In 2015 a report “Small City, Regional and Global Predictions for 2015, 2016 and 2017” was sealed and lodged by Predict Ability (PAL) in the safe of a legal firm. The report was generated programmatically, using PAL’s proprietary prediction technology: this machine-generated report was designed to identify qualifying ‘disaster events’ predicted to occur within specified time and distance windows. Once the reporting period had passed, a verification of PAL’s predictions, compared to the GLIDE online database confirmed an accuracy approaching 90% of those vulnerable cities pinpointed by PAL’s program.
As shown in the pie chart and table below, criteria for this report were set for ‘hits’ and for ‘near hits’, depending on proximity and time-line. Note that there were 160 disaster events that impacted 43 vulnerable cities. While this is a small sample of the 1000 or so disasters each year, it is nevertheless of huge importance to the millions of citizens living in those cities, as well as to UN relief agencies – for whom advance warning of where resources need to be allocated would be invaluable.
Table 1. Proximity and time-line criteria for event predictions.
|Dataset: 7A15||Event(s) occurred (within 250 km and 182 days)||Event(s) occurred (within 250 km and 365 days)||Event(s) occurred (within 500 km and 182 days)||Event(s) occurred (within 500 km and 365 days)||No event|
|Prediction outcome %||29%||8%||42%||10%||10%|
Extreme weather-related disasters occur when an ‘event’ (storm, flood, landslides, drought, forest fires…) impacts a centre of population – typically a city – that lacks resilience and hence is vulnerable.
Of course, it is well known where extreme weather-related hazards are likely to occur. A classic example is 2017’s Hurricane Irma, which rose off the west coast of Africa and crossed the Atlantic to make landfall in the Caribbean chain and the contiguous US. Together with Harvey and Maria, also in 2017, Irma made up three of the five most costly hurricanes in US history according to USA Today. The term ‘costly’ here is taken to mean the cost of loss and damage to property and infrastructure, and consequently to insurers and to government agencies. Irma, described as a ‘monster’ hurricane, caused 144 deaths overall, 100 in the US. Post Irma, the US quickly returned to business as usual. Yet, post Maria, San Juan in Puerto Rico is still devastated and the city asks, “Why don’t they help us?”
PAL’s algorithm for predicting the proximity and time-line for the ‘near where and near when’ of extreme weather disaster likelihood, has filters set to focus on those centres of population that are the most vulnerable i.e. lacking in resilience. For example, Hong Kong, a coastal city of 7 million, has high resilience. In 2003, Typhoon Azalea struck HK. This ‘double eyed monster’, as it was described at the time, was the worst storm to hit the region in 24 years. In Hong Kong, nobody died. This was not a disaster because of Hong Kong’s resilience. Taiwan and mainland China were not so fortunate.
A particular PAL case study is that of New Orleans. On back testing, New Orleans kept popping up through the filters as being at risk. Year on year the probability increased until, when Hurricane Katrina hit in 2005, the predicted probability was 86%. So even a ‘miss’ can become a ‘hit’ in due course as probability ratchets up. Significantly, for New Orleans, it is starting to ratchet up again.
Bruce Menzies, Chairman, Predict Ability Ltd (PAL)
© Copyright Predict Ability Ltd 2018. All rights reserved.
Author: Bruce Menzies
Bruce Menzies is Chairman and co-founder of PAL. He founded Global Digital Systems Ltd that won the Queen’s Award For Enterprise 2011. Bruce is co-author of six books on geotechnics and geology, one of which won the British Geotechnical Association Prize 2002. He holds doctorates from the Universities of London and Auckland, and is a Fellow of the Institution of Civil Engineers.