研究目的
To create annual seamless cloud-free Landsat mosaics for large-area forest change assessment over a 30-year period, addressing challenges like data gaps and radiometric inconsistencies.
研究成果
The method successfully produces seamless, gap-free, radiometrically consistent Landsat composites for forest change assessment, offering state-wide consistency and reduced processing/storage needs, but has limitations in error handling and sensor calibration that could be improved with enhanced algorithms and multi-index approaches.
研究不足
Reliance on a single spectral index (NBR) for breakpoint detection may not capture all disturbances; errors can occur from residual cloud noise misidentified as disturbances; method may fail to capture abrupt changes coinciding with data gaps; large data volumes (4 TB input) require significant processing; not fully calibrated for OLI and ETM+ sensor differences.
1：Experimental Design and Method Selection:
The method involves creating Best Available Pixel (BAP) composites from Landsat scenes, analyzing time-series with Normalized Burn Ratio (NBR) to detect breakpoints (e.g., disturbances), and fitting piece-wise linear regression models to remove gaps and anomalies.
2：Sample Selection and Data Sources:
Surface reflectance products (Tier 1, TM, ETM+ & OLI, <70% cloud-cover, Jan-Mar, 1988-2017) for 19 Landsat tiles covering Victoria, Australia, downloaded from the USGS archive.
3：List of Experimental Equipment and Materials:
Landsat satellite data, software packages LandsatLinkr and Raster in R statistical software.
4：Experimental Procedures and Operational Workflow:
Scenes were ranked by proximity to February 15, merged into BAP composites, NBR time-series created and processed to find breakpoints using criteria (negative deviations >
5：5 times interquartile range), spatial filter applied to remove small disturbances, segmented linear models fitted to optical bands for forested areas, and simple regression for non-forest areas to generate pseudo composites. Data Analysis Methods:
Statistical methods including linear regression, interquartile range for outlier detection, and visual validation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
