Controls
Vector Index Areas
District-Wide MLE Trend
District-Wide MIR Trend
Abundance by Area Over Time
District-Wide Vector Index Trend
Vector Index by Area Over Time
Area Data Table
Analysis Controls
What This Does
Scoring: Rates each trap's WNV detection value (yield, pools, positivity, consistency).
Risk Surface: Interpolates trap scores into an area-wide WNV risk heatmap.
Coverage: Grades each VI area's surveillance coverage (Good/Adequate/Thin/Gap).
Causal Factors: Identifies which trap features drive performance.
Based on Chakravarti et al. (2026):
'A novel approach to determine mosquito trap placement for WNV surveillance'
J. Med. Entomol. 63(2)
Spatial Risk & Trap Scores
Area Coverage Summary
Factor Importance
Score Distribution
Summary
Dose-Response: Pools → Score
Dose-Response: Detection Rate → Score
Trap Scorecard & Recommendations
Data Sources & Methods
About This Dashboard
This dashboard displays mosquito surveillance metrics aggregated by Vector Index Area (VI Area) — geographic zones used for surveillance analysis across the MMCD district.
The Causal Analysis tab implements an analytical framework inspired by Chakravarti et al. (2026) : 'A novel approach to determine mosquito trap placement for WNV surveillance' — J. Med. Entomol. 63(2). The implementation adapts the paper's concepts to MMCD's data and operational context.
Data Sources
- Abundance: dbadult_mon_nt_co2_forvectorabundance — Monday night CO2 trap counts by species, location, and week.
- MLE: dbvirus_mle_yrwk_area / dbvirus_mle_yrwk_area_spp — Pre-calculated Maximum Likelihood Estimates of infection rate.
- MIR: dbvirus_mir_yrwk_area / dbvirus_mir_yrwk_area_spp — Pre-calculated Minimum Infection Rates.
- Virus Pools: dbadult_insp_current, dbvirus_pool, dbvirus_pool_test — Pool collection and WNV testing results per trap.
- Trap Locations: loc_mondaynight — Trap coordinates (PostGIS geometry).
- Geometry: loc_vectorindexareas_sections_a — Section polygons dissolved into 12 VI areas.
Core Metrics (Map & Trend Tabs)
Abundance (N)
Average mosquitoes per trap per night for the selected week and species.
N = Total Count / Number of Traps
MLE (Maximum Likelihood Estimate)
Statistical estimate of the probability a single mosquito is infected, calculated from pooled testing data using bias-reduced methods.
Pre-calculated in the database using the Firth method (bias-reduced MLE).
MIR (Minimum Infection Rate)
Simple proportion: (positive pools / total mosquitoes tested) × 1000
MIR = (Positives / Total Mosquitoes) × 1000
Vector Index (VI = N × P)
Combined measure of mosquito abundance and infection risk.
VI = N × P
Where N = average mosquitoes/trap, P = infection rate (MLE or MIR/1000)
Causal Analysis Methods
The Causal Analysis tab runs four linked analyses in sequence. Full equations with derivations are in the TRAP_ANALYSIS_EQUATIONS.md document.
1. Trap Performance Scoring
Each trap is scored on a composite performance metric (0–1) based on four weighted factors:
| Component | Weight | Description |
|---|---|---|
| Yield Score | 25% | Avg mosquitoes per week, normalized by max across all traps |
| Testing Score | 30% | Total WNV pools collected, normalized by max |
| Detection Score | 30% | Pool positivity rate (% positive), normalized by max |
| Consistency Score | 15% | Weeks active, normalized by max |
composite = 0.25×yield + 0.30×testing + 0.30×detection + 0.15×consistency
Tiers: High (≥ 0.6) | Medium (0.3–0.6) | Low (< 0.3)
Weights reflect the paper's finding that testing volume and detection are the strongest predictors of trap utility.
2. Spatial Risk Smoothing
Computes a spatial risk score for each trap using exponential kernel smoothing of neighboring trap signals. This approximates the spatial correlation structure (Matérn covariance) used in the paper's GLMM, using Haversine distances between traps.
For each trap i , the spatial risk is a weighted average of neighbor signals:
spatial_risk_i = Σ(w_j × signal_j) / Σ(w_j)
Where:
-
w_j = exp(-dist_ij / bandwidth)— Exponential decay kernel (bandwidth default 3 km) -
signal_j = 0.6×detection + 0.4×testing— Neighbor's detection signal -
dist_ij— Haversine (great-circle) distance in km between traps i and j
The combined Risk Index integrates the trap's own metrics with its spatial context:
risk_index = 0.35×yield + 0.30×spatial_risk + 0.35×detection
Tiers: High Risk (≥ 0.5) | Moderate (0.2–0.5) | Low (< 0.2)
3. Risk Surface & Coverage Assessment
The Risk Surface interpolates trap-level risk scores onto a regular grid (~500m resolution) spanning the district, visualizing WNV risk between trap locations.
grid_risk(x,y) = Σ(w_j × risk_index_j) / Σ(w_j)
Grid cells farther than 2× bandwidth from any trap are flagged as coverage gaps .
Area Coverage Grades aggregate trap metrics per VI area:
| Grade | Criteria |
|---|---|
| ✅ Good | ≥ 10 traps AND avg score ≥ 0.3 |
| ⚠️ Adequate | ≥ 5 traps AND avg score ≥ 0.2 |
| ⚠️ Thin | ≥ 3 traps (any avg score) |
| ❌ Gap | Fewer than 3 traps |
4. Causal Factor Analysis
Identifies which trap-level variables have the strongest association with performance using Spearman rank correlations and binned dose-response curves (approximating ADRFs from the paper).
Factors analyzed:
- Pools Collected — total WNV pools submitted
- Positive Pools — number of pools testing positive
- Avg Yield/Week — average mosquitoes captured per collection week
- Weeks Active — number of distinct collection weeks
- Total Mosquitoes — total specimens collected
Key finding from the paper: Three variables showed significant causal effects: (1) total population in 1,500m buffer (not available in MMCD data), (2) number of pools collected , and (3) number of positive pools . Landcover and socioeconomic variables did NOT show significant direct causal effects.
Per-trap recommendations are generated based on score thresholds and factor values.
Species
- Total Cx Vectors: Sum of all Culex species — uses area-level MLE (all species pooled together).
- Cx. pipiens (33): Primary WNV vector in urban areas.
- Cx. restuans (34): Early-season WNV vector.
- Cx. tarsalis (36): Efficient WNV vector, more rural.
- Cx. restuans/pipiens (372): Unresolved Culex identification.
Temporal Resolution
Data is organized by epiweek (yrwk) — epidemiological weeks. Each yrwk value encodes year and week number (e.g., 202537 = year 2025, week 37).
Reference
Chakravarti, Li, Bartlett, Irwin, Smith (2026). 'A novel approach to determine mosquito trap placement for West Nile virus surveillance.' Journal of Medical Entomology, 63(2).