Mapping Uzon caldera vegetation using remote sensing data and machine learning

DOI: 10.35595/2414-9179-2025-2-31-83-99

View or download the article (Rus)

About the Authors

Nikita V. Maksimovich

St. Petersburg State University, Institute of Earth Sciences,
7–9, Universitetskaya emb., St. Petersburg, 199034, Russia,
E-mail: nekit.maksimovich@mail.ru

Artem A. Tarasov

St. Petersburg State University, Institute of Earth Sciences,
7–9, Universitetskaya emb., St. Petersburg, 199034, Russia,
E-mail: artrar.90@mail.ru

Anton P. Korablev

Komarov Botanical Institute of the Russian Academy of Sciences,
2, Professora Popova str., St. Petersburg, 197376, Russia,
E-mail: akorablev@binran.ru

Natalia A. Pozdnyakova

St. Petersburg State University, Institute of Earth Sciences,
7–9, Universitetskaya emb., St. Petersburg, 199034, Russia,
E-mail: n.pozdnyakova@spbu.ru

Tatyana A. Andreeva

St. Petersburg State University, Institute of Earth Sciences,
7–9, Universitetskaya emb., St. Petersburg, 199034, Russia,
E-mail: t.andreeva@spbu.ru

Olga V. Artemeva

St. Petersburg State University, Institute of Earth Sciences,
7–9, Universitetskaya emb., St. Petersburg, 199034, Russia,
E-mail: o.artemeva@spbu.ru

Abstract

Remote sensing data (RSD) and the machine learning methods used with them are currently used in mapping geographic processes and objects of the earth’s surface associated with the functioning of ecosystems and landscapes. RSD is often used to map one of the landscape components—vegetation, often representing the main source of information about it. This also applies to such hard-to-reach areas as the Uzon caldera. Thus, the purpose of this work was to create a cartographic model of the vegetation of a part of the Uzon caldera. For this purpose, the work used a generalized map of the caldera vegetation compiled by A.O. Pesterov, data from geobotanical descriptions provided by employees of the Botanical Institute of the Russian Academy of Sciences as auxiliary materials for creating a training sample for machine learning. The work also used data on the height of the vegetation cover, aboveground biomass, as well as several Sentinel-2 and Landsat-9 channels, from which a composite raster was created. In addition to channels 4–8 of Sentinel-2 and channel 10 of Landsat-9, the classified raster included ArcticDEM data. The training sample was created based on auxiliary data; the test sample was created by expert assessment based on high-resolution remote sensing data and geobotanical description data. K-means and Random Forest were selected as the classification methods used. For the first, the elbow method was used to assess the expression and separability of classes from each other in the classified raster. This method showed the inconsistency of the unsupervised classification method and the need to use methods with training. For the second, an accuracy assessment was used on test data with pre-determined optimal parameters of the trained model; an error matrix for each class was also compiled and classification quality metrics were calculated in Python 3.0. Thus, the overall accuracy of the model was 90 %. Statistical characteristics of the model were calculated and the features of its operation with the resulting vegetation classes were identified.

Keywords

Earth remote sensing, machine learning, vegetation, accuracy metrics, confusion matrix

References

  1. Boateng E.Y., Otoo J., Abaye D.A. Basic Tenets of Classification Algorithms K-nearest-neighbor, Support Vector Machine, Random Forest and Neural Network. Journal of Data Analysis and Information Processing, 2020. V. 8. No. 4. P. 341–357. DOI: 10.4236/jdaip.2020.84020.
  2. Boitard S., Mialon A., Mermoz S., Rodríguez-Fernández N.J., Richaume P., Salazar-Neira J.C., Tarot S., Kerr Y.H. Aboveground Biomass Dataset from SMOS L-band Vegetation Optical Depth and Reference Maps. Earth System Science Data, 2025. V. 17. No. 3. P. 1101–1119. DOI: 10.1186/s13021-022-00214-w.
  3. Kotlov I.P., Chernenkova T.V., Belyaeva N.G. Mapping Experience on the Forest Typological Diversity in the Moscow Region. Geobotanical Mapping, 2024. P. 25–43 (in Russian). DOI: 10.31111/geobotmap/2024.25.
  4. Malysheva N.V. Automated Interpretation of Aerospace Images of Forest Stands. Moscow: Publishing House of Moscow State University of Forestry, 2012. 154 p. (in Russian).
  5. Mohammadpour P., Viegas D.X., Viegas C. Vegetation Mapping with Random Forest Using Sentinel-2 and GLCM Texture Feature—A Case Study for Lousã Region, Portugal. Remote Sensing, 2022. V. 14. No. 18. P. 4585. DOI: 10.3390/rs14184585.
  6. Neshataeva V.Yu., Pesterov A.O. The Vegetation Cover Structure of Uzon Volcano Caldera (Eastern Kamchatka). Izvestiya Samarskogo nauchnogo centra RAN (News of the Samara Scientific Center of the Russian Academy of Sciences), 2012. V. 14. No. 1-4. P. 1080–1083 (in Russian).
  7. Neshataeva V.Yu., Pesterov A.O., Korablev A.P. Vegetation of the Eastern Volcanic Belt of Kamchatka (Within the Kronotsky Reserve). St. Petersburg: Marathon, 2021. 67 p. (in Russian).
  8. Pal M. Random Forest Classifier for Remote Sensing Classification. International Journal of Remote Sensing, 2005. V. 26. No. 1. P. 217–222. DOI: 10.1080/01431160412331269698.
  9. Pesterov A.O. Coenotic Diversity and Structure of Vegetation Cover of the Eastern Volcanic Belt of the Kamchatka Peninsula: PhD dissertation of biological sciences. St. Petersburg, 275 p. (in Russian).
  10. Polevshchikova Yu.A., Kirillov V.A. Multiclass Classification of Vegetation Types Using the Google Earth Engine Cloud-Based Platform. Information Technologies and High-Performance Computing: Proceedings of the VII International Scientific and Practical Conference, 2021. P. 18–21 (in Russian).
  11. Puzachenko M.Yu., Kotlov I.P., Chernenkova T.V. Technological Scheme for Monitoring Natural Objects Using Remote Sensing and GIS Technologies. Monitoring the Biological Diversity of Russian Forests. Moscow: Nauka, 2008. P. 347–359 (in Russian).
  12. Shashkov M.P., Ivanova N.V., Ermolov S.A. Usability of Raster Spatial Environmental Datasets as Predictors for Species Distribution Modelling of Earthworms. Proceedings of the International Conference “Mathematical Biology and Bioinformatics”, 2024. P. 228–232 (in Russian). DOI: 10.17537/icmbb24.55.
  13. Sheloukhin O.I., Polkovnikov M.V. Improving the Classification Efficiency of Network Traffic of Mobile Devices Using the Random Forest Algorithm Technologies of the Information Society. Materials of the XIII International Industrial Scientific and Technical Conference, 2019. P. 415–418 (in Russian).
  14. Thanh Noi P., Kappas M. Comparison of Random Forest, K-nearest-neighbor, and Support Vector Machine Classifiers for Land Cover Classification Using Sentinel-2 Imagery. Sensors, 2017. V. 18. No. 1. P. 18. DOI: 10.3390/s18010018.
  15. Tolan J., Yang H.I., Nosarzewski B., Couairon G., Vo H.V., Brandt J., Spore J., Majumdar S., Haziza D., Vamaraju J., Moutakanni T., Bojanowski P., Johns T., White B., Tiecke T., Couprie C. Very High Resolution Canopy Height Maps from RGB Imagery Using Self-Supervised Vision Transformer and Convolutional Decoder Trained on Aerial Lidar. Remote Sensing of Environment, 2024. V. 300. P. 113888. DOI: 10.1016/j.rse.2023.113888.

For citation: Maksimovich N.V., Tarasov A.A., Korablev A.P., Pozdnyakova N.A., Andreeva T.A., Artemeva O.V. Mapping Uzon caldera vegetation using remote sensing data and machine learning. InterCarto. InterGIS. Moscow: MSU, Faculty of Geography, 2025. V. 31. Part 2. P. 83–99. DOI: 10.35595/2414-9179-2025-2-31-83-99 (in Russian)