Using the Cloud to Collaboratively Mind the Gap between High and Medium-Resolution Tree Cover Data

The U.S. Forest Service (USFS) builds and maintains the tree canopy cover (TCC) component of the National Land Cover Database (NLCD). The NLCD is Landsat-based and available at 30-m resolution for the conterminous United States, coastal Alaska, Hawai'i, Puerto Rico, and the U.S. Virgin Islands.

High-resolution datasets have become accessible in recent years due to wider availability of earth observation data and increased computing capacity. For example, NOAA has produced 1-m land cover datasets (including a tree class) for the Coastal Change Analysis Program. State agencies also have high-resolution land cover data available for analysis.

The USFS receives inquiries from stakeholders and program planners about how the NLCD-TCC and higher-resolution data compare. To support the USFS in decision-making related to NLCD-TCC efforts, we conducted a comparative assessment of standard USFS-produced NLCD-TCC datasets and several aggregated high-resolution land cover datasets. Specifically, we aggregated the tree class of the high-resolution land cover data to create a 30-m dataset with each pixel's assigned value representing the pixel's proportion of tree cover. We analyzed the results to identify differences and similarities across several ecoregions in the United States. For example, in the mid-Atlantic region, the average per-pixel difference between NLCD-TCC and the aggregated high-resolution tree cover data was 20%. We also compared the 30-m datasets to independent estimates of TCC derived from photo-interpretation of high-resolution imagery.

Our analysis was cloud-based and performed in Google Earth Engine (GEE). Common access to datasets and code via cloud computing eased collaboration among the authors, who are located in different states. If the datasets included in our analysis were not already available in GEE, we uploaded the datasets into a shared GEE Asset collection. Use of powerful cloud computing resources over desktop computing facilitated rapid aggregation of numerous statewide and regional high-resolution datasets, which allowed us to reduce data processing time and focus more time and effort on the analysis and comparison of the datasets. Results from our analysis will inform near- and long-term planning efforts and decision-making for USFS tree canopy cover data production.