Wed, Aug 17, 2022: 8:30 AM-8:45 AM
520E
Background/Question/MethodsHistorical land use, land cover and vegetation dynamics are critical determinants of current ecosystem processes and structure as well as biodiversity patterns observed at ecological network sites. Unfortunately, obtaining spatially and temporally detailed information on past land cover changes and vegetation disturbances can be difficult and labor intensive. Remote sensing can assist in this task, at least for the last few decades when systematically acquired remotely sensed datasets are available. We present a set of spatial data layers of vegetation disturbance detection and attribution covering the conterminous United States at 30 m resolution. We first processed Landsat data acquired since 1984 in Google Earth Engine through the LandTrendr temporal segmentation algorithm. We then used the USGS LCMAP (Land Change Monitoring, Assessment, and Projection) reference data product in a Random Forest classification to classify each Landsat pixel at a yearly timescale. This was achieved with a two-stepped approach: first a binary disturbance detection, then a classification of disturbance agent (e.g., harvest, fire, hydrology) for disturbed pixels.
Results/ConclusionsOur disturbance product consists of yearly data layers for the period 1984-2021, and provides for each 30 m CONUS pixel and year, an indication whether the pixel has undergone a disturbance event or not, and which disturbance agent was attributed to the event. We additionally provide a probability estimate for the binary disturbance detection product and the attribution product. In addition to the yearly disturbance and attribution layers we provide summary layers and additional layers of disturbance regime for the entire period 1984-2021, including frequency of disturbances, time since last disturbance, and duration of disturbances. We also provide the open and reproducible workflow used to generate these data products in R and Google Earth Engine. Landsat-derived layers of disturbance detection and attribution for the contiguous United States since 1984 will be made freely accessible through the Environmental Data Initiative (EDI) repository, as Google Earth Engine assets, and/or through our project website. We will present vignettes on our project’s GitHub (https://github.com/NEON-biodiversity) to show users how to easily access and process this large dataset.
Results/ConclusionsOur disturbance product consists of yearly data layers for the period 1984-2021, and provides for each 30 m CONUS pixel and year, an indication whether the pixel has undergone a disturbance event or not, and which disturbance agent was attributed to the event. We additionally provide a probability estimate for the binary disturbance detection product and the attribution product. In addition to the yearly disturbance and attribution layers we provide summary layers and additional layers of disturbance regime for the entire period 1984-2021, including frequency of disturbances, time since last disturbance, and duration of disturbances. We also provide the open and reproducible workflow used to generate these data products in R and Google Earth Engine. Landsat-derived layers of disturbance detection and attribution for the contiguous United States since 1984 will be made freely accessible through the Environmental Data Initiative (EDI) repository, as Google Earth Engine assets, and/or through our project website. We will present vignettes on our project’s GitHub (https://github.com/NEON-biodiversity) to show users how to easily access and process this large dataset.