Folders: ROOT > ScienceBase Catalog > Pacific Island Ecosystems Research Center > PIERC Public Data Releases > High-Resolution Land Cover Maps of Lāna‘i, Hawai‘i, 2020 > Lāna‘i Landcover Maps ( Show all descendants )
4 results (9ms)
Location
Folder
ROOT _ScienceBase Catalog __Pacific Island Ecosystems Research Center ___PIERC Public Data Releases ____High-Resolution Land Cover Maps of Lāna‘i, Hawai‘i, 2020 _____Lāna‘i Landcover Maps Filters
Date Range
Extensions Types Contacts Categories Tag Types Tag Schemes |
This raster integrates the species-specific and community classifications using a hierarchical approach based on classification certainty. A 0.66 probability threshold was applied, with pixels assigned the finest species-specific class as long as the probability exceeded the threshold. Pixels below the threshold were assigned to the broader community class meeting the threshold. This approach displays the most detailed class possible given a minimum confidence, providing a map that balances specificity and certainty. Please note that to reduce the inherent 'salt and pepper' noise in the final land cover classification map, we applied a 3x3 pixel moving window majority filter to the final classification results.
This raster depicts the distribution of broader community-level vegetation classes across Lāna‘i. To generate this map, the species-specific class probabilities were summed to get total probability of membership in each defined community class. Each pixel was then assigned to the community class with the highest probability. This generalized map allows for an assessment of vegetation patterns at a coarser categorical level across the island. Please note that to reduce the inherent 'salt and pepper' noise in the final land cover classification map, we applied a 3x3 pixel moving window majority filter to the final classification results.
This raster stack contains 15 probability layers representing the pixel-level predicted probability of membership in each species-specific vegetation class from 0 to 1. These probability layers can be used to generate class membership uncertainty maps or probabilistic class cover maps from the model outputs. They provide additional information beyond the discrete categorial land cover assignments.
Categories: Data;
Types: Downloadable,
GeoTIFF,
Map Service,
Raster;
Tags: Conservation,
Hawaiian Islands,
Hawai‘i,
Invasive species,
Land cover,
This raster depicts the distribution of 15 species-specific vegetation classes across the island of Lāna‘i at 2m resolution. It represents the final selected neural network model predictions with expert-adjusted posterior probabilities. Each pixel is assigned to the most likely species-specific class based on the model. Overall and class-specific accuracy assessments indicate this map has generally over 95% accuracy. It provides detailed species-level vegetation mapping to support conservation planning and monitoring. Please note that to reduce the inherent 'salt and pepper' noise in the final land cover classification map, we applied a 3x3 pixel moving window majority filter to the final classification results.
|
|
