An automated compositing method for producing annual clear images from Landsat Collection 2 for annual NLCD production

• The new compositing algorithm is streamlined, pixel-based, and directly run on Landsat Collection analysis ready data. • The compositing method automates extra clouds and clouds shadow detection and gap-filling. • Automatically produce cloud-free and gap-free Landsat composite imagery. Quality image input is fundamental to the quality of derived land cover products. Substantial time and effort are usually required to prepare images. Here, we present a novel and streamlined compositing algorithm that ingests Landsat Collection 2 Analysis Ready Data (ARD) and outputs cloud-free and gap-free composite imagery, which can be directly used for classification. This method leverages and improves the previous National Land Cover Database (NLCD) Virtual Median Value Point (VMVP) compositing method, the first part of the image preparation for NLCD 2019 operational production. The NLCD 2019 image preparation approach includes a second part, a residual cloud and cloud shadow detection and gap-filling method, to produce final cloud-free and gap-free composite imagery. The second part requires one clear reference image for each target year. Additional reference images are needed for producing reasonable observations for perennial ice/snow areas because Pixel QA (Quality Assessment) from ARD has difficulties differentiating ice/snow areas from clouds. Unlike the NLCD 2019 image preparation approach, our new compositing method, which is referred to as Automated VMVP (AVMVP), uses Landsat ARD as the only input and does not require reference images and extra steps. In this method, we developed new spectral filter criteria coupled with counts of clear observations using Pixel QA to identify potential cloud and cloud shadow observations on initially selected observations from the NLCD VMVP compositing algorithm. We also automate “gap-filling” using clear observations retrieved from a maximum of ±2 years around the target year when needed. Finally, a percentile-filtered compositing method was developed for the perennial ice/snow areas. All these steps are streamlined, pixel-based, and directly run on Landsat Collection 2 ARD. We have run successful tests on the conterminous United States (CONUS). Composite images derived from our innovative method were used to produce the CONUS Annual NLCD Collection 1 product suite that covers the period from 1985 to 2023.