Version 4 of the Vegetation Index and Phenology Earth Science Data Records

Friday, 19 December 2014
Armando Barreto-munoz1, Kamel Didan1, Tomoaki Miura2 and Javzandulam Tsend-Ayush3, (1)University of Arizona, Tucson, AZ, United States, (2)Univ Hawaii, Honolulu, HI, United States, (3)University of Hawaii-NREM, Honolulu, HI, United States
This is the culmination of a 6-year effort to create a seamless multi-sensor data record about vegetation dynamic from different synoptic imagers. In this work we devised a set of sophisticated science algorithms to combine these disparate data records into an unique and single dataset. To address the cross sensor continuity problem, several published methods were tested. These methods ranged from a single global linear transfer equation (V1), to a land cover based approach (V2, V3), and to a complex hybrid method that looks at the behavior of each pixel separately and considers the impact of vegetation seasonality by adjusting the continuity by the vegetation dynamic of each phase of the growing season (V4). Because of the 1999 break between AVHRR and MODIS we considered SPOT-VGT data in bridging the two sensors using the periods of overlap. However, we found that due to the lack of an efficient cloud algorithm and accurate QA data in SPOT-VGT the method performed rather poorly. In V4 we matched the AVHRR and MODIS medium term average single pixel temporal profile and created spatially explicit transfer equations. This approach was far superior to all other methods and was adopted in our V4 reprocessing effort.

Furthermore, the issue of data filtering was found to be critical to the overall approach. The stricter the data filtering the better the performance of the continuity algorithm. Filtering lead to spatial gaps that were addressed using a simple linear interpolation algorithm, in case of long temporal gaps we used a long term average substitution technique. All data were assigned per pixel quality information that captured the input quality and processing performance.

This filtering, continuity, and gap filling package was applied to a 30+ year record of daily global observations from the LTDR-V4 AVHRR (1981-1999) and MODIS C5 (2000-2013) to generate records of NDVI, EVI2, and phenology metrics at CMG resolution. Version V4 of these records are available at daily, 7day, 15day, and monthly steps. The Phenology metrics are available at 1 and 3-year cycles. Additional long term average reference data of these records are available at 5, 10, 20 and full record through an interactive “Data Explorer” visualization system [http://vip.arizona.edu/viplab_data_explorer] and separately through the LP DAAC.