kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Supervised and unsupervised machine learning approaches using Sentinel data for flood mapping and damage assessment in Mozambique
KTH, School of Architecture and the Built Environment (ABE), Urban Planning and Environment.ORCID iD: 0000-0003-4448-6180
KTH, School of Architecture and the Built Environment (ABE), Urban Planning and Environment, Geoinformatics.ORCID iD: 0000-0001-9692-8636
KTH, School of Architecture and the Built Environment (ABE), Urban Planning and Environment, Geoinformatics.ORCID iD: 0000-0002-0001-2058
KTH, School of Architecture and the Built Environment (ABE), Urban Planning and Environment, Geoinformatics.ORCID iD: 0000-0003-1369-3216
Number of Authors: 42023 (English)In: Remote Sensing Applications: Society and Environment, E-ISSN 2352-9385, Vol. 32, article id 101015Article in journal (Refereed) Published
Abstract [en]

Natural hazards, such as flooding, have been negatively impacting developed and emerging economies alike. The effects of floods are more prominent in countries of the Global South, where large parts of the population and infrastructure are insufficiently protected from natural hazards. From this scope, a lot of effort is required to mitigate these impacts by continuously providing new and more reliable tools to aid in mitigation and preparedness, during or after a flood event. Flood mapping followed by damage assessment plays an important role in all these stages. In this work we investigate a new dataset provided by DrivenData Labs based on Sentinel-1 (S1) imagery (VH, VV imagery and labels) to help map floods in the city of Beira in Mozambique. Exploiting Google Earth Engine (GEE), we deployed supervised and unsupervised machine learning (ML) methods on a dataset comprising imagery from 13 countries worldwide. We first mapped the floods country-by-country including Mozambique. This first part was helpful to understand the sensitivity of each method when applied to data from different regions and with different polarizations. We then trained the supervised model globally (in all 13 countries) and used it to predict floods in Beira. To assess the accuracy of the experiments we used the intersection over the union (IoU) metric, results of which we compared with the benchmark IoU achieved by the winner in the DrivenData competition for flood mapping in 2021. The implementation of unsupervised and supervised ML using VH and VV+VH produced satisfactory results, and showed to be better than using VV imagery; in Cambodia and Bolivia with VH polarization yielded IoUs values ranging from 0.819 to 0.856 which is above the benchmark (0.8094). The predictions in Beira using VH imagery yielded IoU of 0.568, which is a reasonable outcome. The proposed approach is a reliable alternative for flood mapping, especially in Mozambique due to its low cost and time effectiveness as even with unsupervised approaches, relatively high-quality results are yielded in near real-time. Finally, we used Sentinel-2 (S2) imagery for a land cover classification to perform damage assessment in Beira and integrated population data from Beira to enhance the quality the results. The results show that 20% of agricultural area and about 10% of built up area were flooded. Flooded built up area includes highly populated neighborhoods such as Chaimite and Ponta Gea that are located in the center of the city.

Place, publisher, year, edition, pages
Elsevier BV , 2023. Vol. 32, article id 101015
Keywords [en]
Classification, Damage assessment, DrivenData dataset, Flood mapping, Sentinel-1 and Sentinel-2
National Category
Earth Observation
Identifiers
URN: urn:nbn:se:kth:diva-333894DOI: 10.1016/j.rsase.2023.101015ISI: 001054671800001Scopus ID: 2-s2.0-85164383013OAI: oai:DiVA.org:kth-333894DiVA, id: diva2:1791178
Note

QC 20230824

Available from: 2023-08-24 Created: 2023-08-24 Last updated: 2025-02-10Bibliographically approved
In thesis
1. Multi-Temporal Sentinel-1 SAR and Sentinel-2 MSI Data for Flood Mapping and Damage Assessment in Mozambique
Open this publication in new window or tab >>Multi-Temporal Sentinel-1 SAR and Sentinel-2 MSI Data for Flood Mapping and Damage Assessment in Mozambique
2024 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

Floods are one of the most frequent natural disasters worldwide. Althoughthe vulnerability varies from region to region, all countries are susceptible toflooding. Mozambique was hit by several tropical cyclones (TCs) in the lastfew decades, and in 2019, after TCs Idai and Kenneth, the country becamethe first one in southern Africa to be hit by two cyclones in the same rainyseason. In 2023, Mozambique was slammed twice by the same cyclone (TCFreddy) which was also recorded as the longest one. Aiming to provide thelocal authorities with tools to yield better responses before and after any disasterevent, and to mitigate the impact and support in decision making forsustainable development, it is fundamental to continue investigating reliablemethods for disaster management. In this thesis, two approaches for floodmapping (FM) are proposed. The first is a fully automated method for FM innear real-time utilizing multi-temporal Sentinel-1 Synthetic Aperture Radar(SAR) data acquired in the Beira municipality and the Macomia district.The second approach relies on supervised and unsupervised machine learning(ML) methods as we investigate a dataset provided by DrivenData Labsbased on Sentinel-1 (S1) imagery (VH, VV imagery and labels from 13 countriesworldwide). By exploiting the processing capability of the Google EarthEngine (GEE) platform, both approaches are presented as an alternative todeep learning (DL) methods due to cost effectiveness and low computationalpower requirement. The first approach is implemented by finding the differencesof images acquired before and after the flooding and then use Otsu’sthresholding method to automatically extract the flooded area from the differenceimage, while the second one is based on ML methods such as SVMand K-Means. To validate and compute the accuracy of the proposed techniques,we compare our results with the Copernicus Emergency ManagementService (Copernicus EMS) data available in the study areas. Furthermore, weinvestigated the use of a Sentinel-2 (S2) multi-spectral instrument (MSI) toproduce a land cover (LC) map of the study area and estimate the percentageof flooded areas in each LC class. The results show that the combinationof S1 and S2 data is reliable for near real-time flood mapping and damageassessment. We automatically mapped flooded areas with an overall accuracyof about 87–88% and kappa of 0.73–0.75 for the first approach. The secondapproach produced satisfactory results, and showed to be better than usingVV imagery; in Cambodia and Bolivia with VH polarization yielded IoUs valuesranging from 0.819 to 0.856. The predictions in Beira using VH imageryyielded IoU of 0.568, which is a reasonable outcome. The LC classification isvalidated by randomly collecting over 600 points for each LC, and the overallaccuracy is 90–95% with a kappa of 0.80–0.94. With these results we wereable to detect areas that are prone to flooding and where floods recede fasterfor improving the planning; we were also able to determine the percentageof flooded LC such as Agriculture, Mangrove and Built as their destructionnegatively impacts on food security and socio-economic development plans.

Abstract [sv]

Översvämningar är en av de vanligaste naturkatastroferna i världen. Även omsårbarheten varierar från region till region är alla länder mottagliga för översvämningar.Moçambique drabbades av flera tropiska cykloner (TC) underde senaste decennierna, och 2019, efter cyklonerna Idai och Kenneth, blevlandet det första i södra Afrika som drabbades av två cykloner under sammaregnperiod. 2023 slog samma cyklon (TC Freddy) ner över Moçambiquetvå gånger, som också registrerades som den tidsmässig längsta. I syfte attförse de lokala myndigheterna med verktyg för att ge dem bättre möjligheteratt planera och genomföra hjälpinsatser före och efter varje katastrofhändelse,och för att mildra påverkan och ge stöd i beslutsfattande för hållbarutveckling, är det viktigt att fortsätta att utveckla tillförlitliga metoder förkatastrofhantering. I denna avhandling föreslås två metoder att genomföraöversvämningskartering (FM). Den första metoden är en helt automatiseradmetod för FM i nästan realtid som använder multi-temporala Sentinel-1 SyntheticAperture Radar (SAR)-data från European Space Agency (ESA) överBeira kommun och Macomia-distriktet. Den andra metoden bygger på övervakadoch oövervakad maskininlärning (ML) där vi undersöker en datamängdsom tillhandahålls av DrivenData Labs som är baserat på Sentinel-1 bilder(S1) (VH, VV-bilder och signaturer (små områden i bilder som markeratssom översvämmade/icke-översvämmade från 13 länder över hela världen)).Genom att använda Google Earth Engine (GEE)-plattformen framstår bådadessa metoder som alternativ till Deep Learning-metoder – de är kostnadseffektivaoch har låga krav på datorkraft. Den förstnämnda metoden implementerasgenom att hitta skillnaderna mellan bilder som tagits före och efteröversvämningen och sedan använda Otsus tröskelmetod för att automatisktextrahera det översvämmade området från skillnadsbilden; den andra baseraspå machine learning metoder som SVM och K-Means . För att valideraoch beräkna noggrannheten hos de föreslagna metoderna jämför vi våra resultatmed Copernicus Emergency Management Service (Copernicus EMS) datasom finns tillgängliga i studieområdena. Dessutom undersökte vi användningenav data från Sentinel-2’s (S2) multispektrala instrument (MSI) för attproducera en marktäckeskarta (LC) över studieområdet och kunna uppskattaandelen översvämmade områden i varje marktäckesklass. Resultaten visaratt kombinationen av S1- och S2-data är tillförlitlig för översvämningskarteringoch skadebedömning i nästan realtid. Vår automatiska kartläggning avöversvämmade områden resulterade i en total noggrannhet på cirka 87–88 %och ett kappavärde på 0,73–0,75 för den första metoden. Den andra metodengav tillfredsställande resultat och visade sig vara bättre än att använda VVbilder;i Kambodja och Bolivia med VH-polarisering erhöll vi IoUs-värdenfrån 0,819 till 0,856. Förutsägelserna i Beira med VH-bilder gav ett IoU på0,568, vilket är ett rimligt resultat. Marktäckesklassificeringen valideras genomatt slumpmässigt sampla över 600 poäng för varje marktäckesklass; dentotala noggrannheten blev 90–95 % med ett kappavärde på 0,80–0,94. Meddessa resultat kunde vi upptäcka områden som är utsatta för översvämningoch där översvämningar avtar snabbare för att förbättra markanvändningsplaneringen.Vi kunde också bestämma procentandelen översvämmade marktäckesklasser som jordbruk, Mangrove och byggt miljö, eftersom deras förstörelsenegativt påverkar livsmedelssäkerheten och socioekonomiska utvecklingsplaner.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2024. p. 66
Series
TRITA-ABE-DLT ; 2418
Keywords
Remote Sensing, Sentinel 1 and 2, Flood Mapping, Classification
National Category
Earth and Related Environmental Sciences
Research subject
Geodesy and Geoinformatics, Geoinformatics
Identifiers
urn:nbn:se:kth:diva-346482 (URN)978-91-8040-902-5 (ISBN)
Presentation
2024-06-05, 1515, 5th floor, Teknikringen 74 D, KTH Campus, public video conference link [MISSING], Stockholm, 10:00 (English)
Opponent
Supervisors
Note

QC 20240521

Available from: 2024-05-21 Created: 2024-05-16 Last updated: 2025-02-07Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Nhangumbe, ManuelNascetti, AndreaGeorganos, StefanosBan, Yifang

Search in DiVA

By author/editor
Nhangumbe, ManuelNascetti, AndreaGeorganos, StefanosBan, Yifang
By organisation
Urban Planning and EnvironmentGeoinformatics
Earth Observation

Search outside of DiVA

GoogleGoogle Scholar

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 54 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf