Skip to content

Adding News Reports

Satellite detections show you where heat is. News coverage tells you what people on the ground were dealing with, from evacuations to road closures and damage. Putting the two side by side gives a fuller picture of a fire than either one alone.

The GDELT Project builds a public database of world events from news coverage, and all of its datasets are available in Google BigQuery, with the live tables updated every 15 minutes.

Before You Run Anything

BigQuery charges by the amount of data a query reads, and GDELT is large. The GDELT team once warned that a single combined query of theirs read 221 GB, over a quarter of the monthly free BigQuery quota at the time. The events table isn't partitioned by date, so a date filter alone won't shrink the scan. Select only the columns you need, and run a dry run first to see the cost.

Query Events Inside Your Box

The query below pulls news events located inside the Big Grass Fire box between July 23 and July 31, 2026, keeping only articles whose web address mentions fire.

from google.cloud import bigquery

client = bigquery.Client()  # uses your Google Cloud project and credentials

SQL = """
SELECT SQLDATE, ActionGeo_FullName, ActionGeo_Lat, ActionGeo_Long, SOURCEURL
FROM `gdelt-bq.gdeltv2.events`
WHERE SQLDATE BETWEEN @start AND @end
  AND ActionGeo_Long BETWEEN @west AND @east
  AND ActionGeo_Lat BETWEEN @south AND @north
  AND LOWER(SOURCEURL) LIKE '%fire%'
"""

params = [
    bigquery.ScalarQueryParameter("start", "INT64", 20260723),
    bigquery.ScalarQueryParameter("end", "INT64", 20260731),
    bigquery.ScalarQueryParameter("west", "FLOAT64", -118.3),
    bigquery.ScalarQueryParameter("south", "FLOAT64", 41.9),
    bigquery.ScalarQueryParameter("east", "FLOAT64", -116.4),
    bigquery.ScalarQueryParameter("north", "FLOAT64", 44.6),
]

dry = client.query(SQL, job_config=bigquery.QueryJobConfig(query_parameters=params, dry_run=True))
print(f"This query will read {dry.total_bytes_processed / 1e9:.1f} GB")

news = client.query(SQL, job_config=bigquery.QueryJobConfig(query_parameters=params)).to_dataframe()

Install the client with pip install google-cloud-bigquery pandas db-dtypes.

Match News to Detections

GDELT places an event at a named location, often a town, rather than at the fire itself. Matching by distance, with a generous radius, works better than expecting exact overlap.

import pandas as pd
from firms import flag_within

# July detections from Pulling Fires for an Area
fires = july

news = news.rename(columns={"ActionGeo_Lat": "latitude", "ActionGeo_Long": "longitude"})
news["date"] = pd.to_datetime(news["SQLDATE"].astype(str), format="%Y%m%d").dt.date
fires["date"] = pd.to_datetime(fires["acq_date"]).dt.date

matched = []
for day, day_news in news.groupby("date"):
    day_fires = fires[fires["date"] == day]
    matched.append(flag_within(day_news, day_fires, radius_km=25))

result = pd.concat(matched, ignore_index=True)
print(result[result["within_radius"]][["date", "ActionGeo_FullName", "nearest_fire_km", "SOURCEURL"]])

Read the Output With Care

GDELT codes events from news text automatically, so a story about a fire department budget can match as easily as an evacuation order. Open the source links before drawing conclusions from them. The 25 km radius is a starting point for a rural fire, and a dense area may need a smaller one.