Time series analysis in remote sensing
We develop time series methodology for handling of time series data from satellites. Early on we investigated Fourier series (Olsson and Eklundh 1994), but subsequently moved on to curve-fitting methodology that allowed for accurate estimation of phenological parameters and which led to the development of the TIMESAT package (Jönsson and Eklundh, 2002, 2004). Our focus has been on data from coarse resolution sensors (MODIS, AVHRR), but today we also focus on the new ESA Sentinel satellite Sentinel-2, which combines high spatial and temporal resolutions. The new smoothing and fitting algorithms will utilize both the spatial and temporal domains for obtaining noise-free data. We are currently investigating different methods, and have tested our fitting using simulated data from SPOT (Eklundh et al. 2012). One spatio-temporal method we have tested is a based on smoothing splines (Eklundh et al. 2013).
We test our methods against flux-tower data and data from our optical sensor network. The aim is to investigate the ability of Sentinel-2 data to estimate variations in carbon balance and phenology. However, we also foresee many other interesting applications of these time-series data within a range of fields.
We are interested in methodology for managing long time series, and have studied how the spatial domain can increase accuarcy of estimated trend parameters (Bolin et al. 2009). We are also developing trend analysis methodology for accurate description of non-linear trends (Jamali et al. 2014, 2015). These methods are powerful for understanding environmental changes and their causes and consequences.
PolyTrend is a polynomial fitting-based algorithm for classifying trend in time series of satellite vegetation data. It classifies the trends into linear, quadratic, cubic, concealed, and no-trend types. The "concealed trends" are those trends that possess quadratic or cubic forms, but the net change from the start of the time period to the end of the time period hasn't been significant. The "no-trend" category includes simple linear trends with statistically insignificant slope coefficient.
- Web-App: http://polytrend.gis.lu.se/
- R Software: CRAN – PolyTrend: https://cran.r-project.org/web/packages/PolyTrend/index.html
- PolyTrend code in MATLAB & R
DBEST (Detecting Breakpoints and Estimating Segments in Trend) is a program for analysing change in time series of satellite vegetation data. It has two main algorithms:
- change detection algorithm: that detects changes in trend, determines type of the changes (abrupt or gradual), and estimates timing, magnitude, number, and direction of the changes;
- generalization algorithm: that simplifies trend into main features.
The user can set the number of breakpoints or magnitude of greatest changes of interest for detection, and can control the generalization process by setting an additional parameter of generalization-percentage.
How to get DBEST?
- R Software: CRAN – Package DBEST: https://cran.r-project.org/web/packages/DBEST/index.html
- MATLAB code: contact Sadegh Jamali (Sadegh [dot] Jamali [at] tft [dot] lth [dot] se)
- Bolin, D., Lindström, J., Eklundh, L. and Lindgren, F., 2009, Fast estimation of spatially dependent temporal vegetation trends using Gaussian Markov Random Fields. Computational Statistics and Data Analysis, 53, 2885-2896. http://dx.doi.org/10.1016/j.csda.2008.09.017
- Eklundh, L. and Jönsson, P., 2013, A new spatio-temporal smoother for extracting vegetation seasonality with TIMESAT, The 35th International Symposium on Remote Sensing of Environment, 22-26 April 2013, Beijing, China.
- Eklundh, L., Sjöström, M., Ardö, J. and Jönsson, P., 2012, High resolution mapping of vegetation dynamics from Sentinel-2, Proceedings of the First Sentinel-2 Preparatory Symposium, 23-27 April 2012, Frascati, Italy.
- Jamali S., Jönsson P., Eklundh L., Ardö J., and Seaquist J., 2015, Detecting changes in vegetation trends using time series segmentation. Remote Sensing of Environment, 156, 182-195. http://dx.doi.org/10.1016/j.rse.2014.09.010
- Jamali S., Seaquist J., Eklundh L. and Ardö J., 2014, Automated mapping of vegetation trends with polynomials using NDVI imagery over the Sahel. Remote Sensing of Environment 141, 79-89. http://dx.doi.org/10.1016/j.rse.2013.10.019
- Jönsson, P. and Eklundh, L., 2002, Seasonality extraction by function fitting to time-series of satellite sensor data.IEEE Transactions on Geoscience and Remote Sensing, 40, 1824-1832.
- Jönsson, P. and Eklundh, L., 2004, TIMESAT - a program for analysing time-series of satellite sensor data.Computers and Geosciences, 30, 833-845.
- Olsson, L. and Eklundh, L., 1994, Fourier transformation for analysis of temporal sequences of satellite imagery.International Journal of Remote Sensing, 15, 3735-3741.
- Tomov H (2016). Automated temporal NDVI analysis over the Middle East for the period 1982 – 2010. http://lup.lub.lu.se/student-papers/record/8871893
- Wei, Yufei (2016) Developing a web-based system to visualize vegetation trends by a nonlinear regression algorithm https://lup.lub.lu.se/student-papers/search/publication/8882588