Philipp Gärtner

Evaluating the temporal stability of synthetically generated time-series for crop types in Central Germany

The presented work is evaluating the temporal stability of STARFM generated time-series in an intensively agriculturally used area in Central Germany. 10 Landsat 5 or 7 scenes from 2011 used to generate a daily synthetically time-series, based on the MODIS terra product (500 m). Similarly, 13 RapidEye scenes were acquired. The synthetic Landsat product was correlated to the RapidEye imagery on a class-wise basis for the 12 crops that most frequently occur in the region. The results vary strongly between the Day of Year and the crop type. As conclusion it can be stated that the pairing date of a synthetic scene should not deviate strongly from the date of interest within the time-series. Moreover, classes with a stable phenological development can be predicted with a higher precision compared to classes with rapid changes mostly due to multiple harvest or swath per year.

Monitoring forest recovery with change metrics derived from Landsat time series stacks

Ecological restoration of degraded riparian Tugai forests in north-western China is a key driver to combat desertification in this region. The Chinese government launched its largest environmental restoration project in 2000 by transporting 27.98 × 108 m3 (until 2011) of fresh water from an upstream dam through the dried up lower reaches of the Tarim until the end lake. The diverted water recharged the groundwater, which is considered to be the main driver for the apparent forest recovery. The presented study evaluates the long term trend in forest growth using a time series of 14 years from the Landsat satellites. Three different vegetation metrics were used as input for the pixel based trend analysis. The Theil-Sen estimator served as trend indicator. The results give restoration managers an important insight of the vegetation response and serve as decision support during the restoration project.

Prediction of NDVI for grassland habitats by fusing RapidEye and Landsat imagery

Data fusion is currently one of the most frequently implemented approaches for dealing with inconsistent timesseries. However, approaches to fuse multi-temporal satellite data with higher spatial resolution for the analysis of smaller landcover classes are still missing. Therefore, in this study the ESTARFM algorithm was used to fuse high spatial resolution RapidEye images with medium resolution acquisitions from Landsat 5 TM. The main objective was to predict NDVI values for heterogeneous semi-natural grassland classes for the year 2011. It can be shown, that it is possible to use the ESTARFM algorithm for predicting valuable and accurate NDVI values of small heterogeneous land-cover classes. However, the quality of the estimations depends on the temporal quality of the used input pairs and the date of prediction.