|M.Sc Student||Mozhaeva Sofia|
|Subject||Time-Series Analysis of Land-Cover Using Spectral Indices|
Derived from Remotely Sensed Imagery
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Maxim Shoshany|
|Full Thesis text|
Land-cover and land-use (LCLU) change is one of the most significant features of the global environmental transformation. Monitoring LCLU changes provides better understanding of the processes taking place.
Change detection is the process of identifying variations in surface conditions by observing it at different times. Remote sensing provides cost-effective means for monitoring change over wide regions. The concept of using remotely-sensed images for change detection is that LCLU changes will cause changes in surface reflectance recorded by a sensor.
Over recent decades, a variety of change detection methods were developed, most of them focus on bi-temporal change detection and very few deal with the problem of multi-temporal detection of change.
The aim of this study is to develop a multi-temporal change detection method based on time-series analysis in addition to examining the variability of several vegetation indices such as NDVI and fractional abundance of vegetation. The research is based on the concept of identifying changes in a relation to the general trend determined from time-series of these indices, calculated using moving averages.
For this purpose, 16 Landsat TM images of a Mediterranean to arid transition zone in central Israel, captured between 1994 and 2011, were used, while images from 2001 and 2002 were excluded from the study due to cloud cover.
The first step in the research methodology employed moving average smoothing. Afterwards, ISODATA classification algorithm was applied on the time-series of moving averages in order to identify the main types of temporal trends, where each class represents a different trend of change. Statistical post-classification analysis of each multi-temporal class facilitated determination of the mean, standard deviation and histograms of change variations of a specific class. These parameters were used to determine whether a pixel deviates significantly from temporal trend class. Preliminary analysis of the results indicated that fractional abundance of vegetation is a more suitable indicator compared to NDVI.
This procedure was conducted separately for three areas along the Mediterranean to arid transition zone: Avisur Highland with mean annual rainfall of 450 mm/year, Amazia area with 350 mm/year, and Lehavim with 250 mm/year. 10 to 12 temporal trends were identified for each area, where each trend represents variation of discrete levels of vegetation cover.
Results indicate that the developed algorithm detected multi-temporal changes efficiently. The algorithm was especially effective to detect abrupt changes, such as the forest disappearance due to fire in the Amazia area and changes in agricultural fields in Avisur Highland and Lehavim area, which occurred as a result of transition from vegetation to non-vegetation areas, and vice-versa. Thus, it can be concluded that fractional abundance of vegetation can provide a useful vegetation index for change monitoring. Altogether, the proposed method has successfully identified changes despite of a great amount of noise present in the images.