Salt marsh mapping based on a short-time interval NDVI time-series from HJ-1 CCD imagery

Abstract:

Salt marshes are regard as one of the most dynamic and valuable ecosystems in coastal zone. It is crucial to obtain accurate information on the species composition and spatial distribution of salt marshes in time since they are experiencing tremendous replacement and disappearance. However, discriminating various types of salt marshes is a rather difficult task because of the strong spectral similarities. In previous studies, salt marsh mappings were mainly focused on high-spatial and hyperspectral resolution imageries combined with auxiliary information but this method can hardly extend to a large region. With high temporal and moderate spatial resolutions, Chinese HJ-1 CCD imagery would not only allow monitoring phenological changes of salt marsh vegetation in short-time intervals, but also cover large areas of salt marshes. Taking the middle coast of Jiangsu (east China) as an example, our study first constructed a monthly NDVI time-series to classify various types of salt marshes. Then, we tested the idea of compressed time-series continuously to broaden the applicability and portability of this particular approach. The results showed that (1) the overall accuracy of salt marsh mapping based on the monthly NDVI time-series reached 90.3%, which increased approximately 16.0% in contrast with a single-phase classification strategy; (2) a compressed time-series, including NDVI from six key months (April, June to September, and November) demonstrated very little decline (2.3%) in overall accuracy but led to obvious improvements in unstable regions; (3) Spartina alterniflora identification could be achieved with only a scene NDVI image from November, which could provide an effective way to regularly monitor its distribution. Besides, by comparing the calibrated performance between HJ-1 CCD and other sensors (i.e., Landsat TM/ETM+, OLI), we certified the reliability of HJ-1 CCD imagery, which is expected to pave the way for laws expansibility from this imagery.