Geographic Phenomena and Spatial Data Models
Open in the Padhaiko app2.1 Geographic phenomena: fields and objects, boundaries, scale and resolution
A geographic phenomenon is any real-world feature, property, or process that exists at or near the Earth's surface and can be named, measured, or located in space and time. To store such phenomena inside a finite computer memory, GIS conceptualizes reality using two complementary views: geographic fields and geographic objects.
Fields vs. Objects Conceptualization
- Geographic Fields: Phenomena that have a well-defined value at every point in a continuous spatial domain, modeled mathematically as a function (or ).
- Continuous fields vary smoothly without abrupt jumps (e.g., elevation, air temperature, barometric pressure, terrain slope).
- Discrete fields divide space into exhaustive, mutually exclusive zones where every point belongs to exactly one category (e.g., land-cover classes, soil types, administrative districts).
- Geographic Objects (Discrete Entities): Identifiable, bounded entities scattered across an otherwise empty space. Each object is characterized by its location, shape, size, and orientation (e.g., buildings, road segments, fire hydrants, lakes).
Boundaries, Scale, and Resolution
- Boundaries:
- Crisp (Sharp) Boundaries: Clear, exact transitions determined by measurement or legal survey (e.g., cadastral parcel lines, administrative borders, road edges).
- Fuzzy (Indeterminate / Transition) Boundaries: Gradual transition zones where one class blends into another without a sharp line (e.g., forest-to-grassland ecotones, soil moisture zones, landslide susceptibility zones).
- Map Scale: The ratio of distance on a map to the corresponding ground distance, expressed as a Representative Fraction (RF), e.g., .
- Large-scale maps (e.g., ) cover a small area with high geometric detail.
- Small-scale maps (e.g., ) cover a large region with heavy generalization.
- Spatial Resolution: The smallest ground feature or cell size that can be distinguished and represented (for rasters, the ground side length of one pixel, e.g., for Sentinel-2; for vectors, the minimum mapping unit and coordinate precision). Coarser resolution smooths out local variability and can alter area/perimeter measurements.
2.2 Attribute data and measurement scales
Non-spatial attributes attached to spatial entities or field cells are classified into four measurement scales (Stevens' taxonomy), which dictate which mathematical, statistical, and cartographic operations are valid:
| Measurement Scale | Nature of Values | Ordering | Meaningful Arithmetic | GIS Examples | Valid Statistics & Symbolization |
|---|---|---|---|---|---|
| Nominal | Qualitative labels or names; no inherent rank | No | Equality (, ) only | Land-use code (Forest, Urban), road class, soil ID | Mode, frequency count; unique hue/category colors |
| Ordinal | Ranked categories; intervals between ranks are unequal or unknown | Yes (, ) | Comparison/rank only (differences meaningless) | Road hierarchy (Highway > Collector > Local), hazard level (Low < Medium < High) | Median, percentile, mode; ordered color lightness |
| Interval | Quantitative numbers with equal unit intervals, but an arbitrary zero | Yes | Addition and subtraction (, ) | Temperature in or , elevation relative to arbitrary local datum, calendar year | Mean, standard deviation, differences; graduated/choropleth |
| Ratio | Quantitative numbers with equal intervals and a true (absolute) zero | Yes | All operations (, , , , ratios) | Population, rainfall (), distance (), Kelvin temperature, parcel area () | All statistics (geometric mean, coefficient of variation, ratios) |
2.3 Raster data model
The raster data model partitions 2D geographic space into a regular, tessellated matrix of equally sized rectangular (usually square) cells (pixels) arranged in rows and columns.
Cells, Resolution, and Georeferencing
- Each cell stores a numeric value representing either a continuous measurement (elevation in meters, reflectance) or an integer class code (land cover), plus a designated
NoDatasentinel for unobserved areas. - Spatial location is implicit from the grid origin , cell width , cell height , and array indices :
where is the center of cell (assuming top-left origin and North-up orientation).
- Spatial resolution equals : halving cell side length quadruples () the total cell count and raw storage for the same footprint.
Raster Encoding and Compression
Uncompressed raw grids waste huge storage when large regions share identical values (e.g., classified land cover or binary masks). Two classic lossless raster compression techniques are:
- Run-Length Encoding (RLE): Scans the raster row by row and replaces consecutive runs of identical cell values with
(run_length, value)pairs.- Best for: Categorical/thematic rasters with large homogeneous patches (soil maps, binary masks).
- Poor for: Continuously varying imagery (satellite bands, DEMs) where adjacent cells rarely match exactly.
- Quadtree Encoding: Recursively subdivides a square raster into four equal quadrants (NW, NE, SW, SE) until every quadrant is homogeneous (all cells share one value) or reaches single-pixel resolution ().
- Homogeneous blocks become leaf nodes storing the single value; mixed blocks become internal gray nodes with four child pointers.
- Supports rapid spatial indexing and multi-resolution retrieval directly on compressed quadrants.
Multi-Band Rasters and Raster Pyramids
- Multi-Band Rasters: Store multiple coregistered layers (bands) for the same grid—for example, Red, Green, Blue, and Near-Infrared satellite bands. Three common disk interleaving layouts exist:
- BSQ (Band Sequential): Entire Band 1 matrix, followed by Band 2, then Band 3 (fast single-band extraction).
- BIL (Band Interleaved by Line): Row 1 of Band 1, Row 1 of Band 2, Row 1 of Band 3, etc.
- BIP (Band Interleaved by Pixel): All band values for pixel , then pixel , etc. (fast multi-band spectral signature lookup per pixel).
- Raster Pyramids (Overviews): Precomputed, downsampled copies of the original raster stored alongside it at successively coarser resolutions (, e.g., by averaging or nearest-neighbor subsampling). When zoomed out on screen, the GIS reads only the appropriate overview level instead of scanning millions of full-resolution pixels, keeping pan/zoom rendering fast.
1
Consider a thematic raster grid representing two land-cover classes ( = Forest, = Water):
Row 0: 1 1 1 1 1 1 2 2
Row 1: 1 1 1 1 2 2 2 2
Row 2: 2 2 2 2 2 2 2 2
Row 3: 2 2 1 1 1 1 1 1- Encode the grid row-by-row using Run-Length Encoding (RLE) as
(length, value)pairs. - If each uncompressed cell uses integer and each RLE pair uses integers, compute the compression ratio ().
Encode each row into (length, value) pairs:
- Row 0: six
1s followed by two2s(6, 1), (2, 2)( runs). - Row 1: four
1s followed by four2s(4, 1), (4, 2)( runs). - Row 2: eight
2s(8, 2)( run). - Row 3: two
2s followed by six1s(2, 2), (6, 1)( runs).
Summing across all four rows gives runs in total.
Compute storage and compression ratio:
- Uncompressed storage integers.
- RLE storage integers.
- Compression ratio:
(Space savings .)
The row-by-row RLE encoding is:
- Row 0:
(6,1), (2,2) - Row 1:
(4,1), (4,2) - Row 2:
(8,2) - Row 3:
(2,2), (6,1)
Total RLE storage is integers compared to uncompressed integers, giving a compression ratio of ( space savings).
Wait—notice the @verify on ex1 above had 2*9! Let's fix that @verify right away when we check it, or fix it before publishing! Actually let's check our draft carefully after writing, and make sure @verify approx((4*8)/(2*7), 16/7) is exact! Good catch! Let's fix it cleanly in the file before moving to content/.
2.4 Vector data model
The vector data model represents discrete geographic objects using exact coordinate pairs (or ) connected into geometric primitives:
Vector Geometric Primitives
- Point (): A single coordinate pair representing features too small to show area at the map scale (e.g., wells, survey benchmarks, street lamps). Has zero length and zero area.
- Line / Polyline (): An ordered sequence of two or more coordinate vertices connected by straight segments (e.g., roads, rivers, power lines). Has length, zero area.
- Polygon (): A closed sequence of vertices where the first and last vertices coincide , enclosing an interior area (and optionally inner rings/holes for islands or courtyards). Has both perimeter and area.
Spaghetti vs. Topological Vector Models
- Spaghetti Model: Stores each feature independently as a raw list of coordinates without recording shared boundaries or connectivity. Fast to draw, but duplicates shared polygon borders (causing sliver gaps/overlaps) and cannot directly answer adjacency or network flow queries.
- Topological Vector Model: Explicitly stores nodes (endpoints/intersections), arcs/edges (directed line segments with
from_node,to_node,left_polygon,right_polygon), and polygons (lists of bounding arcs). Shared edges are stored only once.
Triangulated Irregular Network (TIN)
A Triangulated Irregular Network (TIN) is a vector-based surface representation composed of contiguous, non-overlapping triangles formed from irregularly spaced sample points .
- Construction (Delaunay Triangulation): Connects sample points such that the circumcircle of every triangle contains no other sample points in its interior. This maximizes the minimum interior angle, avoiding long thin "sliver" triangles.
- Advantages over DEM grids: Adapts vertex density to terrain complexity (dense vertices along ridges, valleys, and breaklines; sparse vertices on flat plains) and preserves exact surveyed spot heights and linear breaklines.
- Euler–Poincaré Relation for Planar TINs: For a planar triangulation with vertices ( of which lie on the convex hull boundary and interior vertices), the number of triangles and edges satisfy:
2.5 Topology and spatial relationships
Topology studies the spatial properties of geometric objects that remain invariant under continuous deformations (stretching, bending, rotating, scaling without tearing or gluing)—namely connectivity, containment, and adjacency.
Topological Rules
Geodatabases enforce topological integrity rules during editing to keep vector layers geometrically valid:
- Polygon rules: Must Not Overlap (adjacent land parcels cannot overlap), Must Not Have Gaps (exhaustive subdivisions like wards must leave no sliver voids), Must Be Covered By (buildings inside parcel boundaries).
- Line rules: Must Not Have Dangles (no undershoots/overshoots in road networks), Must Not Self-Intersect, Must Not Overlap.
- Point rules: Must Be Covered By Boundary Of (survey monuments on parcel corners), Must Coincide With (valves snapped onto water pipes).
Dimensionally Extended 9-Intersection Model (DE-9IM)
The DE-9IM formalizes the binary topological relationship between two geometries and by decomposing each geometry into three point-sets:
- Interior (): Points inside the geometry (for a polygon, its inner region; for a line, points excluding endpoints; for a point, itself).
- Boundary (): Points on the border (for a polygon, its bounding ring(s); for a non-closed line, its two endpoints; for a point, ).
- Exterior (): All remaining points in outside .
DE-9IM evaluates the dimension of all nine pairwise intersections in a matrix:
Written row-major as a -character string (e.g., FF2F11212), it uniquely defines named OGC predicates such as Equals, Disjoint, Intersects, Touches, Crosses, Within, Contains, and Overlaps. For instance, two polygons Touch (FT*******, F**T*****, or F***T****) when their interiors do not intersect () while at least one boundary intersects the other's boundary or interior.
OGC Simple Features Geometry Model
The OGC Simple Features specification defines a standard object-oriented class hierarchy for vector geometries (with optional / coordinates) implemented by PostGIS, Shapely, GDAL, and GeoPackage:
| Geometry Class | Dimension | Description & Validity Rules |
|---|---|---|
Point | Single coordinate vertex | |
LineString | Sequence of vertices joined by linear segments (LinearRing if closed & simple) | |
Polygon | One closed outer LinearRing plus zero or more non-overlapping inner hole rings; rings must not self-intersect | |
MultiPoint | Collection of Points | |
MultiLineString | Collection of LineStrings (e.g., braided river channels) | |
MultiPolygon | Collection of non-overlapping Polygons (e.g., an archipelago nation) | |
GeometryCollection | Mixed | Heterogeneous collection of any geometry types |
2
Let polygon and polygon be two adjacent square parcels sharing the vertical edge . Construct the -character DE-9IM matrix string for and determine which OGC spatial predicate holds.
Identify Interior, Boundary, and Exterior of and :
- , perimeter of , .
- , perimeter of , .
Evaluate each row of the intersection matrix:
- Row 1 ( vs. ):
- (2D region)
- Row 2 ( vs. ):
- (a 1D line segment)
- remaining three sides of (1D)
- Row 3 ( vs. ):
- (2D)
- remaining three sides of (1D)
- unbounded outer plane (2D)
Assemble the matrix and classify the predicate:
Because interiors do not intersect () and boundaries share a edge (), the polygons satisfy Touches (and also Intersects).
The DE-9IM matrix string is FF2F11212, so polygon Touches (and Intersects) polygon along a shared boundary edge.
2.6 Raster vs. vector comparison and spatio-temporal representation
| Aspect | Raster Data Model | Vector Data Model |
|---|---|---|
| Spatial Representation | Regular 2D grid of equal-sized cells (pixels) | Discrete coordinates forming points, lines, and polygons |
| Best Suited Phenomena | Continuous fields (elevation, temperature, rainfall, satellite imagery) | Discrete objects with crisp boundaries (parcels, roads, buildings, pipes) |
| Data Structure Complexity | Simple matrix array; fast overlay and map algebra | Complex topological tables (nodes, arcs, polygons); fast network routing |
| Positional Precision | Limited by cell size (stair-stepped edges) | High coordinate precision; crisp smooth boundaries at any zoom |
| Storage Volume | Large (stores every cell unless compressed), independent of feature complexity | Compact for sparse features; grows with vertex density |
| Cartographic Output | Blocky/pixelated when zoomed past native resolution | Clean, crisp graphics at arbitrary map scales |
Temporal and Spatio-Temporal Data Representation
Real-world geographic phenomena evolve continuously or discretely over time (), requiring spatio-temporal data models beyond static snapshots:
- Snapshot Model: Stores a sequence of complete layer states at discrete timestamps (e.g., annual land-cover rasters). Simple to query at a fixed date, but duplicates unchanged features and loses explicit cause-and-effect transitions.
- Space-Time Cube (2D + T): Adds time as a vertical third axis so stationary objects appear as vertical lines and moving trajectories form continuous 3D polylines inside the cube.
- Timestamped / Versioning (Valid-Time & Transaction-Time) Model: Attaches
valid_fromandvalid_toattributes to individual vector features or edges (or records state-change deltas/events), enabling historical rollback and change tracking without copying unchanged geometries. - Event-Based & Moving-Object Models: Represents discrete events (
Split,Merge,Created) or time-parameterized trajectories for vehicles, storms, or wildlife telemetry.
3
A surveyor collects elevation spot-height points across a terrain patch to build a Delaunay TIN, of which points lie on the outer convex hull boundary.
- Calculate the exact number of triangles and edges in the resulting TIN.
- Explain briefly why a TIN is preferred over a uniform raster DEM along sharp mountain ridges and cliffs.
Apply the Euler–Poincaré planar triangulation formulas: Given vertices and convex hull vertices:
Explain the advantage of TIN over raster DEM for sharp terrain: A TIN allows adaptive triangle density and explicitly incorporates breaklines (ridges, river banks, cliff crests) as constrained triangle edges. A uniform raster grid smooths sharp breaklines across fixed square cells unless cell size is made uniformly tiny everywhere, which wastes storage on flat areas.
The TIN contains triangles and edges. Unlike a fixed-resolution raster DEM, a TIN adapts triangle density to terrain roughness and preserves exact breaklines along sharp ridges and cliffs without wasting storage in flat regions.
added
Distinguish between geographic fields and geographic objects with two examples each, and explain how crisp vs. fuzzy boundaries arise in spatial modeling.
Geographic fields vs. objects:
- Geographic fields assign a value to every location across space—either continuous (elevation, temperature) or discrete (soil class, land-cover map).
- Geographic objects are discrete, bounded entities separated by empty space, each possessing geometric identity and attributes (buildings, road segments, fire hydrants).
Crisp vs. fuzzy boundaries:
- Crisp boundaries mark sharp transitions defined by physical edges or legal surveys (e.g., cadastral property lines, paved road curbs).
- Fuzzy boundaries occur where phenomena change gradually over a transition zone (e.g., forest-to-savanna vegetation gradients, air pollution plumes, or soil moisture zones), making an exact dividing line artificial.
Fields model phenomena defined everywhere in space (, e.g., elevation and temperature), while objects model discrete bounded entities (e.g., buildings and roads). Crisp boundaries represent sharp legal or physical limits, whereas fuzzy boundaries represent gradual transition zones.
Classify the following GIS attributes into nominal, ordinal, interval, or ratio measurement scales, and justify each choice: (a) Ward number (1, 2, 10), (b) Landslide hazard rank (Low, Moderate, High), (c) Mean January temperature in , (d) Annual rainfall in .
Analyze each attribute:
- (a) Ward number (
1,2,10): Nominal. Even though digits are used, they serve only as categorical labels; Ward 10 is not "five times" Ward 2, and averaging ward numbers is meaningless. - (b) Landslide hazard rank (
Low,Moderate,High): Ordinal. Categories have a clear directional order (), but the difference betweenLowandModerateis not quantitatively equal to that betweenModerateandHigh. - (c) Mean January temperature in : Interval. Equal differences () represent equal thermal intervals, so subtraction is valid, but is an arbitrary freezing-point reference rather than an absence of thermal energy (ratios like are invalid).
- (d) Annual rainfall in : Ratio. Has equal intervals and a true absolute zero ( means complete absence of rain), making multiplication and ratios valid ( is twice ).
(a) Nominal (identifier label), (b) Ordinal (ranked levels without fixed spacing), (c) Interval (equal units with arbitrary zero), (d) Ratio (equal units with true absolute zero).
A study area is covered by a -band multispectral satellite raster at spatial resolution, where each pixel value per band is stored as a -bit (-byte) unsigned integer.
- Compute the uncompressed size of this image in megabytes (, ).
- If the spatial resolution is sharpened to , find the new uncompressed size and explain the scaling factor.
Compute grid dimensions and size at resolution:
Compute grid dimensions and size at resolution: Halving the cell side length doubles both rows and columns ( pixels), increasing storage by :
At resolution, the uncompressed 4-band image occupies (). At resolution, doubling linear resolution in both axes quadruples () the pixel count to ().
Explain the DE-9IM framework and derive the -character DE-9IM string when a closed polygon strictly contains a smaller polygon (without touching its boundary).
DE-9IM framework: Every geometry is partitioned into Interior , Boundary , and Exterior . DE-9IM records the maximum spatial dimension of the intersections between and .
Evaluate all 9 intersections for strictly inside :
- Row 1 (): (2D ); (1D ); (2D ).
- Row 2 (): Since does not touch , ; ; (1D ).
- Row 3 (): ; ; (2D ).
DE-9IM evaluates the dimensions of pairwise intersections between Interior, Boundary, and Exterior sets. When strictly contains , the matrix is , giving 212FF1FF2 (satisfying Contains(A, B) and Within(B, A)).
Compare the raster and vector data models across five criteria, and explain two methods for representing temporal (spatio-temporal) change in a GIS.
Raster vs. Vector comparison (five criteria):
- Structure: Raster uses a regular cell matrix; Vector uses coordinates as points, lines, and polygons.
- Best Use: Raster excels for continuous fields (DEM, satellite imagery); Vector excels for discrete features (parcels, roads).
- Spatial Precision: Raster accuracy is bounded by cell size; Vector preserves exact coordinates at any scale.
- Analytical Strength: Raster enables fast pixel-by-pixel map algebra; Vector enables topological adjacency and network shortest-path routing.
- Storage: Raster volume depends on resolution and extent; Vector volume depends on feature and vertex count.
Two spatio-temporal representation methods:
- Snapshot Model: Stores complete layer states at discrete timestamps (); easy to inspect a fixed date, though redundant for static features.
- Timestamped / Versioning Model: Attaches
valid_fromandvalid_totemporal attributes (or event logs) to individual features so historical states and changes are queried without duplicating unchanged geometries.
Raster models continuous fields as regular grids suited to map algebra, whereas vector models discrete objects as exact points, lines, and polygons suited to topology and network analysis. Spatio-temporal change is represented using snapshot layers at discrete dates or feature-level timestamp versioning (valid_from/valid_to intervals and event histories).