Publications

Wang, YT; Yan, GJ; Xie, DH; Hu, RH; Zhang, H (2022). Generating Long Time Series of High Spatiotemporal Resolution FPAR Images in the Remote Sensing Trend Surface Framework. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 60, 4401915.

Abstract
To improve our capacity to map long-term vegetation dynamics in heterogeneous landscapes, this study proposed a new prior knowledge-based spatiotemporal enhancement method, namely, PK-STEM, to fuse MODIS and Landsat FPAR products following the remote sensing trend surface framework. PK-STEM uses historical Landsat FPAR images as prior knowledge and fuses them with new satellite-derived FPAR data. PK-STEM can work in three modes: 1) using only MODIS data; 2) using only Landsat data; and 3) using both MODIS and Landsat data. This study retrieved FPAR from Landsat images using a scaling-based method and tested the performance of PK-STEM in a regional application. For the entire year of 2012, we compared the performance of PK-STEM in different modes and with that of two typical spatiotemporal fusion methods, the enhanced spatial and temporal adaptive reflectance model (ESTARFM) and unmixing-based linear mixing growth model (LMGM). Then, a long time series FPAR data set at 30-m resolution and eight-day intervals was generated for 13 years (2000x2013;2012). Our results show that PK-STEM in mode III is the most robust and accurate (root mean squared error (RMSE) x003D; 0.062; mean $R = 0.851$ ) among the three modes and more accurate than ESTARFM (mean RMSE x003D; 0.065; mean $R = 0.776$ ) and LMGM (mean RMSE x003D; 0.074; mean $R = 0.734$ ). For the 12 years (2000x2013;2011), PK-STEM also achieves high accuracies with mean RMSE x003D; 0.066 and $R = 0.938$ . PK-STEM is very flexible with a continual update mechanism and is efficient for long time series applications.

DOI:
10.1109/TGRS.2021.3067913

ISSN:
1558-0644