DATimeS

Understanding the spatiotemporal variability of vegetation phenology is key to study the timing of recurring biological events over large areas and different crop types. The success of these studies depends on two factors: frequent observations, and accurate identifications of pheno-phases. Optical remotely sensed data are spatially and temporally discontinuous, with missing values due to cloud cover. Consequently, gap-filling solutions are needed for accurate crop phenology characterization.

The so-called Decomposition and Analysis of Time Series Software (DATimeS) is a stand-alone image processing GUI toolbox written in MATLAB. DATimeS enables to perform different advanced time series tasks for: (1) generating spatially continuous maps from discontinuous data using advanced Machine Learning Regression Algorithms (MLRA) (e.g. Gaussian Process Regression - GPR) and synergy of multiple sensors; and (2) detecting heterogeneous spatial patterns of phenological indicators (i.e., crop key growth stages) throughout multiple seasons.

DATimeS software is structured in multiple modules. The modular architecture, which has been inherited from the ARTMO toolbox, offers the following advantages: (1) it guides the user through the processing steps, i.e. subsequent processing modules are activated once current module has been completed, (2) modules can be easily modified or extended without affecting the main architecture, (3) and new modules with new functionalities can be easily added to the toolbox. An overview of the DATimeS’ modules contained in this first official version (v.1.10) is shown below. 

Figure 1. Hierarchical design of DATimeS.

The first module, ”Input”, is responsible for reading the data. Time series processing for spatiotemporal analysis can take place either on a single pixel, or on a stack of images. For single pixels, a.txt file is required. Imagery can be entered in TIFF or ENVI format from a specific folder. When loaded, the data will appear sorted in the DATimeS main window (Fig. 2).

Figure 2. DATimeS graphical user interface (GUI).

The second module, ”Time Series Analysis”, is in charge of constructing composite images with any time step by applying the gap- filling (e.g. due to clouds) (Fig. 3) and the synergy between different sensors (Fig. 4) using a broad variety of advanced interpolation and smoothing methods, such as MLRA,, double logistic curve, harmonic analysis and conventional methods.

Figure 3. Original and reconstructed time series of LAI using several gap-filling techniques. The example applies different time settings. Interpolated values of time series at a higher sampling frequency (every 20 days) (top). Interpolated values are computed only for embedded missing values in the input time series (bottom). The GPR uncertainties are shown in red shade areas.

Figure 4. Original NDVI time series and their Multi Output GPR predictions of one pixel estimated on the union of Landsat-7, Landsat-8 and Sentinel-2 acquisition dates provided by the MOGP model trained on input time series.

Besides, it enables to calculate all kinds of phenological indicators such as (Fig. 5 and Fig. 6):

• The amplitude (difference between the maximum and the average of the left and right minimum values per season),
• Maximum value (largest value per cycle),
• Day of maximum value (when the largest value per cycle occurs),
• Start of season (SOS),
• End of season (EOS),
• Seasonal integral (area under the curve between SOS and EOS), and
• Length of season (difference between SOS and EOS).

 Figure 5. Automatic identification of some seasonal patterns computed in DATimeS by using the structed LAI curve (red line) with GPR. Red shade area shows the associated GPR uncertainties (standard deviation). Purple and green colors indicate the areas under the curve between SOS/EOS (red angles) and the left/right minimum values, respectively. Blue lines show approximately the length of seasons (LOS). Maximum value (MaxV), day of maximum value (DOM) and amplitude (Amp) are represented with orange dashed lines.

Figure 6. Maps of phenological indicators estimated by using reconstructed Sentinel-2 LAI images with GPR.

The third module, ”Post-processing”, improves the obtained maps by spatial interpolation and provides the possibility to create videos, thus enabling animation of temporal trends.

Altogether, DATimeS is freely available as a powerful image time series software, offering strong potential for crop monitoring, even in cloudy conditions, i.e. by means of gap filling, and in case of persistent clouds with fusion optical data with radar data.

For more information see: promotional and tutorial video!

Novelties:

This table provides an overivew of DATime’S novelties starting from 28/02/2019