LESSON 12.4 — Decision Support Systems and Urban Analytics

A. Standard Map

B. Mechanism in Words

1. Define the decision problem — identify whether it is structured (one right answer from rules), semi-structured (competing criteria, expert judgement needed), or unstructured (no agreed framework); DSS is designed for the semi-structured class, which covers most strategic planning decisions.
2. Identify alternatives and criteria — enumerate the planning options (sites, alignments, policy instruments, investment sequences) and the criteria by which they will be evaluated; criteria must be measurable, mutually exclusive, and collectively exhaustive of planning objectives.
3. Assign weights and scores — elicit weights from stakeholders or analysts (AHP pairwise comparison is the most defensible method); score each alternative on each criterion using a consistent scale (1–5 or 0–100); standardise scores if raw data units differ across criteria.
4. Compute weighted scores — apply WSM (Weighted Sum Method): Weighted Score = Σ(wᵢ × sᵢ) for each alternative; rank alternatives by total weighted score.
5. Test sensitivity — vary the most influential weight(s) by ±10–20%; if the ranking of alternatives does not change, the conclusion is robust; if ranking flips on a small weight change, the conclusion is fragile and additional validation is required.
6. Apply scenario planning — construct three scenarios (optimistic/high-growth, baseline/medium, pessimistic/low-growth) by varying key assumptions (population projection method, land cost trajectory, infrastructure cost); assess whether the preferred alternative remains preferred across all scenarios.
7. Document and present — every MCDA output must state: the criteria, their weights, the scoring methodology, the sensitivity test result, and the scenario range; outputs without this documentation are not defensible in statutory plan proceedings.

C. Core Concept Explanations

C1. MCDA — Criteria, Weights, Scoring, Normalisation

Multi-Criteria Decision Analysis (MCDA) is the primary quantitative technique within a DSS for ranking alternatives against multiple, often conflicting, objectives. The Weighted Sum Method (WSM) is the most commonly tested MCDA variant.

Procedure:

Step 1: Define n criteria and m alternatives
Step 2: Assign weights wᵢ to each criterion (Σwᵢ = 1.00; if not, normalise)
Step 3: Score each alternative on each criterion on a common scale (1–5 or 0–100)
         Higher score = more preferred on that criterion (consistent direction)
Step 4: Compute weighted score for each alternative:
         WSᵢ = Σ(wⱼ × sᵢⱼ) for j = 1 to n criteria
Step 5: Rank alternatives by WS (highest = most preferred)

Normalisation requirement: If raw criterion data is in different units (distance in km, cost in ₹, area in ha), raw values cannot be directly multiplied by weights. Each criterion must be converted to a dimensionless score on the common scale before weighting.

Two normalisation approaches:
– Linear scaling: Score = (Value − Min) / (Max − Min) × scale_max
– If higher value = better (e.g., accessibility score), use this directly
– If lower value = better (e.g., land cost), invert: Score = (Max − Value) / (Max − Min) × scale_max
– Rank scoring: Assign ordinal ranks (1 = worst, 5 = best) based on relative performance — simpler but loses information about magnitude of differences

AHP (Analytic Hierarchy Process) — weight derivation:
AHP derives weights by asking analysts to compare criteria in pairs: “How many times more important is Criterion A than Criterion B?” The pairwise comparison matrix is normalised to yield weights that are internally consistent. AHP also computes a Consistency Ratio (CR) — values < 0.10 indicate acceptable consistency. AHP is the most defensible weight derivation method for multi-stakeholder planning decisions.

TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution):
TOPSIS ranks alternatives by their geometric distance from the ideal best solution (highest score on every criterion) and ideal worst solution (lowest on every criterion). The alternative closest to the ideal best and farthest from the ideal worst is ranked first. TOPSIS is more sophisticated than WSM but rarely appears in GATE NATs — know the concept, not the full computation.

C2. Weighted Overlay — Raster Suitability Model

Weighted overlay is the spatial implementation of MCDA within a GIS raster environment. Instead of evaluating discrete alternatives (Site A vs Site B), it evaluates every cell in a study area simultaneously, producing a continuous suitability surface.

Procedure:

Step 1: Identify criterion layers (slope, flood risk, proximity to roads,
        land cover, distance from water body, etc.)
Step 2: Convert all layers to RASTER format with identical:
        - Cell size (resolution)
        - Spatial extent
        - Coordinate reference system
Step 3: Reclassify each raster to a common ordinal scale
        (e.g., 1 = least suitable → 5 = most suitable)
        Higher value = more suitable on that criterion (consistent direction)
Step 4: Apply weighted sum using raster map algebra:
        Suitability = (w₁ × Raster₁) + (w₂ × Raster₂) + ... + (wₙ × Rasterₙ)
        where Σwᵢ = 1.00
Step 5: Classify output suitability raster into zones:
        e.g., High (4–5), Medium (3–4), Low (1–3), Excluded (masked out)

Key differences from tabular MCDA:
– Output is a continuous map (every pixel has a suitability score), not a ranked list of discrete sites
– Exclusionary constraints (flood zones, protected areas, existing development) are applied as masks before the overlay — excluded pixels receive a score of 0 regardless of other criteria
– Criteria must be in raster format; vector layers (road lines, parcel boundaries) are converted using proximity/distance rasters or rasterisation

Common planning applications:

C3. Sensitivity Analysis — Robust vs Fragile Conclusions

Sensitivity analysis tests how much the output of an MCDA or model changes when input parameters (weights, scores, assumptions) are varied. It distinguishes robust conclusions — which hold across a range of assumptions — from fragile conclusions — which reverse when a single parameter is changed.

Standard sensitivity test for MCDA:
1. Run the baseline MCDA → record the ranking of alternatives.
2. Increase the weight of the most influential criterion by Δw = 0.10 (e.g., from 0.40 to 0.50); decrease another weight proportionally to maintain sum = 1.
3. Re-run the MCDA → check if the ranking changes.
4. Repeat for each criterion individually.
5. Identify the “pivot weight” — the weight value at which the top-ranked alternative changes.

Interpretation table:

Sensitivity analysis in scenario planning:
When key model inputs are uncertain (population growth rate, infrastructure cost escalation, land market response), sensitivity analysis is applied to the entire model output, not just MCDA weights. The result is a range (band) of outcomes rather than a single projected value. Infrastructure should be sized for the upper end of the range; financial planning should use the central estimate.

C4. Lorenz Curve and Gini Coefficient — Spatial Inequality Measurement

The Lorenz curve (Max Lorenz, 1905) visualises the distribution of any quantity across a population or set of spatial units. In urban planning, it measures inequality in access to services, open space, infrastructure, or income across wards, zones, or population groups.

Construction of the Lorenz curve:

Step 1: Order all spatial units (wards/zones) from lowest to highest
        value of the variable per capita (open space/person, income/household, etc.)
Step 2: Compute cumulative % of population (x-axis)
Step 3: Compute cumulative % of the variable (y-axis)
Step 4: Plot (x, y) pairs; add origin (0, 0) and endpoint (100, 100)
Step 5: The line of perfect equality is the 45° diagonal from (0,0) to (100,100)
        The Lorenz curve lies below this diagonal; greater bowing = greater inequality

Gini coefficient formula:

Gini = A / (A + B)

Where:
  A = area between the line of equality and the Lorenz curve
  B = area below the Lorenz curve
  A + B = total area below the line of equality = 0.5 × base × height
          (in percentage units: = 0.5 × 100 × 100 = 5,000)

Practical computation using trapezoid method:
  B = Σ [(xᵢ − xᵢ₋₁) × (yᵢ + yᵢ₋₁) / 2]
  A = 5,000 − B
  Gini = A / 5,000

Gini = 0 → perfect equality (every ward has identical per-capita access)
Gini = 1 → perfect inequality (one ward holds all resources; rest have none)

Interpretation benchmarks (planning context):

Planning applications of Gini:
– Open space Gini: Distribution of park area per person across wards — a high value (e.g., 0.60) demonstrates that parks are clustered in high-income areas.
– Infrastructure Gini: Distribution of water supply hours per day, drainage coverage, or road length per person — used to justify investment in infrastructure-deficit wards.
– Income Gini: Distribution of household income across zones — used to identify affordable housing demand clusters and inform housing cross-subsidy policy.

C5. Scenario Planning — High/Medium/Low; Stress Testing

Scenario planning acknowledges that the future is uncertain and constructs multiple plausible futures rather than a single forecast. It provides a policy-testing framework: a plan that performs acceptably across all scenarios is more defensible than one optimised for a single forecast that may not materialise.

Three-scenario structure:

Scenario construction steps:
1. Identify the two or three key uncertainties that most affect the plan outcome (population growth rate, economic base composition, infrastructure cost escalation).
2. Define plausible ranges for each uncertainty — not best/worst guesses, but internally consistent combinations.
3. Run each scenario through the demand model (population × per-capita demand) to get service demand ranges.
4. Test the preferred planning alternative (from MCDA) against each scenario.
5. Report the scenario band for each key output variable.

Stress testing: A specific type of scenario analysis that applies an extreme-but-plausible shock (economic crash, climate event, pandemic demand disruption) to test whether the plan remains functional under worst-case conditions. Any plan element that fails under a stress test requires either a design modification or an explicit risk management provision.

PERT/CPM cross-reference: Schedule uncertainty in implementation programmes (quantified through PERT expected times and variances — covered fully in Ch 2.4) feeds directly into the scenario planning timeline; construction delay risk shifts the scenario band for when infrastructure becomes operational. Do not replicate CPM network calculations here.

C6. Scenario Planning — How it Connects to MCDA and Sensitivity

MCDA and scenario planning operate at different levels of the analytical hierarchy:

MCDA → answers: "Given ONE future, which alternative is best?"
Sensitivity analysis → answers: "Does the MCDA ranking change when weights shift?"
Scenario planning → answers: "Does the best alternative remain best across DIFFERENT futures?"

A complete DSS workflow integrates all three:
1. Run MCDA for the baseline scenario → preferred alternative identified.
2. Run sensitivity test on MCDA weights → confirm robust ranking.
3. Run the preferred alternative through high/medium/low scenarios → confirm the alternative remains performant.
4. If it passes all three tests, the recommendation has high analytical credibility.

D. Worked Numericals and Parameter Tables

Section D contains two worked problems: a Gini NAT and an MCDA weighted-score NAT.

D1 — Worked NAT: Gini Coefficient for Open Space Inequality

Problem: A city is divided into 4 wards. The planning department has
recorded the following data. Calculate the Gini coefficient for
open space (park area) distribution across the city's population.

| Ward | Population | Park Area (ha) |
|------|-----------|----------------|
| W1   | 25,000    | 2.5            |
| W2   | 25,000    | 3.75           |
| W3   | 25,000    | 7.5            |
| W4   | 25,000    | 11.25          |
| Total| 1,00,000  | 25.0           |

Step 1 — Park area per capita (to determine ordering):
  W1: 2.5 ha / 25,000 = 1.00 sq m/person
  W2: 3.75 ha / 25,000 = 1.50 sq m/person
  W3: 7.5 ha / 25,000 = 3.00 sq m/person
  W4: 11.25 ha / 25,000 = 4.50 sq m/person

  Order ascending (worst → best per-capita access):
  W1 (1.0) → W2 (1.5) → W3 (3.0) → W4 (4.5)

Step 2 — Compute cumulative population % and cumulative park area %:
  Each ward = 25,000/1,00,000 = 25% of total population

| Cumulative step | Cum. pop % (x) | Cum. park % (y) |
|----------------|----------------|-----------------|
| Origin          | 0              | 0               |
| After W1        | 25             | 2.5/25.0 × 100 = 10.0 |
| After W1+W2     | 50             | 6.25/25.0 × 100 = 25.0 |
| After W1+W2+W3  | 75             | 13.75/25.0 × 100 = 55.0 |
| All 4 wards     | 100            | 100             |

Lorenz curve points: (0,0), (25,10), (50,25), (75,55), (100,100)

Step 3 — Compute area B under Lorenz curve using trapezoid method:
  B = Σ [(xᵢ − xᵢ₋₁) × (yᵢ + yᵢ₋₁) / 2]

  T1: (25 − 0) × (10 + 0) / 2 = 25 × 5 = 125.0
  T2: (50 − 25) × (25 + 10) / 2 = 25 × 17.5 = 437.5
  T3: (75 − 50) × (55 + 25) / 2 = 25 × 40 = 1000.0
  T4: (100 − 75) × (100 + 55) / 2 = 25 × 77.5 = 1937.5

  B = 125.0 + 437.5 + 1000.0 + 1937.5 = 3500.0

Step 4 — Compute A and Gini:
  Total area below diagonal = (100 × 100) / 2 = 5000.0
  A = 5000.0 − 3500.0 = 1500.0
  Gini = A / (A + B) = 1500.0 / 5000.0

┌────────────────────────────────────────┐
│  Gini coefficient = 0.30               │
│  (Moderate inequality in park access)  │
└────────────────────────────────────────┘

Interpretation: Gini = 0.30 indicates moderate inequality.
The worst-served ward (W1) has 10% of the park area serving 25% of
the population. Targeted investment in W1 and W2 is warranted.

D2 — Worked NAT: MCDA Weighted Score for Site Selection

Problem: A planning authority is evaluating three sites (A, B, C)
for a new district hospital. Four criteria are used with the
weights and scores shown below (1–5 scale; 5 = most preferred).
Determine the preferred site.

| Criterion          | Weight (wᵢ) | Score: Site A | Score: Site B | Score: Site C |
|--------------------|------------|--------------|--------------|--------------|
| Accessibility       |   0.40     |      4       |      3       |      5       |
| Environmental risk  |   0.25     |      3       |      5       |      2       |
| Land cost           |   0.20     |      5       |      3       |      2       |
| Infrastructure cap. |   0.15     |      3       |      4       |      4       |
| Sum of weights      |   1.00     |              |              |              |

Step 1 — Verify weight sum: 0.40 + 0.25 + 0.20 + 0.15 = 1.00 ✓

Step 2 — Compute weighted score for each site:

  Site A: (0.40×4) + (0.25×3) + (0.20×5) + (0.15×3)
        =  1.60  +  0.75  +  1.00  +  0.45  = 3.80

  Site B: (0.40×3) + (0.25×5) + (0.20×3) + (0.15×4)
        =  1.20  +  1.25  +  0.60  +  0.60  = 3.65

  Site C: (0.40×5) + (0.25×2) + (0.20×2) + (0.15×4)
        =  2.00  +  0.50  +  0.40  +  0.60  = 3.50

Step 3 — Rank by weighted score:
  Site A = 3.80 (1st) → Site B = 3.65 (2nd) → Site C = 3.50 (3rd)

┌────────────────────────────────────────┐
│  Preferred site: A (WS = 3.80)         │
└────────────────────────────────────────┘

Step 4 — Quick sensitivity test:
  If weight of Accessibility rises from 0.40 to 0.50
  (and Environmental risk falls from 0.25 to 0.15):

  Site A: (0.50×4)+(0.15×3)+(0.20×5)+(0.15×3) = 2.0+0.45+1.0+0.45 = 3.90
  Site B: (0.50×3)+(0.15×5)+(0.20×3)+(0.15×4) = 1.5+0.75+0.60+0.60 = 3.45
  Site C: (0.50×5)+(0.15×2)+(0.20×2)+(0.15×4) = 2.5+0.30+0.40+0.60 = 3.80

  Site A still leads (3.90 vs 3.80 vs 3.45). Conclusion is ROBUST
  under this weight perturbation.

Note: If the questioner asks for the weighted score of Site A only,
the answer is 3.80.

E. Common Confusions

• GIS ≠ DSS. GIS answers spatial questions (“where are flood-prone areas?”). A DSS adds decision logic to evaluate alternatives (“which site should receive the housing investment?”). A planning DSS incorporates GIS but also has a model base, knowledge base, and user interface — GIS alone lacks these.
• MCDA weight sum must equal 1. If weights are given as 40, 30, 20, 10 (appearing to sum to 100 but expressed as integers), divide by 100 before computing — use 0.40, 0.30, 0.20, 0.10. Never multiply raw scores by un-normalised weights.
• Weighted overlay requires rasterisation before weighting. Vector layers (road buffers, flood zone polygons) must be converted to raster with the same cell size and extent as other layers before they can be entered into the weighted sum. This is a pre-processing requirement, not optional.
• Gini = 0 is perfect equality, not perfect inequality. The most common reversal error in exams: Gini = 0 → zero inequality (all units have identical per-capita share); Gini = 1 → maximum inequality (one unit holds everything). The Lorenz curve at Gini = 0 coincides with the 45° diagonal.
• Sensitivity analysis ≠ scenario planning. Sensitivity analysis varies model inputs within a single alternative or model to test output stability. Scenario planning constructs alternative futures (different combinations of key assumptions) and tests whether the preferred plan performs acceptably in each. They answer different questions and complement each other.
• TOPSIS ≠ WSM. WSM (Weighted Sum) ranks by total weighted score directly. TOPSIS ranks by distance from ideal best/worst solutions — it can give different rankings than WSM, especially when one alternative dominates on some criteria and is very weak on others.

F. Exam Traps

G. Answer-Writing Cues

NAT — Gini calculation (show all steps):

MCQ — DSS vs GIS distinction:

MCQ — Sensitivity analysis interpretation:

MSQ / written — Scenario planning justification:

H. PYQ Linkage Note

I. Mini-Check — Lesson 12.4

Q1. (NAT) A city’s park area is distributed across three wards as follows (wards already ordered from worst to best per-capita access):

Calculate the Gini coefficient for park area distribution. (Express as a decimal to 2 decimal places.)

Solution:

B = Σ [(xᵢ − xᵢ₋₁) × (yᵢ + yᵢ₋₁) / 2]

T1: (25 − 0) × (5 + 0) / 2 = 25 × 2.5 = 62.5
T2: (50 − 25) × (15 + 5) / 2 = 25 × 10 = 250.0
T3: (75 − 50) × (40 + 15) / 2 = 25 × 27.5 = 687.5
T4: (100 − 75) × (100 + 40) / 2 = 25 × 70 = 1750.0

B = 62.5 + 250.0 + 687.5 + 1750.0 = 2750.0
A = 5000.0 − 2750.0 = 2250.0
Gini = 2250.0 / 5000.0

Interpretation: Gini = 0.45 indicates high-moderate inequality — the bottom 25% of the population by park access holds only 5% of the park area. Targeted open space investment in underserved wards is required.

Q2. (MSQ — select ALL correct) A planning authority is setting up a Decision Support System (DSS) to evaluate alternative routes for a new metro corridor. Which of the following statements about the DSS methodology are correct?

(A) The DSS evaluates alternatives against weighted criteria — this is a GIS-only function that does not require a model base
(B) Weights assigned to criteria must sum to 1.00 (or 100%) before scores are computed
(C) If the ranking of alternatives reverses when one criterion’s weight is changed by only 5%, the MCDA conclusion is considered fragile
(D) A GIS suitability map showing corridor potential is a sufficient output — no further MCDA or sensitivity analysis is needed
(E) Three scenarios (high-growth, baseline, low-growth) should be tested to check whether the preferred corridor performs acceptably under different demand assumptions

Answer: B, C, E

Explanation: (A) Incorrect — DSS includes a model base, knowledge base, and decision logic; GIS alone cannot evaluate and rank alternatives with weights. (B) Correct — MCDA weights must sum to 1.00; un-normalised weights corrupt weighted scores. (C) Correct — a <10% weight shift causing a rank reversal classifies the conclusion as fragile. (D) Incorrect — a suitability map is one analytical input; it must be supplemented by MCDA, sensitivity analysis, and scenario testing before a recommendation is defensible. (E) Correct — scenario planning tests plan robustness across multiple plausible futures.

Q3. (MCQ) A Lorenz curve for water supply access per household across city wards closely follows the 45° line of equality. The Gini coefficient for this distribution is most likely:

(A) 0.70
(B) 0.50
(C) 0.30
(D) 0.05

Answer: (D) 0.05

Explanation: When the Lorenz curve closely follows the 45° diagonal (line of perfect equality), there is very little area between the curve and the diagonal. Area A → 0, so Gini = A/(A+B) → 0. A Gini near 0.05 indicates near-perfect equality in water access distribution across wards — all wards have roughly the same per-household supply. Options A (0.70) and B (0.50) would indicate high inequality with the Lorenz curve bowing far below the diagonal.

Q4. (MCQ) In the weighted overlay raster suitability analysis, an analyst forgets to reclassify the slope raster (raw values: 0–45 degrees) to a 1–5 suitability scale before combining it with other reclassified criterion rasters. The most likely consequence is:

(A) The slope criterion will have no influence on the final suitability score
(B) The slope criterion will dominate the composite suitability score because its raw values are far larger than the 1–5 scores of other criteria
(C) The analysis will produce an error message and refuse to compute
(D) The composite suitability surface will have higher resolution than intended

Answer: (B)

Explanation: Raw slope values (0–45 degrees) are of a completely different magnitude than reclassified 1–5 scores from other criteria. When multiplied by their respective weights and summed, the slope criterion’s large values will dwarf the contributions of other criteria (whose scores range only 1–5), effectively making slope the dominant driver of the composite score regardless of its assigned weight. This is the cardinal error of weighted overlay without prior standardisation. The analysis will compute — it will not crash — it will simply produce a meaningless output that misrepresents the intended multi-criteria balance.

Q5. (MCQ) In a scenario planning framework for a metropolitan master plan, which set of inputs is MOST appropriate for the “low/pessimistic” scenario?

(A) Geometric population projection; low infrastructure cost escalation; high economic growth
(B) Arithmetic population projection; high infrastructure cost escalation; below-trend economic growth
(C) Incremental increase population projection; stable infrastructure costs; average economic growth
(D) Logistic population projection; medium cost escalation; average economic growth

Answer: (B)

Explanation: The low/pessimistic scenario is designed to stress-test the plan under adverse conditions. Arithmetic projection gives the lowest population growth assumption (conservative demand side); high infrastructure cost escalation compresses the plan’s financial viability (supply-side pressure); below-trend economic growth reduces revenue projections and public sector investment capacity. Together, these represent a coherent pessimistic future that tests whether the plan remains financially feasible and physically adequate even when conditions are unfavourable. Option (A) is an optimistic scenario. Option (C) is a baseline/central scenario. Option (D) uses logistic — inappropriate unless carrying capacity is explicitly constrained.