Wildfire prediction & management on GCP for a leading energy company

Challenge

Utilities operating in fire-prone regions face mounting pressure from regulators, insurers, and the public to reduce wildfire ignition risk. The customer needed a system that could:

• Predict ignition risk at high spatial resolution across thousands of miles of grid
• Fuse heterogeneous data: weather forecasts, vegetation indices, terrain, asset health, and historical incidents
• Update predictions in near-real-time as conditions changed
• Surface actionable decisions for de-energization and field crews — not just risk heatmaps

Existing tooling was a patchwork of GIS layers and manual analyst workflows that could not meet seasonal scaling demands.

Approach

VerticalServe stood up an AI/ML POD that delivered the full stack on Google Cloud:

• Data fabric on BigQuery and Cloud Storage, ingesting weather (NOAA, ECMWF), satellite imagery, asset telemetry, and historical incident records
• Feature engineering pipelines on Dataflow producing fire-weather indices, fuel-moisture estimates, and asset-health scores
• Risk models blending gradient-boosted classifiers with deep models on Vertex AI, evaluated with class-imbalance-aware metrics
• Geospatial scoring service that produced per-segment risk scores updated on a sub-hourly cadence
• Operations console integrating scores into the customer’s existing GIS tooling and PSPS (public-safety power shutoff) workflow
• MLOps with continuous retraining, drift detection, and model governance baked in

Outcome

The platform went live ahead of the next wildfire season. Operations teams replaced spreadsheets and manual GIS overlays with a single, governed view of risk — making PSPS decisions faster and with greater defensibility. Model performance improved iteration over iteration as new incident labels flowed back in.

Impact

Stack

GCP BigQuery Dataflow Vertex AI Cloud Storage Cloud Pub/Sub TensorFlow XGBoost GeoServer Looker

Read the full long-form on Medium