Each year, landslides (rocks, soil, and debris down the slopes) kill thousands, cost billions of dollars, and disrupt roads and power lines. Topography, rock and soil characteristics, weather and climate all contribute to landslide activity, making it difficult to pinpoint the most dangerous locations of these hazards at any given time. Early warning systems are generally regional based on region-specific data provided by ground sensors, field observations, and total rainfall. But what if you could always identify an endangered region anywhere in the world?
Participate in NASA’s Global Landslide Risk Assessment (LHASA) model And mapping tools.
LHASA version 2 released last month, along with the corresponding research Machine learningA model based on analyzing a collection of individual variables and satellite-derived datasets to generate a customizable “nowcast”. These timely and targeted nowcasts are potential predictions. Landslide Near real-time activity in each 1 square kilometer area between the poles. The model considers land slopes (higher slopes are more likely to cause landslides), distance to geological faults, rock composition, past and present precipitation, and data on soil moisture and snow from satellites. ..
“This model processes all the data and outputs probabilistic estimates of landslide risk in the form of an interactive map,” said NASA’s Goddard Space Flight Center (Greenbelt, Maryland), who led the study. Thomas Stanley, a scientist at the University Space Research Association of Maryland, said. “It’s worth it because it provides a relative scale of landslide risk, rather than just saying if there is a landslide risk. Users define the area of interest and tailor it to their needs. You can adjust the category and probability thresholds. “
To “teach” the model, researchers enter a table that contains all relevant landslide variables and many locations where landslides have been recorded in the past. Machine learning algorithms take tables, test different scenarios and results, and output a decision tree when they find the one that best fits the data. Then identify the errors in the decision tree and calculate another tree to fix those errors. This process continues until the model is “trained” and improved 300 times.
“As a result, this version of the model is approximately twice as accurate as the first version of the model, making it the most accurate global nowcasting tool available,” says Stanley. “The accuracy of large-scale landslide events caused by tropical cyclones is the highest (often 100%), but has been significantly improved in all inventories.”
Version 1, released in 2018, was not a machine learning model. Nowcasting was created by combining satellite precipitation data and a global landslide susceptibility map. We made predictions using a single decision tree, primarily based on previous week’s rainfall data, and categorized each grid cell into low-risk, medium-risk, and high-risk.
“In this new version, the 300 trees contain even better information compared to the first version. version“Version 2 incorporates more variables than before, such as soil moisture and snow volume data,” Stanley said.
In general, soil can only absorb large amounts of water before it becomes saturated, and in combination with other conditions there is a risk of landslides. By incorporating soil moisture data, the model can identify the amount of water already present in the soil and the amount of additional precipitation above that threshold. Similarly, if the model knows the amount of snow present in a particular area, it can consider additional water that invades the soil as the snow melts. This data comes from the Soil Moisture Active Passive (SMAP) satellite managed by NASA’s Jet Propulsion Laboratory in Southern California. Launched in 2015, it will continue to be available. Soil moisture coverage.
LHASA version 2 also adds a new exposure feature that analyzes the distribution of roads and populations within each grid cell to calculate the number of people and infrastructure at risk of landslides. The exposure data is available for download and is integrated into the interactive map. Adding this type of information about exposed roads and people vulnerable to landslides can help stakeholders, from international organizations to local authorities, improve awareness and behavior.
LHASA version 2 has been tested in the real world up to its official release by the NASA disaster program and stakeholders, based on years of research and applications. When hurricanes Eta and Iota hit Central America in November 2020, researchers working on NASA’s Global Applied Science Disaster Program predicted Guatemala and Honduras using LHASA version 2. We have created a map of the danger of landslides. Researchers have superimposed district-level population data on the model to better assess potential hazards and proximity to densely populated communities. The disaster program coordinator shared information with national and international emergency response agencies and provided ground personnel with better insights into hazards.
According to Stanley, it’s a useful tool for planning and risk mitigation purposes, but the model is intended to be used with a global perspective, not as a regional emergency alert system in a particular region. I am. However, future research may extend that goal.
“We are working on incorporating precipitation forecasts into LHASA version 2 and hope to provide more information for advanced planning and action ahead of major rainfall events,” said Stanley. According to Stanley, one of the challenges is to get enough archives of precipitation forecast data that the model can train.
In the meantime, governments, relief agencies, emergency responders, and other stakeholders (and the general public) have access to LHASA Version 2’s powerful risk assessment tools.
NASA Goddard Space Flight Center
Quote: The machine learning model is the global landslide “Nowcast” (Nowcasts) obtained from https://phys.org/news/2021-06-machine-accuracy-global-landslide-nowcasts.html on June 10, 2021. Double the accuracy of June 10, 2021)
This document is subject to copyright. No part may be reproduced without written permission, except for private research or fair trade for research purposes. The content is provided for informational purposes only.
Machine learning models double the accuracy of the global landslide “nowcast”
Source link Machine learning models double the accuracy of the global landslide “nowcast”