Beyond Standalone Geo-Embeddings: Weighted Multi-Model Ensemble Prediction for Tropical Land-Cover Mapping.
Saved in:
| Title: | Beyond Standalone Geo-Embeddings: Weighted Multi-Model Ensemble Prediction for Tropical Land-Cover Mapping. |
|---|---|
| Authors: | Havinga, Alessandra1,2,3 (AUTHOR) alessandra.havinga@geneve.ch, Giuliani, Gregory2,4,5 (AUTHOR), Nobel, Sophie3,4 (AUTHOR), Ranirison, Patrick3,4,6 (AUTHOR), Nusbaumer, Louis1,2,3,5 (AUTHOR), Martin, Pascal1,2,6 (AUTHOR) |
| Source: | Remote Sensing. Sep2026, Vol. 18 Issue 17, p2952. 24p. |
| Subjects: | Ensemble learning, Geospatial data, Random forest algorithms, Tropical ecosystems |
| Geographic Terms: | Madagascar |
| Abstract: | Highlights: What are the main findings? Alpha Earth Foundations Geo-embeddings outperformed all conventional Sentinel-1 and 2 time-series classifiers for detailed tropical land-cover mapping, reaching an overall accuracy of 0.74 across 18 land-cover classes. A reproducible workflow was developed to integrate AlphaEarth-derived probability maps into the SITS framework, enabling an ensemble model that increased overall accuracy by 8 percentage points (0.74 to 0.82). What are the implications of the main findings? Geo-embeddings may complement, rather than replace, conventional satellite image time-series approaches. Understanding how Geo-embeddings and EO time series provide complementary information is a key challenge for future land-cover mapping. Accurate land-cover mapping in tropical regions remains challenging because of high environmental heterogeneity, complex vegetation structure, and persistent cloud cover. Recent geospatial foundation models provide pre-computed Geo-embeddings that offer a new representation of Earth Observation data (EO), yet their potential for detailed tropical land-cover mapping and their performance relative to, and in combination with, conventional Sentinel-1 and Sentinel-2 satellite image time series remains largely unexplored. This study evaluated AlphaEarth Foundation Geo-embeddings for detailed vegetation mapping in northern Madagascar and investigated whether combining them with satellite image time series processed using the Satellite Image Time Series (SITS) R package could improve classification performance. Geo-embeddings were classified using a Random Forest model (GEO), while four supervised classifiers were trained on Sentinel-1 and Sentinel-2 time series. GEO achieved the highest standalone performance (OA = 0.74), outperforming all classifiers trained on the satellite image time series. Probability-level ensemble models were then used to assess whether both data representations could be beneficially combined. The best-performing ensemble, combining GEO (75%) with Temporal Convolutional Neural Network (TempCNN; 25%), increased overall accuracy from 0.74 to 0.82. These results suggest that Geo-embeddings provide an effective standalone representation for detailed tropical land-cover mapping and indicate that combining them with conventional satellite image time series may exploit complementary information. As geospatial foundation models continue to evolve, understanding how they can be integrated with established Earth observation workflows may support future operational land-cover mapping. [ABSTRACT FROM AUTHOR] |
| Copyright of Remote Sensing is the property of MDPI and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| FullText | Links: – Type: pdflink Text: Availability: 1 |
|---|---|
| Header | DbId: egs DbLabel: Engineering Source An: 197043802 AccessLevel: 6 PubType: Academic Journal PubTypeId: academicJournal PreciseRelevancyScore: 0 |
| IllustrationInfo | |
| Items | – Name: Title Label: Title Group: Ti Data: Beyond Standalone Geo-Embeddings: Weighted Multi-Model Ensemble Prediction for Tropical Land-Cover Mapping. – Name: Author Label: Authors Group: Au Data: <searchLink fieldCode="AR" term="%22Havinga%2C+Alessandra%22">Havinga, Alessandra</searchLink><relatesTo>1,2,3</relatesTo> (AUTHOR)<i> alessandra.havinga@geneve.ch</i><br /><searchLink fieldCode="AR" term="%22Giuliani%2C+Gregory%22">Giuliani, Gregory</searchLink><relatesTo>2,4,5</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Nobel%2C+Sophie%22">Nobel, Sophie</searchLink><relatesTo>3,4</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Ranirison%2C+Patrick%22">Ranirison, Patrick</searchLink><relatesTo>3,4,6</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Nusbaumer%2C+Louis%22">Nusbaumer, Louis</searchLink><relatesTo>1,2,3,5</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Martin%2C+Pascal%22">Martin, Pascal</searchLink><relatesTo>1,2,6</relatesTo> (AUTHOR) – Name: TitleSource Label: Source Group: Src Data: <searchLink fieldCode="JN" term="%22Remote+Sensing%22">Remote Sensing</searchLink>. Sep2026, Vol. 18 Issue 17, p2952. 24p. – Name: Subject Label: Subjects Group: Su Data: <searchLink fieldCode="DE" term="%22Ensemble+learning%22">Ensemble learning</searchLink><br /><searchLink fieldCode="DE" term="%22Geospatial+data%22">Geospatial data</searchLink><br /><searchLink fieldCode="DE" term="%22Random+forest+algorithms%22">Random forest algorithms</searchLink><br /><searchLink fieldCode="DE" term="%22Tropical+ecosystems%22">Tropical ecosystems</searchLink> – Name: SubjectGeographic Label: Geographic Terms Group: Su Data: <searchLink fieldCode="DE" term="%22Madagascar%22">Madagascar</searchLink> – Name: Abstract Label: Abstract Group: Ab Data: Highlights: What are the main findings? Alpha Earth Foundations Geo-embeddings outperformed all conventional Sentinel-1 and 2 time-series classifiers for detailed tropical land-cover mapping, reaching an overall accuracy of 0.74 across 18 land-cover classes. A reproducible workflow was developed to integrate AlphaEarth-derived probability maps into the SITS framework, enabling an ensemble model that increased overall accuracy by 8 percentage points (0.74 to 0.82). What are the implications of the main findings? Geo-embeddings may complement, rather than replace, conventional satellite image time-series approaches. Understanding how Geo-embeddings and EO time series provide complementary information is a key challenge for future land-cover mapping. Accurate land-cover mapping in tropical regions remains challenging because of high environmental heterogeneity, complex vegetation structure, and persistent cloud cover. Recent geospatial foundation models provide pre-computed Geo-embeddings that offer a new representation of Earth Observation data (EO), yet their potential for detailed tropical land-cover mapping and their performance relative to, and in combination with, conventional Sentinel-1 and Sentinel-2 satellite image time series remains largely unexplored. This study evaluated AlphaEarth Foundation Geo-embeddings for detailed vegetation mapping in northern Madagascar and investigated whether combining them with satellite image time series processed using the Satellite Image Time Series (SITS) R package could improve classification performance. Geo-embeddings were classified using a Random Forest model (GEO), while four supervised classifiers were trained on Sentinel-1 and Sentinel-2 time series. GEO achieved the highest standalone performance (OA = 0.74), outperforming all classifiers trained on the satellite image time series. Probability-level ensemble models were then used to assess whether both data representations could be beneficially combined. The best-performing ensemble, combining GEO (75%) with Temporal Convolutional Neural Network (TempCNN; 25%), increased overall accuracy from 0.74 to 0.82. These results suggest that Geo-embeddings provide an effective standalone representation for detailed tropical land-cover mapping and indicate that combining them with conventional satellite image time series may exploit complementary information. As geospatial foundation models continue to evolve, understanding how they can be integrated with established Earth observation workflows may support future operational land-cover mapping. [ABSTRACT FROM AUTHOR] – Name: AbstractSuppliedCopyright Label: Group: Ab Data: <i>Copyright of Remote Sensing is the property of MDPI and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract.</i> (Copyright applies to all Abstracts.) |
| PLink | https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=egs&AN=197043802 |
| RecordInfo | BibRecord: BibEntity: Identifiers: – Type: doi Value: 10.3390/rs18172952 Languages: – Code: eng Text: English PhysicalDescription: Pagination: PageCount: 24 StartPage: 2952 Subjects: – SubjectFull: Ensemble learning Type: general – SubjectFull: Geospatial data Type: general – SubjectFull: Random forest algorithms Type: general – SubjectFull: Tropical ecosystems Type: general – SubjectFull: Madagascar Type: general Titles: – TitleFull: Beyond Standalone Geo-Embeddings: Weighted Multi-Model Ensemble Prediction for Tropical Land-Cover Mapping. Type: main BibRelationships: HasContributorRelationships: – PersonEntity: Name: NameFull: Havinga, Alessandra – PersonEntity: Name: NameFull: Giuliani, Gregory – PersonEntity: Name: NameFull: Nobel, Sophie – PersonEntity: Name: NameFull: Ranirison, Patrick – PersonEntity: Name: NameFull: Nusbaumer, Louis – PersonEntity: Name: NameFull: Martin, Pascal IsPartOfRelationships: – BibEntity: Dates: – D: 01 M: 09 Text: Sep2026 Type: published Y: 2026 Identifiers: – Type: issn-print Value: 20724292 Numbering: – Type: volume Value: 18 – Type: issue Value: 17 Titles: – TitleFull: Remote Sensing Type: main |
| ResultId | 1 |
