Temporal consistency is one of the pre-requisite when two satellite images are used for land cover change detection, since both inter-annual (yearly) and intra-annual (seasonal) changes are quite common for some land covers. Global Land Survey (GLS) images, mainly based on Landsat MSS/TM/ETM+, have been widely used for land cover change detection due to its 40-year global coverage, relatively high quality, 30-80 m spatial resolution, and open policy. However, few works have been conducted to investigate the effect of lower temporal resolution on change detection. There may be two reasons for this oversight: for some applications, land cover change can be detected using classification, which may circumvent the temporal inconsistency problem; for others, the difference resulted from minor temporal change may be regarded as negligible. This paper obtained the actual year and date of GLS Landsat data from Global Land Cover Facility (GLCF) for nominal year (epoch) 1975, 1990, 2000 and 2005, and analyzed date inconsistency effect on change detection using 16-day MODIS-NDVI serials. The investigation showed that real year (year-difference, YD) covered 1975±4, 1990±4.5, 2000±1.5 and 2005±1.5 for epoch 1975, 1990, 2000 and 2005, respectively. And the average date difference (DD) was 47 days, roughly 3 composite periods for 16-day MODIS-NDVI. What do the Landsat YD and DD mean to MODIS-NDVI? For YD, the study in NECT using 2000-2008 MODIS-NDVI showed that, 4 years among 9 years, 7% sparse grasslands have a >7% yearly variance. For DD, when using growth peak (mid-summer) as reference, 16-48 d difference may lead to MODIS-NDVI difference over 0.4 for cropland and grassland, 0.1-0.2 for forest, while <16 d difference will lead to <0.1 MODIS-NDVI difference. However, when using actual MODIS-NDVI date as reference, even a DD <16 d may also lead to MODIS-NDVI difference with seasonal patterns: only 2.93% pixels with a more than 0.1 NDVI difference resulted from DD<16 d during July 12 to August 28, indicating a perfect period for change detection. Yet this number increased to 11.42% during June 10-September 29 (excluding July 12 to August 28), showing a worst period for change detection.
This study concluded that the difference resulted from minor temporal change can not be regarded as negligible in certain cases, and MODIS-like high temporal data can improve change detection using low temporal Landsat-like data, in three aspects: 1) when there are enough high-spatial and low-temporal resolution images, MODIS-like data serials can be used to choose data with optimal time for change detection; 2) when there are not enough such images and second-best data are used, MODIS-like data serials can be used to estimate resulted NDVI difference; and 3) when change is only quantitative and simple classification fails to detect, high temporal bio-physiological parameters such as MODIS-NDVI can be directly used to change detection.