Project Funding: 2014 - 2018

NRA: 2013 NASA: Terra and Aqua: Algorithms--Existing Data Products   

Funded by NASA

Abstract:
The MODIS Vegetation Continuous Fields (VCF) product (MOD44B) is an annual characterization of ground cover represented as three fractional components: percent tree, non-tree vegetation, and non-vegetated cover at a spatial resolution of 250 m. This product is derived from custom 16-day composites of daily MODIS satellite imagery (MOD44C). MOD44B and MOD44C are currently running in the MODIS Adaptive Processing System (MODAPS) at NASA Goddard Space Flight Center with distribution of user products via the Land Processes DAAC (LP-DAAC). The VCF product has been adopted by a significant and growing segment of the earth science community, including use in biophysical modeling, hydrology, fire ecology, habitat studies, and agricultural monitoring. The MODIS VCF product received funding under the former ROSES 2009 A.41 Algorithms”Existing Data Product Refinement topic area under the title "Enhanced Landcover Products from MODIS: Vegetation Continuous Fields Maintenance and  Refinements." We are seeking additional funding for the Collection 6 (C6) refinements described below. These refinements will add substantially to the accuracy of the output products, improve error and uncertainty estimates, and result in faster runtimes. Because the VCF products require a complete year of reflectance and thermal data before they can be run, they are produced and tested at the end of the processing stream. MODIS collection 6 (C6) processing will start at the end of June 2013, with the first VCF test products available within one to two months. We expect updates to code and training to be necessary once test products are available. While a few of the necessary changes can be coded and tested during the current grant period, we will need further time to evaluate our products, add training and make necessary code updates. We are receiving feedback from expert users using our Collection 5 products, and there are areas where they have indicated that the products need improvement. In most cases these areas occur in ecosystems which are underrepresented in our training data. We plan to further evaluate our training dataset to identify additional areas that do not have adequate training coverage. We propose to add training data for those areas, then reevaluate and improve our selection of metrics to take full advantage of the new data. We also intend to improve training data in areas with dense forest cover. Our oldest training data has a maximum value of 80% tree cover due to the method used to create it. We will update the 80% data values to reflect the actual ground cover. With early MODIS data it was not possible to distinguish between 80% and 100% tree cover using the MODIS reflectance bands. The improvements in upstream MODIS data products in subsequent collections will allow this determination to be made and give a better estimate of tree cover. Additional improvements need to be made in frequently cloudy tropical regions. These will require algorithm changes. For most of these pixels there are one or more non- cloudy "looks" in a year with sufficient information to determine vegetation cover with good accuracy. Our current regression tree models, expecting a wide range of good- quality metrics, fail to give good estimates when most of the data is cloudy. We will add an alternate regression tree to the code to handle this missing data problem. Our code is currently set up to make it easy to test different metrics and training configurations. Once our metrics selection is finalized and algorithm changes have tested successfully, we propose to streamline the code for better runtimes. Finally, we plan to evaluate the VCF products using independent high-resolution data. These will give us improved error and uncertainty estimates.

2015 NASA Carbon Cycle & Ecosystems Joint Science Workshop Poster(s)

• Accelerated Deforestation in the Humid Tropics from the 1990s to the 2000s   --   (Do-Hyung Kim, Joseph O. Sexton, John R. Townshend)   [abstract]