Long-Term Time Series Analysis of Land Cover Changes in an Arid Environment Using Landsat Data:(A Case Study Of Hamoun Biosphere Reserve, Iran)

Change detection of lakes is important to monitor ecosystem health and wind erosion process in arid environments. The main purpose of this research is to evaluate unsupervised classification based on vegetation indices to monitor Land cover changes (LCCs). The Hamoun Biosphere Reserve is located in the east of Iran and is considered one of the most important wetlands in the center of the Iran Plateau. To detect land cover changes, using Landsat images from the 1990s, 2000s, 2010s and 2020s ground control points (GCP) and spectral profiles, four major land cover classes were obtained (sparse vegetation, dense vegetation, bare land, and water bodies). To create AOIs, the pure pixels were selected using obtained spectral profiles of the main land types by GCPs in 2020. The separability of representative AOIs by classes was examined by Jeffries–Matsushita distances and scattering ellipse parameters. A maximum likelihood classifier (MLC) was applied to Landsat images in 2020 with an overall accuracy of 93% and a Kappa statistic of 0.90. Subsequently, based on Soil Adjusted Vegetation Index (SAVI) maps, as additional input data, unsupervised classification was used to classify the same images in 2020.  The observed accuracy and kappa statistic of the used classification technique was up to 0.91 and 0.89 respectively. The finding indicated that in 2000, the area of arid land increased (90% of all areas) and became a major land use type, whereas water bodies (74% of all areas in the 1990s) reached zero in this year. Yearly water body changes revealed a severe dryness condition in this wetland. After 2000, in most cases in subsequent years, the water body completely dried up and in the seasonally flooded years, it did not exceed 10% of the total wetland’s area. On the other hand, before 2000, on average, 60% of the wetland’s area was dominated by the water class. Our study showed that in the time series without GCP for monitoring past changes, an unsupervised SAVI-based technique could provide acceptable accuracy in this region.