine coordinate transformation 1s usually fit using @ statistical method that Chapter 4: Maps and Data Entry 55 | i | minimizes the root mean square error, | RIMSE. The RMMSE is defined as: : Tee ne retest Rese = [Set BaF On a5) ‘where the e; are the residual distances | between the true E and N coordinates and | the E and N coordinates im the output data layer: i CNRS He) 4) ‘This residual is the difference between the true Coordingtes Xq, ¥4, and the transformed output coordinates X4, ¥g. Figure 4-24 ; shows examples of this lack of fit. Individual | residuals may be observed at each control | point location. ‘A statistical method for estimating | transformation equations is preferred because it aiso-identifies transformation ecror, Control point coordinates contain unavoidable measurement errors. A statisti- cal process provides an RMSE, a summary of the difference between the “true” (mea- sured) and predicted control point coordi- | nates, It provides one index of transformation quality. Transformations are -@:true location: + fit location | SYerron | Figure 4-24: Examples of control points, pre- “ide outa ioaton, ad estdoas Soar dinate wransformation.156 GIS Fundamentals fit (Figure 4-25). The RMSE will usually be tess than the true transformation error at a randomly selected point, because we are actively minimizing the N and E residua) errors when we statistically fit the transfor mation equations. However, the RMSE is an index o accuracy, and a lower RMSE gener- ally indicates a more accurate affine trans- formation. Estimating the coordinate transforma- tion parameters is often an iterative process. Conirol points are rarely exact, and x and y coordinates may not be precisely digitized, Poor eyesight, a shaky hand, fatigue, lack of attention, mis-identification of the control ovation, or a blunder may result in errone- ‘ous x and y values, There may aiso be errors in the E and N coordinates. Typically, con- + tro! points are entered, the affine transforma- tion parameters estimated, and the overall RMSE and individual point £ and N errors evaluated (Figure 4-24, Figure 4-25). Sus- pect points are fixed, and the transformation, re-estimated and errors evaluated until a final transformation is estimated. The wans- formation is then applied to all features to convert them from digitizer to map coordi- nates. Other Coordinate Transforma- tions Other coordinate transformations are sometimes used. The conformal coordinate ‘transformation is similar to the affine, and has the form: E=Teptex-dy (47) N= Twtdercy (4.8) ‘The coefficients Tg, Ty, ¢, and d are esti- sated from control point data. Like the affine transformation, the conformal trans- formation is also a first-order polynomial. Unlike the affine, the conformal transforma- tion requires equal scale changes in the x and y directions, Note the symmetry in the equations 4.7 and 4.8, in that the x and y coefficients match across equations, and there is a change in sign for the d coefficient. This results in 2 system of equations with only four unknown parameters, and so the conformal may be estimated when only wo control points are available, Higher-order polynomial transforma- tions are sometimes used to transform among coordinate systems, An exemple of a 2nd-order polynotnial is; E = bytbyxebyytbyx*ebsy*+byxy (49) Note that the combined powers of the x and y variables may bé up to 2. This allows for curvature in the transformation in botis the x and y directions. 4 minimum of six conttrol points is required to St this Ind-order polynomial transformation, and seven are required when using « statistical fit. The esti- mated parameters Te, Ty, 4, 42, by, and bz will be different in equations 4.1 and 4.2 when conspared to 4.9, even if the same set of contro! points is used for both statistical fits, We change the form of the equations by including the higher-order squared and xy crost-produet terms, and all estimated parameters will vary. A Caution When Evaluating Transformations Selecting tite "best" Goordinate transfor- mation to apply is @ subjective process, guided by multiple goals. We hope to develop an accurate transformation based on 2 large set of well-distributed control points, Isolated control points that substantially ‘improve our coverage may also contribute substantially 10 our transformation errr, ‘There are-no clear nileson the.number of points versus distribution of points trade- off, but if is typically best to strive for the ‘widest distribution of points, We want at least two control points in cach. quadrant of the working area, with a targot of 20% inModel Fit 1: E=1.3328289* x = 0.0058654* y ~ 2068512 N= 0002886 * x + 13296981 * y = 1660286 pMst 36 Examine paints 15 & 17 adjust noted Dhincers. refit model Model Fit 2: 13319386 * x + 000057393" y + 2058124 , N= -0.002462 * x #1329962 "y = 1161855 Rast: 772 Exomine points 4 4 22, adjust noted blunders, refit rode! ‘Model Fit 3: yasue6ay* x +0.0056629* y - 2054903 N= - 0.03816 * x +1,3297296 * y 60243 RMSE= 5.78 Examine points, ro more Binders four. Chepter 4: Maps end Data Entry 157 2a Boiiaove Spleszes 3 i So oloarr Lay ogress Bo38 9086 ie CETL 9 oat “Bouegeas a %. en a (CCRC Tse cei eee orl siegova/“gnnageen TO aap ore. COE SOT negee cao ass [ibaa amar” Ta Beet ae aang smiosrs 8 Big7ist | a00b esto eae joigerbe"~-eggees “1 samara em Tosi eo gpa ass Yaa “aa “anaes Sm9eS Sree 1s “Sus070 roars | Sab wigan saws sony (as ae “ames | | SR TTanagee sans “hair mae a8} CORT Borba gouge [07 Siisoea 1 Boon ceas sor) Figure 4.25; lrerative fiting of ax affine transformation, Control points were examined after each fit, to discover mers entry oF poor matching of potas, Como] points wala lng residale were ext 80, {nod to determine if the cause for the error may be identified. the control point cootdinases:may be ‘modified, and transformation re-fit.158 GIS Fundamentals each quadrant. This is often not possible This latter reason is less common with the development of GNSS. ‘The transformation equation should be developed with the fol- lowing observations in mind, Hirst, bad control points happen, but we should thoroughly justify the removal of any control point, Every attempt should be made to identify the source of the error, either in the collection or in the processing of field coordinates, the collection of image coordi- nates, or in some blunder in coordinate tran- seription. A common error is the mis- identification of coordinate location on the image or map, for example, when the control location is placed on the wrong side of a zoad. Second, a lower RMSE does not mean 2 better transforruation, The RMSE is a useful tool when comperiig among transformations that have the same model form, for example, when comparing one affine to another affine First order transformation as in Figure 4-25. The RMSE is not useful when comparing among different model forms, for example, when comparing an affine to ¢ 2nd-order polynomial, The RMSE 4s typically lower for a 2nd and other higher order poiynomials then au affine transforma- tion, but this does not mean the higher-order polynomiai provides a more accurate trans- formation. The higher-order polynomial will introduce more error than an affine transfor ‘mation on most orthographic maps, and an affine transformation is preferred. High- order polynomials allow more flexibility in warping the surface to fit the control points. ‘Unfortunately, this warping may signifi- cantly deform the non-control-point coordi nates, and add large errors when the transformation is applied to all data in a layer (Figure 4-26). Tus, high order poly- nomials and others should be used with cau- tion. . Finally, independent tests of the trans- formations make the best comparisons 3rd order transformation RMSE = 4.2 m 2 ‘Figure 426: Anvllustration that RMSE: should not be used to compare different otder transfor: ‘mations, nor should il be used as the sols criterion for selecting the best transformation. Above are portions of a transformed image that was registered 1o a road network. This area is interstitial te 18 well distributed conteol points. Because the 3r<-order polyaomiel is-quile flexibledn fitting the points end seducing RMSE, it distorts areas bevweedi the control points, This'is:shown by.the ‘oct mutch berween mage and vector roads, above right. Akhough thas @higher RMSE, the first order transformation on the left is better overall,among transformations. A completely. inde- pendent set of well distributed test points would appear to be ideal, but these rarely exist. The extra points either haven't been collected, or suitable locations do not exist, ‘The best way to test the accuracy of the transformation typically uses a “bootstrap” approach that treats each point as an inde- pendent test point, One point is withheld, the transformation estimated, and the error at the withheld point calculated. The point is replaced in the estimation set, and the next point withheld, fitting the. same type of transformation. The equations will be slightly different. The error at this second withheld point is then calculated. This pro- cess is repeated for each control point, and a mean error calculated. Control Point Sources: Survey- ing Traditional ground surveys based on optical surface measurements are a common, although decreasingiy used method for determining control point locations. Moder surveys use complex instruments such as. transits and theodolites to precisely measure the relative location of points. If the survey starts from a known point, then the coordi- nate location of any survey station may be determined via simple trigonometric func- tions, Federal, state, county, and lecal gov- emments all maintain a set of accurately surveyed locations (Figure 4-27), and these points may be used as control points or as starting points for additional surveys. Many of these known points have been established. using traditional surveying techniques. Indeed, the development of this “control net- work” infrastructure is one of the first and ‘most important responsibilities of govern- ment. These strvey points form the basis for distance, location, and area measurements ‘used to define property, political, and munic- pal boundaries, As a result, this control net work-underlies most commerce, ‘transportation, and land ownership and man- agement. Coordinates, general location, and descriptions are documented for these con- trol networks, and may be obtained from a Chapter 4: Maps and Data Entry 159 number of government sources..Inthe United States these sources include county surveyors, state surveyors, departments of ‘wansportation, and the National Geodetic Survey (NGS) ‘The ground survey network is eften, quite sparse and insufficient for registering many large-scale maps or images. Even when there is a sufficient number of ground- surveyed points in-an area, many may not be suitable for use as control points in a coordi- nate trensformation of spatial data. The con- trol points may not be visible on the maps or images to be registered. For example, a sur- ‘veyed point may fall aiong the edge of road. If the conirol:point is at a mapped road intersection, we_may use the easting and northing coordinates of the road intersection as a control foint during map registration. However, if the surveyed point is along the edge of a road that is not near any mapped feature such as a road intersection, building, or water tower, then it may not be used.as a control point. Our control points must have two characteristics to be useful: first, the point must be visible on the map, data layer, of image that we wish to register, and sec- Figure #227-'Previous surveys ate-a.commion source 6f control points.160 GIS Fundamentals Figure 4-28: Poteatiai control points, indicated here by arrows, may be exiracted from digital reference images. Permanent, well-defined features are identified and coordinates determined from the digital tmnge Nols tne white ers civled i the lower right comer. This is a photogrammettc pane, ypc ‘ppialya plate ‘or painted wooden target placed pricr to photo capture, and wit precisely surveyed coordinates. These tar- guts are used to create the corrected digital image with « known coordinate system, a process deseribed in Chapter 6. ond, we must have precise ground coordi nates in our target map projection. The first requirement, visibility on the source map or photograph, is‘often not met for survey- defined control. Therefore, we must often objaio additional control points. ‘One option for obiaining control points isto perform additional surveys that measure ‘the coordinates of features that are visible on the source materials. Precise surveys are used to.establish the coordinate locations of a well-distributed, sufficient set of points throughout the area covered by the source: map. While sometimes expensive, new sur- ‘veys are the chosen method when the highest. accuracies are required. Costs were probibi- tive with-traditionél-opticai:surveying metb- ods, however, GNSS positioning technologies allow more frequent,.custom collection of control points. Control Points from Existing Maps and Digital Data Registered digital image data are com- mon sources:of ground:control-points;par- ticularly when naturdl zesouiees‘or municipal databases are to be developed for managing large areas. Digital images often provide richly-detaiiedidepiction of surface features (Figure 4-28).Digitil'image data may be obtained that-are registered to a known coordinate system. Typically, the coordinates ofa comerpixel are provided, and the lines and columns for the image run paratle} to the easting (E) and northing (N) direction of the coordinate system, Because the. pixel dimensions:are-known, the caleula- tion ofa pixel:coordinate involvesanultiply- ing the row and-coiumm-mumber-by tho pixe] size; and applying the comer offset, either by addition or subtraction. In this manner, the image row/column may be-converted to-anE, N coordinate pais, and contro! point caor- dinates determined. ‘Existing maps are another common. source of control points. Point locations are plotted and coordinates often printed on maps, for example the comet location coor- dinates are printed on USGS quadrangle maps. Road intersections and other well- defined locations are often represented on maps, If enough recognizable features can be identified, then control points may be obtained from the maps. Control points derived in this manner typically come only from cartomettic maps, those maps pro- duced with the intent of giving an accurate, map-projected representation of features on the Earth’s surface. Existing digital data may also provide control points. A short description of these digital data sources are provided here, and expanded descriptions of these and other digital data are provided in Chapter 7. For example, the USGS has produced Digital Raster Graphics (DRG) files that are scanned images of the 1:24,000-scale quad- rangle maps, These DRGs come referenced toa standard coordinate system, so it isa simple and straightforward task to extract the coordinates of road intersections or other well-defined features that have been plotted on the USGS quadrangle maps. Vector data of roads are often widely available, and if of sufficient accuracy, may be used as a source Control points Chapter 4: Maps and Data. Entry 164 of contro! points at road intersections and other distinct locations. GNSS Control Points The globel positioning system (GPS), GLONASS, and Gatileo are Global Naviga- tion Satellite Systems (GNSS) that allow us to establish control points. GNSS, discussed in detail in Chapter 5, can help us obtain the coordinates of contro! points that are visible ‘on a map or image, GNSS are particularly ‘useful because'we may quickly survey widely-spaced points. GNSS positional accuracy depends on the tecmology and methods employed; it typically ranges from sub-centimeter (tenths of inches) to a few meters (tens of feet). Most points recently added to the NGS and other government- maintained networks were measured using GNSS technologies. To sum up: control points are necessary for coordinate transformation, and typically anumber of control points ate identified for a study area, The x and y coordinates for the control points are obtained from a digitized map or image, and the map projection coor dinates, E and N, are determined from sur- vey, GNSS, or other sources (Figure 4-29), ‘These coordinate pairs are then used with a set of transformation equations to convert data layers‘into-a:desirable map coordinate system. Projection ordinates (UTM) £ N Bos,oss a 5000. 68a 2.082 ah Sno 2.9075 5.002.400. 505,495.31 5001 6735] 302.101.315.001 5540) [sos.o8e 4 50736750] 4.005 840.0) 506, Digitizer coordinates 1005 J [4.908.035.) ‘feos? 9] 4905 47a. Feoiees:il «.s09,176.0 50:360073] «097-296. [soo.ans sl 4so7 eas.) igure 4-29: An example of control point locations from a road data layer, end corresponding dgitizet ‘and map projection coordinates.162 GIS Fundamentals Raster Geometry and Resam- pling Deta often must be resampled when converting between coordinate systems, or changing the cell size of a raster data set (Figure 4-30), Resampling involves reas- signing the cell values when changing raster coordinates or geometry. Resampling is required when changing cell sizes because the new cell centers will not align exactly with old cell centers, Changing coordinate systems may change the direction of the x and y axes, and GIS systems often require that the celi edges align with the coordinate systems axes. Hence, the new celle often do not correspond to the same locations or extents as the old cells. Common resampling approaches include the nearest neighbor (taking the out- put layer value from the nearest input layer cel] center), bilinear imterpolation, (distance- Input rastel based averaging of the four nearest cells), and cubic convolution (a weighted average of the sixteen nearest cells, Figure 4-30). An exampie of a bilinear interpolation is shown in Figure 4-33. This algorithm uses a distance-weighted average of the four near- est celis in the inpat to calculate the value for the output. The new output location is repre- sented by the black: post. Initially, the height, of Zeut Value, of the output location is unknown. Zou is calculated based on the dis- tances between the output locations and the input locations. The distance in the x direc tion is denoted in Figure 4-31 by d;, and the istance in the y direction by d,. The vaiues in the input ate shown as gray posts and are Iabeled as Z; through Z«. Intermediate heights Z,, and Z, are shown, ‘These repre- sent the average of thé input values when taken in pairs in the x’direction. These pairs are, Z, and Za, to yield Z,, and Z, and Za, to yield Z,. Z, and Z, are then averaged to cal- [EF nearest neighbor assigns an surput cell vlue from nest corresponding put eel bilinear interpolation uses the nearest call and nex: three closest cells in a weighted average ‘cubio-comolution includes the tells used fr bilinesr interpolation and lores cells To compute ‘the next iz cl alwelghted avarage for each ot Figure 4-30: Rester resampling. When the orientation or cell size of a raster datz set is changed, ant- ‘pul cell values are calculated oased on the closest (nearest nejghnor), four neaces: (bilinear interpala tion), or sixteen closest (cubie convolution) input celi values.Chapier 4; Maps and Data Entry 163 What is the value of Zou? 2, = 24+ Zy- Zid, t 2,2Ze+ ZZ z Zon 2, Ey-ZpICe € Zy= 14+ (4.0- 1492.9 = 3.26 “e Zy24+ (62 4)2.9= Ble 5 Zou # 3.26 + (B16 - 3.26)"2.2 = 4.1 5 4-31: Te bilinear interpolation method uses e distance weighted average to essign the output ‘value, Zouts based on input values, Z; through Z, crate Zyyp, using the distance dz between ‘the input and output locations to weigint val- ues at each input location, ‘Fhe cubic convo- lution resampling calculation is similar, except that more celis are used, and the weighting is not an average based om linear distance. Map Projection vs. Transforma- tion ~ Map transformations should not be confused with map projections. A map transformation typically employs a statisti- cally-fit linear equation to convert coordi- nates from one Cartesian coordinate system to another. A map projection, described in Chapter 3, differs from a transformation in ‘that it is an analytical, formula-based con- version between coordinate systems, usu- ally from’a curved, Jatitude/iongitude coordinate system to a Cartesian coordinate system. No statistical fitting process is used with a map projection. Map transformations should rarely be used in place of map projection equations when converting geographic data between map projections, Consider.an example when data-are delivered :toan organization | in Universal Transverse Mercator (UTM) coordinates and are to be:converted to State Plane coordinates prior to integration into a GIS database. Two.paths may be chosen. The first involves projection from UTM to164 GIS Fundamentals geographic coordinates (latitude and longi- tude), and then from these geographic coordinates to the appropriate State Plane coordinates. This is the correct, most acce~ sate approach. An aitemate and often less-acourate approach involves using a transformation to convert between different map projec- tions. In this case a set of contro! points would be identified and the coordinates. determined in both UTM and State Plane coordinate systems. The transformation coefficients would be estimated and these equations applied to al] data in the UTM. data layer. This new output data layer would be in State Piane coordinates. This transformation process should be avoided, as a transformation may introduce addi- tional positional error. ‘Transforming between projections is used quite often, inadvertently, when digi- ‘tizing data from paper maps. For exampie, USGS 1:24,000-scale maps are cast on polyconic projection. If these maps are dig- itized, it would be preferable to register them to the appropriate polyconic projec- tion, and then re-project these data to the desired end projection. This is often not done, because the error in ignoring the pro- jection over the size of the mapped area is typically less than the positional error asso- ciated with digitizing. Experience and spe- cific calculations have shown that the spatial errors in using a transformation instead of e projection are small at these map scales under typical digitizing condi- tions. “This second approach, using a transfor- mation when a projection is cailed for, should not be used until it has been tested as appropriate for each new set of condi- tions. Each map projection distorts the su~ face geometry. These distortions are complex and nonlinear, Affine or polyno- mial transformations are unlikely to yemove this non-linear distortion, Excep- ‘tions to this rule occur when the area being transformed is small, particularly when the projection distortion is.small relative to the random uncertainties, transformation errors, of errors in thé spatial data, How- ever, there are no guidelines on what con- stitutes a sufficiently “small” area. In our example above, USGS 1:24,000 maps are often digitized directly into a UTM coordi- nate system with no obvious ill effects, because the errors in map production and digitizing are often much larger than those in the projection distortion for the map area. However, you should not infer this practice is appropriate under all conditions, particularly when working with smaller- scale maps. Output: Hardcopy Maps, Digital.Data, and Metadata We create spatial data to use, share, aud archive. Maps are often produced during data creation and distribution, as intermedi- ‘ate documents while editing, for analysis, or as finished products to communicate some aspect of our date. To be widely useful, we ‘must also generate information, or “meta data.” about the spatial data we've created, and we may have'to convert our data to stan- dard forms. This section describes some characteristics of data output. We start with a brief treatment of cartography and map design, by which we produce hardcopy and digital maps. We then provide a description of metadata, and some observations on data conversion and data transfer standards, Cartography and Map Design Cartography is the.art and techniques of snaking maps. It encompasses both mapmak- ing tools and how these tools may ‘bé.com- ined to communicate spatial information, Cartography is a discipline of much depth and breadth, and there are many books, jour- nal articles, conferences, atid societiesdevoted to the science and art of cartogra- phy, Our aim in the next few pages is 0 pro- Vide a brief overview of cartography with a particular focus on map design. This is both to acqueint new students with the most basic concepts of cartography, aad help tei appiy these concepts in the consumption and production of spatial information. Readers interested in a more complete treatment should consult the references listed at the end of this chapter. A primary purpose of cartography is to communicate spatial information. This requires identification of the -intended audience, ~information to communicate, ~area of interest, . -physical and resource limitations, ia short, the whom, what, where, and bow we may present our information. ‘These considerations drive the major cartographic design decisions we make each time we produce a map. We must consider the: -scale, size, shape, and other general map properties, ~Gata to plot, symbol shapes, sizes, or patterns, -labeling, including type font and size, legend properties, size, and borders, and -the placement of all these elements or amap. ‘Map scale, size, and shape depend pri- marily on the intended map use, Wall maps for viewing at a distance of a meter or more may have few, large, boldly colored features In contrast, commonly produced street maps for navigation in metropolitan areas are detailed, to be viewed at short ranges, and have arich set of additional tables, lists, oF other features. Map scale is often determined in part by the size of the primary objects we wish to display, and in part by the most appropriate Chapter 4: Maps and Data Entry 165 media sizes, such as the page or screen size possibie for a document. As noted earliez, the map scale is the ratio of lengths on a map to true lengths, If we wish to display-en area that spans 25 kilometers (25,000 meters) on z yorcen (hal spans 25 centimeters (0.25 meters), tite map scale will be near 0.25:t0 25,000, or 1:100,000, This decision on size, area, and scale then drives further map design. For example, scale limits the features we may display, and the size, mmmbex, and labeling of features, At e 1:100,000 scale we may no: be able to show all cities, burgs, aud towns, as thete may be too many to fit at a readable size. Maps typically have a primary theme or purpose that is. determined by the intended audience. Is the map for a general popuia- tion, or for a target audience with specific expectations Yor map features and design? General purpose maps typically have a wide range of features represented, including ‘transportation networks, towns, elevation or other common features (Figure 4-322). Spe- cial purpose maps, such as road maps, focus on a more limited set of features, in this instance road locations and names, town names, and large geographic features (Figure 4-32b}, Once the features to include on a map are defined, we must-choose:the.symbois used:to draw thernSyinibolo, part on the:type:of feature: Forexample;-we have a different set of options when repre senting continuous features such as elevation or pollution concentration than when repre- senting discrete ‘features, We also must. choose among-symbbis for each of the types of discrete Features, for example, the.set. of syntbols for points.arc generally different Srom those for line or area features. Symbol size is an important attribute of mep symbology, often specified in a-unit called a.poitt, One point:is-approximately equal: to 0:467-xam, or about 172 of an inch. Ac specific poisit number is mostioften used to-specify the size of symbols, for example, the dimensions of small-squares to represent houses on a map,or the characteristios.of a specific pattern used to fill areas on.emap’A168 GIS Fundamentals line width may also be specified in points Setting 2 line width of two points means we ‘want that particular fine plotted with a width of 0.93 mm. It is unfortunate that “point” is both the name of the distance unit and a gen- eral. property of 2 geographic feature, as in “q bee is a point feature.” This forces us 10 talk about the “point size” of symbols to rep- resent points, lines, or area fills or patterns, but if we are careful, we may communicate these specifications clearly. The best size, pattern, shape, and color used to symbolize each feature depends on the viewing distance, the number, density, and type of features, and the purpose of the map. Generally, we use larger, bolder, or thicker symbols for maps to be viewed from longer disiances, while we reduce this limit when producing maps for viewing at 50 cn (A8 inches), Most people with normal vision under good lighting may resolve lines down to near 0.2 points at close distances, pro- vided the lines show good contrast with the background. Although size limits depend largely on background color and contrast, point features are typically not resalvable at sizes smaller than about one half point, and distinguishing between shapes is difficult for point features smaller than approximately two points in their largest dimension. The pattern and color of symbols must also be chosen, generally from a set pro- vided by the software (Figure 4-33), Sym- bols generally distinguish among feature type by characteristics, and although most symbols are not associated with a feature type, some are, such as, plane outlines for airports, mumnbered shields for highways, o: a hatched line for railroad ‘We also must often choose whether and how to label features, Most GIS software provides a range of tdols for creating and placing labels, and in all cases we must choose the iabel font type and size, location relative io the feature, and orientation. Pri- mary considerations when labeling point Figure 4-32: Exannple‘of a) a detailed, general-purpose map, here a portion of a US Geological Survey than, ard b) a specialized map foeusmga specie St of selected features, here showing ross. The fee- tures chosen for depiction on the mep lepend on the intended map use.features are label placement relative to the point location, label size, and label otienta- tion (Figure 4-34). We may also use gradu- ated labels, that is, resize them according to some variable associated with the point fea- ture. For example, it is common to have larger features and label fonts for larger cit- ies (Figure 4-34), Labels may be bent, angled, or wrapped around features t© improve clarity and more efficiently use space ina map. Label placement is very much an art, and there is often umich individual editing required when placing and sizing labels for finished maps. Most software provides for automatic labei placement, usually specified relative to feature location. For example, one may specify labels above and to the right of ail points, or lines labels placed over line features, or polygon labeis placed near the Figure 4-33: Examples of point (top), line (mid), and area (bottom) symbois used to distingui: ‘among features of different types. Most GIS soft {ware provide a set of standard symbols for point, line, area, and continuous surface features. ‘Chapter 4: Maps and Data Entry 167 Cordoba . cordoba * Be macne Hsevilia Alcintara wv oe a Figure 4-34; Common labeling options, inelud- ‘ng straight, angled, wrapped text, end grade- ated labels for points (cop nwo sets), and angled, wrapped, fonting, and embedded labels for le and polygor features (bottom, thro se). polygon-centroid. However, these:automatic placements may not be satisfactory because iabels may overlap, labels may fall in clut- tered areas of the map, or features associated with labels maybe ambiguous. Some soft- ware provides options for simple to elabo- rate automatic label placement, including automatic removal or movement of overlap- ping labels. These often reduce manual edit- ing, but sometimes increase it. Figure 4-35 shows a portion of a map of southern Finland, This region presents sev- eral mapping problems, ineluding the high density-of cities nearthe.upper-right, an irregular coastline, aid dense clustering of islands along the coast. Most labels are placed above and to the rightof their corre- sponding city, however some are moved-or168 GIS Fundamentals angled for clarity, Cities near the coast show both, to avoid labels crossing the water/tand boundary where practical. Semi-transparent background shading is added for Parainen and Hanko, cities placed in the island matnx. This example demonstrates the mdrvidual editing often required when placing labels. “Most maps should have legends. The legend identifies map features succinctly and describes the symbols used to depict those features, Legends often include or are grouped with additional map information such as scale bars, north arrows, and descrip- tive text. The carlographer must choose the size and shape of the descriptive symbol, and the font type, size, and orientation for ‘each symbot in the legend. The primary goal is to have a clear, concise, and complete ieg- end. ‘The kind of symbols appropriate for map legends depends on the types of fes- tures depicted, Different choices are avail- able for point, line, end polygon features, or for coutinuously variable features stored as asters. Most software provides a range of Figore 4.38: Exarsple label placement for cities and in southern Fin legend elements and symbols which may be used. Typically these tools allow a wide range of symbolizations, and a compact way of describing the symbolization in a legend (Figure 436). ‘The specific iayout of legené features must be defined, for exampie the point fea- ‘ture symbot size may be graduated based on some attribute for the points. Successively iarger features may be assigned for succes- sively larger cities, This must be noted in the legend, and the symbols nested, shown sequentially, or otherwise depicted (Figure 4-36, top left). ‘The legend should be exhaustive, Exam- ples of each different syminol type that * appears on the map should appear in the leg- end. This means eachfpoint, line, or area symbol is drawn in the legend with some ~ descriptive label. Labels may be next to, wrapped around, or embedded within the features, and sometimes descriptive numbers are added, for example, a range of continu- ous variables (Figure 4-36, upper left). Scale bars, north arrows, and descriptive text boxes are typically included in the legend. ‘Map composition or layout is another primary task. Composition consists of deter- mining the map elements, their size, and their placement. Typical map elements shown in Figure 4-3 and Figure'4-4, include one or more main data panes or-areas, a leg- end, atitle, a scale bar and-north arrow, a grid or graticule, and perliaps descriptive text. These each must be sized and placed on the map. These map elements should be posi- tioned and sized in accordance-with their importance, The map’s most important data pane should be Jargest, and itis often cen- tered or otherwise given visuai dominance. Other elements are typically smatier and Jocated around the periphery orrembedded veithin the main data pane. ‘These-other ¢le- ments include map insets, which are smaller data panes that show larger or smaller scale views ofa region in the primary data pane. Good map compositions usually group related elements and uses empty space.effec-tively. Data panes are often grouped and leg- end elements placed near each other, and grouping is often indicated with enclosing boxes Neophyte cartographers should avoid two tendencies in map composition, both depicted in Figure 4-37. First, it is generally easy to create a map with automatic label and legend generation and placement. The map shown at the top of Figure 4-37 is typi- cal of this automatic composition, and" * inoludes poorly placed legend elements and +100 small, poorly placed labels, Labels crowd each other, are ambiguous, cross water/land or other feature boundaries, and fonts are poorly chosen. You should note that automatic map symbol selection and placement is nearly always sub-optimal, and the novice cartographer should scrutinize ‘these choices and manually improve them. ‘The second common error is poor use of empty space, those parts of the map without tmap elements. There are two opposite ten- dencies: either to leave too much or unbal- anced empty space, or to clutter the map in an attempt to fill all empty space. Note that 109 Graduated Symbols Legend 6 Q 2% 80 78 Line Legend Major . Road ) ‘Figure 4-36; Examples of legend elements and grouped iu a corapact way io commtinioate the values associated with each symbol, Chapter 4: Maps-and Deta Entry “469 the map shown at the top of Figure 4-37 leaves large empty spaces on the left (wesi- em) edge, with the Atlantic Ocean devoid of features. The cartographer may address this in several ways, either by changing the.size, shape, or extent of the area mapped, adding new features, such as data panes as insets, additional text boxes, or other elements, or moving fhe legend ar other map elements to that space, The: map shown at the bottom of Figure 4-37, while not perfect, fixes these Gesign flaws, in part by moving the legend ‘and scale bar, and in part by adding labels fox the Atlantic Ocean and Mediterranean Sea, The empty space is more balanced in that it appears around the major map ele- ments in approximately equal proportions. ‘As noted garlier, this is only 2 brief introduction tb cartography, a subject cov- ered by many good books, some listed at the end of this chapter. Perhaps the best com- pendium of examples is the Map Book Series, by ESRI, published amnually since 1984, Examples are available at the time of this writing at www.esr.con/mapmuseum. ‘You should leaf through several volumes in Area Legend -Grostland North Arrows sentation of symbels.'Some symbols may be sequential or ‘ested prnuated ccles to represent city population siz, ares pte cr colo ils 9 istinguish ‘fferent polygon featares, line an ‘and north arrows. point symbols, and informative clements such es scale bars