carto-hotspot-analysis

Hotspot Analysis with Getis-Ord Gi*

Builds CARTO Workflows that identify statistically significant spatial clusters (hotspots and coldspots) using the Getis-Ord Gi* statistic.

Prerequisites: Load carto-create-workflow for the development process, JSON structure, and validation commands — it covers both access paths (the MCP server's workflow tools such as create_workflow, validate_workflow, and run_workflow when attached; the carto workflows CLI otherwise; routing signals in carto-basics/references/access-paths.md).

Instructions

A hotspot workflow always follows this pipeline:

Source Data → (Filter) → Spatial Indexing → Aggregation → Getis-Ord Gi* → (Filter Significant) → Save

Step 1: Load Source Data

Use native.gettablebyname. The input table typically contains point geometries.

Success: Node outputs a table with a geometry column (e.g. geom).

Step 2: Filter (if needed)

Use native.wheresimplified or native.where to narrow the dataset before analysis (e.g. filter by category, date range, non-null values).

Success: Output contains only the subset relevant to the analysis.

Step 3: Build a Complete Grid

Preferred approach: First polyfill the study area boundary (e.g. district polygons) with native.h3polyfill to create a complete, gap-free grid. Then enrich this grid with the data to analyze (e.g. count points per cell via native.h3enrich or a manual join + group by). This ensures every cell in the study area has a value (even if 0), which Getis-Ord needs — gaps in the grid distort the neighborhood calculations and can produce misleading results.

Simpler alternative (when no study area boundary is available): Convert point geometries directly to grid cells with native.h3frompoint or native.quadbinfromgeopoint. Be aware this only produces cells where data exists, leaving gaps that may affect the statistic.

Resolution guidance — higher resolution = smaller cells = more local patterns:

Success: A contiguous grid covering the study area, with every cell assigned a spatial index column (e.g. h3).

Step 4: Aggregate per Cell

Use native.groupby to produce one row per cell with a numeric value:

• Group by: the spatial index column (h3)
• Aggregation: h3,count (or value_col,sum / value_col,avg)

If using the polyfill approach, cells with no data should have a value of 0 (use COALESCE(count, 0) via native.selectexpression after joining).

Success: Output has exactly one row per unique cell with a count/sum column — no gaps.

Step 5: Run Getis-Ord Gi*

Use native.getisord with:

Kernel options: uniform, triangular, quadratic, quartic, gaussian. Default to uniform (equal weight to all neighbors) unless the user has a reason to decay weight with distance.

K-ring size: Larger = smoother, broader patterns. Smaller = more localized clusters.

Success: Output contains index, gi (z-score), and p_value columns for every cell. (See the Provider casing note in Gotchas — Snowflake surfaces these UPPERCASE.)

Step 6: Filter Significant Results (optional)

Use native.where to keep only statistically significant cells:

• p_value < 0.05 — 95% confidence
• p_value < 0.05 AND gi > 0 — hotspots only
• p_value < 0.05 AND gi < 0 — coldspots only

Success: Only cells with statistically meaningful clustering remain.

Step 7: Save

Use native.saveastable to persist results. The H3/Quadbin column is directly visualizable in CARTO Builder without geometry conversion.

Success: Validated workflow that can be uploaded via create_workflow (MCP) or carto workflows create (CLI).

Output Columns

Gotchas

• Provider casing & SQL dialect. This skill documents columns in lowercase (BigQuery / Databricks / Postgres / Redshift convention). On Snowflake, unquoted identifiers surface UPPERCASE — reference H3, INDEX, GI, P_VALUE, H3_COUNT in expressions. For dialect-specific SQL fragments (e.g. DATETIME_TRUNC below), see carto-create-workflow/references/providers/<provider>.md for the equivalents table.
• The Getis-Ord component requires the Analytics Toolbox. Offline validation (validate_workflow over MCP, or carto workflows validate) can't resolve the AT location — confirm it with the CLI-only carto workflows verify-remote --connection <conn> before running.
• The output column is named index, not h3 or quadbin. If you need to join back to original data, rename it (e.g. with native.renamecolumn).
• If you call native.h3boundary to materialize cell geometries for visualization, the new column is named <h3col>_geo (e.g. index_geo), not geom. Reference it accordingly in downstream nodes.
• The valuecol must be numeric. If you're counting features, the group-by step must produce a count column — don't pass the raw index column as the value.
• Resolution too high + large area = very many cells, which can be slow or hit memory limits. Start with a moderate resolution and refine.
• An empty result from the filter step (Step 6) usually means the k-ring size is too small or the data is too sparse for significant clustering. Try increasing size or lowering the resolution.
• Date columns must be DATETIME type for spacetime Getis-Ord. CAST if your data has DATE or TIMESTAMP.
• Temporal bandwidth choice dramatically affects results. bandwidth=1 detects rapid changes; bandwidth=3+ smooths over longer trends.
• For time-series clustering, pre-filter to only significant cells (the 60% heuristic) to avoid clustering noise.
• The spacetime classification component runs internally on the Gi* output -- do NOT filter by p_value before classification, or the trend test will have incomplete data.

Spacetime Variants

Getis-Ord Spacetime (native.getisordspacetime):

• Extends basic Gi* to detect clusters in both space AND time.
• Additional inputs: kerneltime (uniform/gaussian), bandwidth (number of time steps), timeinterval (week/month/day).
• Data must be pre-aggregated into time bins (e.g. weekly counts per H3 cell).
• Pipeline: points -> H3 -> create time column (BigQuery: DATETIME_TRUNC(CAST(datetime AS TIMESTAMP), WEEK); Snowflake / Databricks / Postgres: DATE_TRUNC('WEEK', datetime)) -> GROUP BY (h3, timebin) -> Getis-Ord Spacetime -> filter `pvalue < 0.05 AND gi > 0`.

Spacetime Hotspot Classification (native.spacetimehotspotsclassification):

• Chains AFTER Getis-Ord Spacetime output.
• Classifies each cell's temporal trend: new hotspot, consecutive, intensifying, diminishing, sporadic, oscillating, historical.
• Uses Modified Mann-Kendall trend test with a significance threshold (default 0.05).
• Pipeline: ... -> Getis-Ord Spacetime -> Spacetime Hotspots Classification.

Time Series Clustering (native.timeseriesclustering):

• Groups locations by similarity of their temporal Gi* pattern.
• Chain: Getis-Ord Spacetime -> filter significant cells -> Cluster Time Series.
• Method: profile (shape-based) or value (magnitude-based).
• Filtering heuristic from the template: keep cells where >=60% of time steps have p_value < 0.05.

Reference Templates

These files are working examples (skill-local, in this skill's folder):

Common Variations