{
    "success": true,
    "data": {
        "id": 1612665,
        "msgid": "googles-ai-system-can-predict-flash-floods-24-hours-in-advance-1773460456",
        "date": "2026-03-14 10:03:00",
        "title": "Google's AI System Can Predict Flash Floods 24 Hours in Advance",
        "author": "Reska K. Nistanto",
        "source": "KOMPAS",
        "tags": "",
        "topic": "Technology",
        "summary": "Google has introduced Groundsource, an artificial intelligence system powered by its Gemini language model that can predict flash flooding up to 24 hours before it occurs across more than 150 countries. The technology analysed approximately 5 million global news articles dating back to 2000 to extract and structure flood event data, creating a dataset of 2.6 million flood records that enables the model to recognise flood patterns by incorporating weather forecasts, urbanisation rates, soil absorption capacity, and topography. The predictions, available through Google's Flood Hub platform, address one of the world's deadliest weather disasters, which claims over 5,000 lives annually, by providing early warning systems for urban areas with population densities exceeding 100 people per square kilometre.",
        "content": "<p>Google has introduced a new artificial intelligence system called\nGroundsource designed to predict the risk of flash flooding up to 24\nhours before it occurs.<\/p>\n<p>The technology was developed using Google\u2019s Gemini AI model, which\nwas utilised to analyse millions of flood reports from news sources\nworldwide and generate predictions for flood-prone areas in more than\n150 countries.<\/p>\n<p>According to Gila Loike, a product manager at Google Research, this\nmarks the first time Google has used its large language model (LLM) to\ncreate a scientific dataset from globally-scaled text reports.<\/p>\n<p>Flash flooding is one of the deadliest weather disasters in the\nworld, claiming an estimated 5,000 lives annually. However, flash floods\nare also among the most difficult to predict. Unlike river flooding,\nwhich can be monitored using water level sensors, urban flash floods\ntypically occur very rapidly when heavy rainfall overflows roads and\ndrainage systems.<\/p>\n<p>The challenge has been that historically, there has been limited\ndetailed data recording flash flood events. Without such data,\nscientists have found it difficult to train AI models to recognise the\npatterns that trigger these disasters.<\/p>\n<p>Gemini was used to examine approximately 5 million global news\narticles from 2000 onwards. The AI extracted information about flood\nevents from these articles and converted them into structured data\ncontaining the location and timing of each occurrence.<\/p>\n<p>Following filtering, deduplication, and translation from various\nlanguages, the millions of news reports were ultimately transformed into\na global flood dataset containing approximately 2.6 million flood event\nrecords from over 150 countries.<\/p>\n<p>The model combines various data points, including hourly weather\nforecasts, urbanisation levels in specific areas, soil water absorption\ncapacity, and topographical conditions.<\/p>\n<p>Using this data, the AI generates flash flood risk alerts for the\nfollowing 24 hours in urban areas with population densities exceeding\n100 people per square kilometre.<\/p>\n<p>Google\u2019s Groundsource flash flood predictions are made available\nthrough the Flood Hub platform. This Google service previously provided\nriver flood warnings to approximately 2 billion people globally. Users\ncan access Flood Hub directly online, search for specific countries or\ncities, and check whether flash flood predictions are in place for the\nfollowing 24 hours.<\/p>",
        "url": "https:\/\/jawawa.id\/newsitem\/googles-ai-system-can-predict-flash-floods-24-hours-in-advance-1773460456",
        "image": ""
    },
    "sponsor": "Okusi Associates",
    "sponsor_url": "https:\/\/okusiassociates.com"
}