{"id":558116,"date":"2025-08-25T20:46:06","date_gmt":"2025-08-25T23:46:06","guid":{"rendered":"https:\/\/revistapesquisa.fapesp.br\/?p=558116"},"modified":"2025-08-25T20:46:06","modified_gmt":"2025-08-25T23:46:06","slug":"artificial-intelligence-identifies-areas-most-vulnerable-to-landslides","status":"publish","type":"post","link":"https:\/\/revistapesquisa.fapesp.br\/en\/artificial-intelligence-identifies-areas-most-vulnerable-to-landslides\/","title":{"rendered":"Artificial intelligence identifies areas most vulnerable to landslides"},"content":{"rendered":"

Artificial Intelligence (AI) techniques can be useful for identifying areas more prone to landslides. A study published in October 2024 in the scientific journal Natural Hazards Research<\/em> compared the effectiveness of five models based on machine-learning algorithms for detecting and predicting regions in the municipality of S\u00e3o Sebasti\u00e3o most susceptible to this type of hazard. According to the article by researchers from S\u00e3o Paulo State University (UNESP) and the Brazilian Center for Natural Disaster Monitoring (CEMADEN), one of the models, named Gradient Boosting, achieved 99.6% accuracy in mapping the stretches most vulnerable to landslides. With an almost identical performance, the Random Forest algorithm came second in the ranking produced by the authors of the research.<\/p>\n

The study covered just over 400 square kilometers (km2<\/sup>) of land along the northern coastline of the state of S\u00e3o Paulo, an area prone to intense rainfall and soil movement from the slopes of the Serra do Mar mountain range. Between February 18 and 19, 2023, during Carnaval, it rained more than 600 millimeters (mm), the equivalent to two months, in S\u00e3o Sebasti\u00e3o. There were landslides, houses collapsed, 2,400 people were left homeless, and 64 lost their lives. To classify the performance of the algorithms, the results of the models were compared with maps of the region showing the areas that are most prone to this type of occurrence.<\/p>\n

The algorithms calculate the risk of a landslide occurring in a location based on the analysis of data relating to environmental factors associated with processes that influence ground stability. The main elements considered are the slope angle of the land, soil moisture, relief dissection (fragmentation), and the geomorphological parameters of the region. \u201cThe machine-learning models enable the integration of different conditional variables and provide a solid foundation for creating susceptibility maps,\u201d says remote sensing expert Enner Alc\u00e2ntara, of the Institute of Science and Technology (ICT) at UNESP\u2019s campus in S\u00e3o Jos\u00e9 dos Campos, lead author of the study. \u201cThey enable complex patterns to be identified that may not be evident in more traditional approaches.\u201d<\/p>\n

The Gradient Boosting algorithm presents a peculiarity: it combines approaches of various simpler models, each specialized in one of the analyzed variables. This more integrated view suggests that the gradient, land fragmentation, and soil moisture index are the factors that impact slope stability the most. \u201cGreater forest cover was associated with lower levels of risk, while areas of pasture presented greater susceptibility to landslides,\u201d comments Alc\u00e2ntara.<\/p>\n<\/div>

\n \n \n \n <\/picture>Alexandre Affonso \/ Revista Pesquisa FAPESP<\/span><\/div>
\n

Based on the Gradient Boosting model, a landslide susceptibility map was created on which the points of S\u00e3o Sebasti\u00e3o were classified into four risk categories: low (74.6% of the municipality\u2019s territory), moderate (15.8%), high (7.9%), and very high (1.7%). Despite the predominance of areas with low susceptibility, isolated pockets of high risk were identified in regions occupied by open scars left in the soil from past landslides, such as in the Serra do Mar State Park, close to Juque\u00ed beach and in Vila Sahy. In the February 2023 tragedy, the majority of deaths occurred in the latter location.<\/p>\n

Other studies have also highlighted the vulnerability of certain areas of S\u00e3o Sebasti\u00e3o to this type of occurrence. In September 2024, an article published in the Brazilian Journal of Geology<\/em> identified 1,000 landslide points in the municipality in the state of S\u00e3o Paulo through the analysis of aerial images taken shortly after the disaster two years earlier. \u201cAlthough a large part of the territory of S\u00e3o Sebasti\u00e3o is relatively safe, the risk of a landslide occurring is really high in the more vulnerable areas,\u201d comments geologist Carlos Henrique Grohmann, from the Institute of Astronomy, Geophysics, and Atmospheric Sciences of the University of S\u00e3o Paulo (IAG-USP), one of the authors of the paper.<\/p>\n

For Alc\u00e2ntara, one of the strengths of machine-learning models is their ability to be adapted to different regions and scenarios, as long as there is enough data to feed the algorithms. \u201cThis flexibility makes AI a powerful tool for developing countries, where resources for disaster mitigation may be limited,\u201d suggests the researcher from UNESP. Recent international studies have explored the potential of machine-learning-based models to map landslide prone zones in different parts of the world, such as in the Himalayas in Asia and the Andes in South America.<\/p>\n

The disasters that occur every year during the period of intense rainfall motivate the search for solutions to help prevent deaths and material losses. In August 2024, S\u00e3o Sebasti\u00e3o was one of 11 Brazilian municipalities selected for the first tests of the Civil Defense Alert. It is a pilot project created by the Federal Government, in partnership with the Brazilian National Telecommunications Agency (ANATEL), which uses the cellphone network to issue audible alerts, without the need for previous registration and even works for cellphones in silent mode, when there is an imminent risk of a disaster in your area. \u201cTwo years after the tragedy, I feel a little safer. But, even with this warning system, I am worried if we have rainfall like in 2023,\u201d comments community leader Rosilene de Jesus Santos, known as Nega Rose, a resident of Vila Sahy for 34 years.<\/p>\n

The story above was published with the title “AI against landslides<\/strong>” in issue in issue 348 of february\/2025.<\/p>\n

Scientific articles
\n<\/strong>ALC\u00c2NTARA, E. et al<\/em>. Machine learning approaches for mapping and predicting landslide-prone areas in S\u00e3o Sebasti\u00e3o (Southeast Brazil)<\/a>. Natural Hazards Research<\/strong>. Oct. 18, 2024.
\nCOELHO, R. D. et al<\/em>. Landslides of the 2023 summer event of S\u00e3o Sebasti\u00e3o, southeastern Brazil: Spatial dataset.<\/a> Brazilian Journal of Geology<\/strong>. Sept. 20, 2024.<\/p>\n","protected":false},"excerpt":{"rendered":"Study compares the effectiveness of five machine-learning models at predicting this type of disaster in S\u00e3o Sebasti\u00e3o","protected":false},"author":719,"featured_media":558117,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_exactmetrics_skip_tracking":false,"_exactmetrics_sitenote_active":false,"_exactmetrics_sitenote_note":"","_exactmetrics_sitenote_category":0,"footnotes":""},"categories":[159],"tags":[217,200,2413],"coauthors":[4223],"class_list":["post-558116","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-science","tag-climate","tag-environment","tag-technology"],"acf":[],"_links":{"self":[{"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/posts\/558116","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/users\/719"}],"replies":[{"embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/comments?post=558116"}],"version-history":[{"count":1,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/posts\/558116\/revisions"}],"predecessor-version":[{"id":558129,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/posts\/558116\/revisions\/558129"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/media\/558117"}],"wp:attachment":[{"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/media?parent=558116"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/categories?post=558116"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/tags?post=558116"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/revistapesquisa.fapesp.br\/en\/wp-json\/wp\/v2\/coauthors?post=558116"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}