Development of the LCMAP land cover mapping and change monitoring data set for Alaska
Abstract
The U.S. Geological Surveys Land Change Monitoring, Assessment, and Projection (LCMAP) has released a series of products to characterize changes in land cover, use, and condition across the conterminous United States (CONUS). These products were generated using Landsat Analysis Ready Data (ARD) and Continuous Change Detection and Classification (CCDC) algorithm. ARD includes data from the TM aboard Landsats 4 and 5, Landsat 7 ETM+, and Landsat 8 OLI. The data have been processed to the highest level of geometric and radiometric quality. The CCDC algorithm estimates statistical breaks and models time series of surface reflectance that is assumed to represent stable land cover. The choice of harmonic model is meant to capture the phenological behavior of the surface. The model coefficients are used as part of a feature set for supervised classification for land cover mapping. In this study, we present our work on preparing the LCMAP products for Alaska. Compared to CONUS, Alaska presents distinct challenges in climate, frequency of cloud cover, and observation density. For example, Alaska has a long term of snow cover in winter and frequent cloud contamination in summer, making it difficult to obtain valid Landsat observations for CCDC. Besides, orbit overlap causes ARD availability to be more varied than in CONUS because satellites have much more frequent overpasses in some regions than others. Moreover, the land cover types are also distinct from CONUS. It is thus important to customize CCDC for better change detection and land cover mapping in Alaska. We first optimized the parameters of the CCDC change detection method using a stochastic gradient descent method named Adam. The parameters being optimized govern the sensitivity of the change detection algorithm, which operates on thresholds controlled by several tunable parameters. The optimization procedure examined change detection results by comparing them with existing fire extent and lake change products. Second, we mapped annual land cover from 2000 to 2020 using the NLCD 2011 land cover product as training data. We mapped land cover in different levels of the classification scheme and evaluated the agreement of our preliminary land cover with the NLCD products to investigate the capability of the current classification procedure.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.B35D1452Z