Tree Genera Classifications in Spain Using Time-Series Sentinel-2 Data Extracted from Plottosat
Miltiadou, Milto ; Lines, Emily R. ; Grieve, Stuart ; Benito, Paloma Ruiz ; Astigarraga, Julen LU
and Cruz, Verónica
(2024)
2024 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2024
In International Geoscience and Remote Sensing Symposium (IGARSS)
p.4953-4956
- Abstract
PlotToSat is a tool that uses the Python API of Google Earth Engine (GEE) to solve the problem of creating time-series of Earth Observation (EO) data at multiple plot locations spread out within a landscape. The current version supports Sentinel-1 and Sentinel-2. It is a flexible framework though that allows easy addition of EO datasets. Using PlotToSat, we generated a Normalised Difference Vegetation Index (NDVI) time-series at 14,822 plot of the fourth Spanish National Forest Inventory (Spanish NFI). Using the non-parametric k-NN algorithm, we classified the NDVI time-series into the six dominant genera in Peninsular Spain. The classification was cross-validated using the evaluation metric F1-score. We show that re-sampling to create... (More)
PlotToSat is a tool that uses the Python API of Google Earth Engine (GEE) to solve the problem of creating time-series of Earth Observation (EO) data at multiple plot locations spread out within a landscape. The current version supports Sentinel-1 and Sentinel-2. It is a flexible framework though that allows easy addition of EO datasets. Using PlotToSat, we generated a Normalised Difference Vegetation Index (NDVI) time-series at 14,822 plot of the fourth Spanish National Forest Inventory (Spanish NFI). Using the non-parametric k-NN algorithm, we classified the NDVI time-series into the six dominant genera in Peninsular Spain. The classification was cross-validated using the evaluation metric F1-score. We show that re-sampling to create a balanced dataset improved classification results. The F1-score for the most common tree genera was low. This is suspected to occur due to subgroups (of tree species and/or stand structural types) being formed within each large genus class. When reducing data to use an equal number of samples per genus class in the classifier, subgroups of large genus classes become under-represented. PlotToSat processed an estimated 18.3 TB of EO data in less than 24 hours. It is useful for any application requiring EO time-series data from multiple spatially disconnected locations.
(Less)
- author
- Miltiadou, Milto
; Lines, Emily R.
; Grieve, Stuart
; Benito, Paloma Ruiz
; Astigarraga, Julen
LU
and Cruz, Verónica
- publishing date
- 2024
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- forests, genera, machine learning, Sentinel-2, time-series
- host publication
- IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, Proceedings
- series title
- International Geoscience and Remote Sensing Symposium (IGARSS)
- pages
- 4 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 2024 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2024
- conference location
- Athens, Greece
- conference dates
- 2024-07-07 - 2024-07-12
- external identifiers
-
- scopus:85204922933
- ISBN
- 9798350360325
- DOI
- 10.1109/IGARSS53475.2024.10641135
- language
- English
- LU publication?
- no
- additional info
- Publisher Copyright: © 2024 IEEE.
- id
- cee96e36-a2db-4b05-bed0-d56ad46ba5d9
- date added to LUP
- 2025-03-25 13:19:02
- date last changed
- 2025-04-22 13:07:31
@inproceedings{cee96e36-a2db-4b05-bed0-d56ad46ba5d9,
abstract = {{<p>PlotToSat is a tool that uses the Python API of Google Earth Engine (GEE) to solve the problem of creating time-series of Earth Observation (EO) data at multiple plot locations spread out within a landscape. The current version supports Sentinel-1 and Sentinel-2. It is a flexible framework though that allows easy addition of EO datasets. Using PlotToSat, we generated a Normalised Difference Vegetation Index (NDVI) time-series at 14,822 plot of the fourth Spanish National Forest Inventory (Spanish NFI). Using the non-parametric k-NN algorithm, we classified the NDVI time-series into the six dominant genera in Peninsular Spain. The classification was cross-validated using the evaluation metric F1-score. We show that re-sampling to create a balanced dataset improved classification results. The F1-score for the most common tree genera was low. This is suspected to occur due to subgroups (of tree species and/or stand structural types) being formed within each large genus class. When reducing data to use an equal number of samples per genus class in the classifier, subgroups of large genus classes become under-represented. PlotToSat processed an estimated 18.3 TB of EO data in less than 24 hours. It is useful for any application requiring EO time-series data from multiple spatially disconnected locations.</p>}},
author = {{Miltiadou, Milto and Lines, Emily R. and Grieve, Stuart and Benito, Paloma Ruiz and Astigarraga, Julen and Cruz, Verónica}},
booktitle = {{IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, Proceedings}},
isbn = {{9798350360325}},
keywords = {{forests; genera; machine learning; Sentinel-2; time-series}},
language = {{eng}},
pages = {{4953--4956}},
publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
series = {{International Geoscience and Remote Sensing Symposium (IGARSS)}},
title = {{Tree Genera Classifications in Spain Using Time-Series Sentinel-2 Data Extracted from Plottosat}},
url = {{http://dx.doi.org/10.1109/IGARSS53475.2024.10641135}},
doi = {{10.1109/IGARSS53475.2024.10641135}},
year = {{2024}},
}