dtwSat: An R Package for Land Cover Classification Using Satellite Image Time Series

Maus, V. ORCID: https://orcid.org/0000-0002-7385-4723 (2017). dtwSat: An R Package for Land Cover Classification Using Satellite Image Time Series. In: EO Open Science 2017, 25-28 September 2017, Frascati, Italy.

[thumbnail of presentation180.pdf]
Preview
Text
presentation180.pdf - Published Version
Available under License Creative Commons Attribution.

Download (6MB) | Preview

Abstract

Open access to satellite data has boosted the development of new approaches to quantify and understand Earth's changes. The large spatiotemporal availability of satellite imagery, for example, has improved our capability to map and monitor land use and land cover changes over vast areas. Given the open availability of large image data sets, the Earth Observation community would get much benefit from methods that are openly available, reproducible and comparable. This paper presents the R package dtwSat, which provides an implementation of the Time-Weighted Dynamic Time Warping (TWDTW) method for land cover mapping using sequences of multi-band satellite images. Methods based on Dynamic Time Warping (DTW) are suitable to handle irregularly sampled and out-of-phase time series, which is frequently the case of those from remote sensing. TWDTW algorithm has achieved significant results using MODIS, Landsat, and Sentinel-2 time series to classify natural vegetation and crop types in different regions. Using existing R packages as building blocks dtwSat supports the full cycle of land cover classification using satellite time series, ranging from selecting temporal patterns to visualizing and assessing the results. To handle the satellite images, dtwSat uses well-known data structures from the R package raster, which offers the option to work with large raster data sets stored on disk instead of loading into memory (RAM) at once. The current version of the dtwSat package provides pixel-based time series classification, i.e., each time series is processed independently from each other, and therefore, the code is easily parallelizable. dtwSat is open source and distributed under a GNU General Public License GPL (≥ 2). A binary version is available from the Comprehensive R Archive Network (https://cran.r-project.org/web/packages/dtwSat) and the development version from GitHub (https://github.com/vwmaus/dtwSat). Future versions of the package envisage new features to reduce border effects, increase spatial homogeneity (i.e., reduce the called 'salt and pepper effect') and improve the temporal consistency of land cover transitions. dtwSat makes it straightforward to apply and compare the TWDTW approach with other methods, contributing to rapid advance automated and semi-automated methods to analyze satellite time series.

Item Type: Conference or Workshop Item (Paper)
Research Programs: Ecosystems Services and Management (ESM)
Depositing User: Luke Kirwan
Date Deposited: 16 Jan 2018 14:22
Last Modified: 27 Aug 2021 17:29
URI: https://pure.iiasa.ac.at/15064

Actions (login required)

View Item View Item