Evaluating Google AlphaEarth and Landsat 8/9 Multisource Data Using Random Forest, Rotation Forest, and Canonical Correlation Forest for Enhanced Land Cover Classification
DOI:
https://doi.org/10.55779/ng62650Keywords:
canonical correlation forest, embeddings, Google AlphaEarth, land cover classification, Landsat 8, Landsat 9, random forest, rotation forestAbstract
Land cover classification is undergoing a major shift driven by advances in Earth observation and machine learning. While Landsat 8 and Landsat 9 continue to provide reliable, physically interpretable, and openly accessible multispectral data, emerging AI-based foundation models such as Google’s AlphaEarth offer a contrasting paradigm built on high-dimensional satellite embeddings. This study addresses a critical gap in current scholarship—the lack of controlled, classifier-agnostic comparisons between traditional multispectral data and foundation model-generated embeddings—through an empirical evaluation of Landsat 8, Landsat 9, and Google Satellite Embeddings V1 for land cover classification across four LULC classes (Urban, Vegetation, Water, and Bare Land) in the Tema Municipal Area, Ghana. A stratified random sample of 339 reference points was partitioned into 70% for training (240 samples) and 30% for independent holdout validation (99 samples), maintaining a consistent split across all datasets and classifiers. Three ensemble classifiers were tested at a common 30 m analysis grid: Random Forest (RF), Rotation Forest (RotFor), and Canonical Correlation Forest (CCF). Results showed that AlphaEarth embeddings outperformed the Landsat datasets across all classifiers. CCF achieved the highest performance on AlphaEarth with an Overall Accuracy (OA) of 98.3% and a Kappa coefficient of 0.977. RF and RotFor achieved OAs of 97.25% (Kappa 0.963) and 98.05% (Kappa 0.974), respectively. Landsat 8 produced lower accuracies, with RF at 92.85% (Kappa 0.90), RotFor at 93.40% (Kappa 0.91), and CCF at 95.10% (Kappa 0.93). Landsat 9 performed slightly below Landsat 8 in our experiments. We attribute this to the selection of acquisition dates and differences in QA_PIXEL masking effectiveness. Landsat 9 produced an OA of 90.6% with RF (Kappa 0.87), RotFor at 91.15% (Kappa 0.88), and CCF at 91.9% (Kappa 0.89).
Metrics
References
Altmann A, Toloşi L, Sander O, Lengauer T (2010). Permutation importance: a corrected feature importance measure. Bioinformatics, 26(10): 1340–1347. https://doi.org/10.1093/bioinformatics/btq134
Alvarez CI, Ulloa Vaca CA, Echeverria Llumipanta NA (2025). Machine learning for urban air quality prediction using Google AlphaEarth Foundations satellite embeddings: a case study of Quito, Ecuador. Remote Sensing, 17(20): 3472. https://doi.org/10.3390/rs17203472
Amraoui M, Mijanovic D, El Amrani M, Kader S, Ouakhir H (2023). Agriculture and economic development of the Ait Werra tribe during the French colonialism period and its local characteristics (1912–1956) within the Middle Atlas region of Morocco. Agriculture and Forestry, 69(4): 91–112. https://doi.org/10.17707/AgricultForest.69.4.07
Ayush K, Uzkent B, Meng C, Tanmay K, Burke M, Lobell D, et al. (2021). Geography-aware self-supervised learning. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 10181–10190. https://doi.org/10.48550/arXiv.2011.09980
Braham NAA, Albrecht CM, Mairal J, Chanussot J, Wang Y, Zhu XX (2025). SpectralEarth: training hyperspectral foundation models at scale. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18: 16780–16797. https://doi.org/10.1109/JSTARS.2025.3581451
Breiman L (2001). Random forests. Machine Learning, 45: 5–32. https://doi.org/10.1023/A:1010933404324
Brown CF, Kazmierski MR, Pasquarella VJ, Rucklidge WJ, Samsikova M, Zhang C, et al. (2025). AlphaEarth Foundations: an embedding field model for accurate and efficient global mapping from sparse label data. arXiv preprint. https://doi.org/10.48550/arXiv.2507.22291
Colkesen I, Kavzoglu T (2017). Ensemble-based canonical correlation forest (CCF) for land use and land cover classification using Sentinel-2 and Landsat OLI imagery. Remote Sensing Letters, 8(11): 1082–1091. https://doi.org/10.1080/2150704X.2017.1354262
Cong Y, Khanna S, Meng C, Liu P, Rozi E, He Y, et al. (2022). SatMAE: pre-training transformers for temporal and multi-spectral satellite imagery. In: Advances in Neural Information Processing Systems, 35: 197–211. https://doi.org/10.48550/arXiv.2207.08051
Ennaji N, Ouakhir H, Abahrour M, Spalevic V, Dudic B (2024). Impact of watershed management practices on vegetation, land use changes, and soil erosion in River Basins of the Atlas, Morocco. Notulae Botanicae Horti Agrobotanici Cluj-Napoca, 52(1): 13567. https://doi.org/10.15835/nbha52113567
Ennaji N, Ouakhir H, Halouan S, Abahrour M (2022). Assessment of soil erosion rate using the EPM model: case of Ouaoumana basin, Middle Atlas, Morocco. IOP Conference Series: Earth and Environmental Science, 1090(1): 012004. https://doi.org/10.1088/1755-1315/1090/1/012004
Fang J, Wu M, Zhang Z, Luo W (2025). Leveraging AlphaEarth Foundations Embeddings for High-Accuracy County-Scale Corn and Soybean Yield Estimation. TechRxiv preprint. https://doi.org/10.36227/techrxiv.175825526.60198358/v1
Fernández-Delgado M, Cernadas E, Barro S, Amorim D (2014). Do we need hundreds of classifiers to solve real world classification problems? Journal of Machine Learning Research, 15(1): 3133–3181. https://doi.org/10.5555/2627435.2697065
Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R (2017). Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202: 18–27. https://doi.org/10.1016/j.rse.2017.06.031
Hong D, Zhang B, Li X, Li Y, Li C, Yao J, et al. (2023). SpectralGPT: spectral remote sensing foundation model. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46: 5227–5244. https://doi.org/10.1109/TPAMI.2024.3362475
Houriez L, Pilarski S, Vahedi B, Ahmadalipour A, Scully TH, Aflitto N, et al. (2025). Scalable geospatial data generation using AlphaEarth Foundations model. arXiv preprint. https://doi.org/10.48550/arXiv.2508.11739
Jakubik J, Roy S, Phillips CE, Fraccaro P, Godwin D, Zadrozny B, et al. (2023). Foundation models for generalist geospatial artificial intelligence. arXiv preprint. https://doi.org/10.48550/arXiv.2310.18660
Janowicz K, Mai G, Huang W, Zhu R, Lao N, Cai L (2025). GeoFM: how will geo-foundation models reshape spatial data science and GeoAI? International Journal of Geographical Information Science, 39(9): 1849–1865. https://doi.org/10.1080/13658816.2025.2543038
Mañas Ó, Lacoste A, Giró-i-Nieto X, Vázquez D, Rodriguez P (2021). Seasonal contrast: unsupervised pre-training from uncurated remote sensing data. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9414–9423. https://doi.org/10.48550/arXiv.2103.16607
Meyer H, Reudenbach C, Hengl T, Katurji M, Nauss T (2018). Improving performance of spatio-temporal machine learning models using forward feature selection and target-oriented validation. Environmental Modelling & Software, 101: 1–9. https://doi.org/10.1016/j.envsoft.2017.12.001
Murakami K (2025). Within- and cross-regional crop classification for cool climate upland agriculture using AlphaEarth. agriRxiv preprint. https://doi.org/10.31220/agriRxiv.2025.00354
Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. (2011). Scikit-learn: machine learning in Python. Journal of Machine Learning Research, 12: 2825–2830. https://doi.org/10.5555/1953048.2078195
Rainforth T, Wood F (2015). Canonical correlation forests. arXiv preprint. https://doi.org/10.48550/arXiv.1507.05444
Rodriguez JJ, Kuncheva LI, Alonso CJ (2006). Rotation forest: a new classifier ensemble method. IEEE Transactions on Pattern Analysis and Machine Intelligence, 28(10): 1619–1630. https://doi.org/10.1109/TPAMI.2006.211
Sahin EK, Colkesen I, Kavzoglu T (2020). A comparative assessment of canonical correlation forest, random forest, rotation forest and logistic regression methods for landslide susceptibility mapping. Geocarto International, 35(4): 341–363. https://doi.org/10.1080/10106049.2018.1516248
Seydi ST (2025). Deep learning-based burned area mapping using bi-temporal Siamese networks and AlphaEarth Foundation Datasets. arXiv preprint. https://doi.org/10.48550/arXiv.2509.07852
Tollefson J (2025). Google AI model mines trillions of images to create maps of Earth ‘at any place and time’. Nature, 644: 313–314. https://doi.org/10.1038/d41586-025-02412-1
Wang Y, Braham NAA, Xiong Z, Liu C, Albrecht CM, Zhu XX (2023). SSL4EO-S12: a large-scale multimodal, multitemporal dataset for self-supervised learning in earth observation. IEEE Geoscience and Remote Sensing Magazine, 11(3): 98–106. https://doi.org/10.1109/MGRS.2023.3268756
Xia J, Yokoya N, Iwasaki A (2016). Hyperspectral image classification with canonical correlation forests. IEEE Transactions on Geoscience and Remote Sensing, 55(1): 421–431. https://doi.org/10.1109/TGRS.2016.2605080
Xiao A, Xuan W, Wang J, Huang J, Tao D, Lu S, et al. (2025). Foundation models for remote sensing and earth observation: a survey. IEEE Geoscience and Remote Sensing Magazine, pp. 2–29. https://doi.org/10.1109/MGRS.2024.3477426
Xie Y, Wang Z, Mai G, Li Y, Jia X, Gao S, et al. (2023). Geo-foundation models: reality, gaps and opportunities. In: Proceedings of the 31st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. New York, NY: Association for Computing Machinery. https://doi.org/10.1145/3589132.3625616
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Saviour Mantey, Isaac S.K. Attipoe, Toffick W. Mohammed
This work is licensed under a Creative Commons Attribution 4.0 International License.
Open Access, Copyright, and Licensing
All content published in Nova Geodesia is made immediately available on an open-access basis under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Authors retain copyright in their published work. By submitting and publishing an article in the journal, the authors grant Nova Geodesia and its publisher, the Society of Land Measurements and Cadastre from Transylvania (SMTCT), a non-exclusive right to publish, distribute, reproduce, archive, preserve, and disseminate the article in all formats and media.
Under the CC BY 4.0 license, users may read, download, copy, distribute, print, search, link to, adapt, and otherwise reuse the published content for any lawful purpose, including commercial use, provided that appropriate credit is given to the authors and the original publication, a link to the license is provided, and any changes are indicated.
The full terms of the license are available at:
https://creativecommons.org/licenses/by/4.0/
Authors are responsible for obtaining permission to reproduce any third-party material that is not covered by the article’s CC BY 4.0 license.
No submission fee, article processing charge, publication fee, or other mandatory publication charge is imposed by the journal.
