MASSACHUSETTS · MUNICIPAL SURVEILLANCE · TEMPORAL HOLDOUT

Where and when is WNV being detected in Massachusetts?

A unified view of reported-positive mosquito surveillance, environmental conditions, and model-based predictions across Massachusetts municipalities.

REPORTED-POSITIVE SAMPLES—selected month
MUNICIPALITIES WITH DETECTIONS—in the selected year/month range
FIRST DETECTION—selected year
MOST RECENT DETECTION—selected year
01 · OBSERVED DETECTIONS

Follow detections through the season.

Play the monthly sequence, select any June–September range, filter a county, or click multiple municipalities to compare their combined reports with their counties.

Year range
to
2026September

Select a start and end month; the map sums reported-positive samples across the range.

KEY TAKEAWAY

Loading the selected observation period…

Interpret with care: Counts reflect reported-positive mosquito samples and are influenced by where and when surveillance occurred. “0 reported” does not confirm that sampling occurred.

02 · ENVIRONMENTAL RELATIONSHIPS

Explore descriptive associations.

Compare monthly environmental conditions and previous-year landscape characteristics with reported-positive samples. The year and county filters remain linked to the observation map.

KEY TAKEAWAY

Loading environmental relationships…

TOWN LEVEL · SELECTED RELATIONSHIP

Town environmental predictor and WNV reports

rₛ —p —Significance —n —
COUNTY LEVEL · SELECTED RELATIONSHIP

County environmental predictor and WNV reports

rₛ —p —Significance —n —

*** p < .001 · ** p < .01 · * p < .05 · ns not significant. Landscape factors use the previous year’s NLCD; county landscape values are area-weighted town summaries. Associations are descriptive; repeated observations, spatial dependence, correlated predictors, and unknown surveillance effort affect interpretation.

TOWN LEVEL · PREDICTOR COMPARISON

Town correlation ranking

COUNTY LEVEL · PREDICTOR COMPARISON

County correlation ranking

03 · MODEL PREDICTIONS AND EVALUATION

Forecast first. Validate before interpretation.

The two-stage Random Forest uses previous-month environmental conditions, prior-year landscape, municipality area, and county identity. Choose a year to inspect its untouched rolling temporal holdout.

Holdout periodSeptember 2026
Decision threshold—
Selected model—
2026June–September

Prediction controls and selected towns are independent from Observations and Correlations.

Predictions represent reported-detection suitability, not mosquito prevalence or human disease risk. Town and county identity can capture persistent geography and surveillance patterns, but they do not measure surveillance effort directly.

AVAILABLE PREDICTORSPrevious-month environment
Prior-year landscape
Area + county proxy
→
STAGE 1Detection Random ForestProbability of a reported-positive month
×
STAGE 2Conditional-count Random ForestCount when a positive report occurs
→
OUTPUTExpected reported-positive countProbability × conditional count

Permutation importance is calculated only on the 2026 holdout.

PRESENCE MODEL

Permutation importance

CONDITIONAL COUNT MODEL

Permutation importance

2026 MODEL ABLATION

What do landscape and geographic proxies add?

ERROR INSPECTION

Largest 2026 disagreements

TownMonthObservedExpectedClass
04 · READING GUIDE

What this dashboard can—and cannot—say.

OUTCOME

Reported-positive detection

The outcome is the number of reported WNV-positive mosquito samples in each municipality. A zero is not proof that the municipality was sampled and found negative.

TWO-STAGE MODEL

Presence, then count

The expected count multiplies the modeled presence probability by the predicted count conditional on reported-positive presence.

PREDICTORS

Available before prediction

Each June–September prediction uses the preceding month’s environmental conditions, prior-year landscape, municipality area, and county identity. No target-month environmental value is used.

LIMIT

Relative suitability

Unknown surveillance effort prevents interpretation as absolute WNV prevalence, mosquito abundance, or human disease risk.

Loading validated dashboard data…