GIS application & spatial analysis
1. GIS links a place to information
A geographical information system (GIS) stores, manages, analyses and displays data linked to locations. A digital map can show where a park is; a GIS can also ask which residents can reach it, what land use surrounds it and how an alternative site would change access. The geographical question should drive the analysis.
Spatial data describe location and shape. Attribute data describe characteristics linked by an identifier, such as a park’s area, opening hours and step-free entrances. A coordinate without the correct reference system can place a feature in the wrong location. A beautifully drawn output is not evidence that the input was accurate.
These notes develop general geographical skills for interpreting GIS outputs and planning enquiries. No particular software brand is prescribed here. All named sites and datasets below are synthetic examples, not actual planning recommendations for Hong Kong.
2. Layers, vector and raster
Layers separate themes while allowing comparison in the same space: roads, buildings, relief, flood exposure and population. A layer is not simply a colour: it contains locations, attributes and metadata. Turning off the flood layer hides it from view; it does not remove flood risk.
Vector data represent discrete features as points, lines and polygons. A bus stop can be a point, a road a line, and a park boundary a polygon. Representation depends on scale: a small park may be a point on a regional map but a polygon in a neighbourhood study. Lines connect only if network topology records the connection; crossing roads at different levels are not necessarily junctions.
Raster data represent space as cells with values, such as elevation or surface temperature. Smaller cells can retain more detail but increase storage and do not create information missing from the source. A \(100\,\mathrm{m}\) cell may mix buildings, roads and trees; resampling it to \(10\,\mathrm{m}\) cannot reveal ten-metre measurements that were never made.
Attribute example, one row per site: A — park, area \(2\,\mathrm{ha}\), entrance open; B — vacant land, area \(3\,\mathrm{ha}\), entrance closed. Joining a visitor table by unique site ID is safer than matching similar names. Check missing IDs, duplicate rows, units and dates before comparing values.
3. From enquiry to trustworthy input
Define the question and criteria first: “Where could a small public park serve residents who currently lack access?” needs residential locations, existing park entrances, a walkable network, barriers and candidate land. A land-price layer may matter later, but an unrelated layer adds complexity without answering the question.
Check metadata: producer, collection date, coordinate reference system, scale/resolution, attribute definitions, completeness and permitted use. Use a suitable projected coordinate system for distances in metres. Latitude and longitude are angles; subtracting their numeric values does not directly give metres. Reproject compatible layers properly rather than merely changing a label.
Validate a few locations against reliable reference data or field observations. Check that polygons close, roads connect where appropriate and categories use consistent spelling. Distinguish “no data” from zero and “not surveyed” from absence. Where possible, document uncertainty and retain the unmodified input alongside the cleaned version.
Workflow schematic: question → relevant layers → align coordinates and dates → clean and check → query/overlay → compare alternatives → field-check and communicate. Record each threshold and transformation so another student can reproduce the result. GIS output should be interpreted using geographical knowledge, not accepted because a computer produced it.
4. Queries, buffers and overlay
An attribute query selects records by values, such as vacant land with area at least \(2\,\mathrm{ha}\). A spatial query selects by location, such as homes inside a flood polygon. “\(\text{Area}\geq2\) AND \(\text{slope}\leq5^{\circ}\)” requires both conditions; “OR” accepts either. Read inclusive boundaries carefully: at most \(500\,\mathrm{m}\) includes exactly \(500\,\mathrm{m}\).
A buffer marks places within a stated distance of a point, line or boundary. A \(500\,\mathrm{m}\) straight-line buffer around a park entrance is a first screen for proximity, not proof of a \(500\,\mathrm{m}\) walk. A river, fence or missing crossing can make the route much longer. Use entrances rather than polygon centres where access is the question.
Overlay combines layers to identify shared locations. Intersect candidate land with suitable-slope areas, then exclude flood-exposed land. “Intersect” keeps overlap; “union” combines coverage and attributes; “clip” limits a layer to a study boundary. State which operation answers the enquiry rather than naming every available command.
Worked selection table: site A — area \(3\,\mathrm{ha}\), slope \(3{}^{\circ}\), entrance distance \(400\,\mathrm{m}\), flood-exposed yes; B — \(2\,\mathrm{ha}\), \(4{}^{\circ}\), \(450\,\mathrm{m}\), no; C — \(4\,\mathrm{ha}\), \(7{}^{\circ}\), \(300\,\mathrm{m}\), no; D — \(2\,\mathrm{ha}\), \(2{}^{\circ}\), \(650\,\mathrm{m}\), no. Criteria are \(\text{area}\geq2\,\mathrm{ha}\), \(\text{slope}\leq5^{\circ}\), \(\text{distance}\leq500\,\mathrm{m}\) AND outside flood exposure. Only B passes. A fails flood exposure, C fails slope and D fails distance. “Closest” alone would select C and ignore the other criteria.
A flood polygon is a model for stated conditions and probability, not a boundary separating certain flooding from perfect safety. Screening out that polygon reduces one identified exposure; it cannot establish that a site has no risk.
5. Networks, spatial patterns and site decisions
Network analysis follows connected routes and can account for travel time, one-way movement or inaccessible steps. The shortest distance route is not necessarily the fastest or most accessible. In a synthetic network, \(\text{A–B}=300\,\mathrm{m}\), \(\text{B–C}=400\,\mathrm{m}\) and \(\text{A–C}=900\,\mathrm{m}\): the shortest A-to-C route via B is \(700\,\mathrm{m}\). If B–C closes, the remaining direct route is \(900\,\mathrm{m}\). Do not draw a shortcut across a river unless a usable connection exists.
Point-in-polygon counting can total facilities inside each district. Divide by population or area when comparing provision, and explain the denominator. A district with \(10\) parks and \(100{,}000\) people has one park per \(10{,}000\) people; another with \(5\) parks and \(20{,}000\) people has one per \(4{,}000\). Park size, quality and accessibility still matter.
Spatial clustering may suggest a relationship, such as high temperatures near dense development, but does not prove a cause. Population concentration, measurement timing and the size of reporting areas can alter the pattern. Changing district boundaries or class intervals can change a choropleth’s appearance without any change on the ground. District averages cannot describe every individual resident.
Multi-criteria selection requires transparent priorities. First exclude sites that fail essential conditions; only then compare eligible sites. For optional scores rated \(1\text{{–}}5\) where larger is better, site P with access \(4\) and cooling need \(2\) scores \(0.6\times 4 + 0.4\times 2 = 3.2\); site Q with access \(2\) and need \(5\) scores \(3.2\) too. With weights \(0.4\) and \(0.6\), P scores \(2.8\) and Q \(3.8\). The ranking depends on priorities, not an objectively best answer produced by GIS.
6. Check, communicate and evaluate
A useful output has a title, legend, units, scale where relevant, orientation, data dates and source notes. Explain what a colour or symbol represents. Use comparable class intervals when comparing maps and avoid using a colour change to imply a sharp physical boundary that the data cannot establish.
Field-check shortlisted sites: an apparently connected path may be locked, a vacant plot may have changed use, and a mapped entrance may have steps. Compare the GIS prediction with observations and revise the dataset. Do not publish precise household locations from a small questionnaire; aggregate data sufficiently to protect respondents.
Evaluate positional accuracy, attribute accuracy, missing data, age, resolution and assumptions separately. A recent layer can still use coarse cells; high resolution can still contain incorrect attributes. Check sensitivity by changing a plausible buffer distance or criterion weight. If the selected site changes easily, describe the decision as sensitive and collect stronger evidence.
Exam approach: identify the geographical purpose, name the required layers and attributes, explain the analysis in order, interpret the result with values, then state a limitation and a check. “Use GIS to find the best place” is incomplete; “overlay vacant land with slope and flood exposure, retain sites meeting the criteria, then check access on foot” explains the process.
Self-check: a satellite receiver obtains location; GIS analyses location-linked data, so GPS and GIS are not synonyms. A buffer measures proximity under its stated distance rule, not automatic accessibility. Overlaying mismatched coordinate systems can create false overlap. A selected site is a candidate requiring judgement, not permission to build.
7. A complete accessibility screen
Synthetic local coordinates are in metres. A park entrance is \(P=(0,0)\); homes \(H_{1}=(300,400)\), \(H_{2}=(600,0)\), \(H_{3}=(0,400)\), with \(4\), \(3\) and \(2\) residents respectively. A \(500\,\mathrm{m}\) Euclidean buffer includes its boundary. Calculate H1 distance as \(\sqrt{300^{2}+400^{2}}=500\,\mathrm{m}\); H2 is \(600\,\mathrm{m}\) and H3 is \(400\,\mathrm{m}\). H1 and H3 are inside or on the boundary.
Attribute join and summary: eligible home IDs H1/H3 → resident counts \(4/2\) → sum \(6\) residents out of \(9\), approximately \(67\%\). The share describes this tiny synthetic dataset only. Counting two of three homes also happens to give about \(67\%\) here, but that equality is accidental and would change if household sizes differed in another way.
Now add a barrier layer: a fenced railway separates H1 from P, and the nearest usable crossing makes its walking route \(850\,\mathrm{m}\). H1 remains in the straight-line buffer but falls outside a \(500\,\mathrm{m}\) network service area. Recalculate using the actual connected route, not by deleting the home from the population table. If H3 has an unobstructed \(400\,\mathrm{m}\) route, only its \(2\) residents meet the walking-distance criterion.
The network result is therefore \(2/9 \approx 22\%\) for the stated criterion. It does not mean the remaining residents never use the park: they may travel farther or use another mode. The criterion is a chosen planning threshold, not a universal behaviour rule or a guarantee of equal access.
Communicate the original screen, the revised network result and the reason for the difference. Before using it for a real decision, check the crossing, entrance opening hours, steps and current resident data. Document that the result assumes all household members use the same entrance and counts each resident once; alternative entrances or mobility needs can change the result.