Unpiloted aerial system (UAS) flight campaigns were conducted at two rangeland sites in Southwestern Montana during the 2018 growing season to classify vegetation and landcover types. A total of nine flights were conducted at the Argenta site and seven at the Virginia City site. To align images in space and time, we used four-dimensional structure from motion (4D SfM) and continued with processing for each flight date based on the full suite of images aligned for the entire growing season. We created dense point clouds, digital terrain models (bare earth), digital elevation models (including vegetation), and orthorectified images for each flight date at each site. We used the orthoimages to calculate the Normalized Difference Vegetation Index (NDVI) for each flight and used the flight at the peak of the growing season to calculate vegetation height and texture. We then used vegetation height and texture, along with different sets of flights as inputs into an Iterative Self-Organized (ISO) unsupervised data analysis algorithm to classify landcover types. We tested four flight frequencies: a single flight, a limited set, spring flights, and biweekly flights using different sets (or subsets) of the flight campaign. For each scenario, we classified the image to identify six functional groups: bare ground, litter, sparse, medium, and dense herbaceous, and sagebrush. For classifications based on multiple flights we tried to further identify subcategories of classes to reflect differences in phenology (timing of green-up and/or senescence).

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

Unmanned Aerial System (UAS) flights were conducted over four stream catchments in Rio Blanco County, Colorado, during the summer of 2016. Two sties had active oil and gas operations within the basin whereas the other two sites did not. Structure from motion (SfM) was used to align raw images and create a dense point cloud, georectified orthoimage, and Digital Elevation Model (DEM) for each basin. A Digital Terrain Model (DTM), or bare earth model, for each basin was created by reclassifying the dense point cloud as either bare ground or other (vegetation, oil and gas infrastructure, etc.) and interpolating the land surface between bare ground points. Ideally, the DTM would always be equal or lower than the DEM; however, the interpolated surface can sometimes be higher than the DEM if bare ground points surround depressions with vegetation or in thick vegetation strands with an undulating surface. Therefore, a final surface model, created by merging the DTM with the DEM for all areas where the DTM was greater than the DEM, was produced for each basin. Lastly, a random forest classification approach was used to classify the orthoimagery on a pixel level into five vegetation/land cover classifications - bare ground, grass, litter, shrub/woody vegetation, and shadow.

In support of mapping ecological conditions (e.g. invasive annual grass) in sagebrush-dominated landscapes of the western United States, we developed weekly (starting from week 7 to week 42 and Week 1 starts January 1 or Day of the year 1 to 7, week 2 is from Day of year 8 to 14, and so on) 30-m cloud-free Normalized Difference Vegetation Index (NDVI) from 2016 to 2019. The data was generated with machine-learning techniques (i.e., regression tree [RT]) and harmonized Landsat and Sentinel -2 (HLS) data. The geographic coverage includes areas in the Great Basin, the Snake River Plain, the state of Wyoming, and contiguous areas. This NDVI collection allows for local-scale detection and analysis such as, fuel breaks in sagebrush ecosystem and wildfire activity, that are not possible with coarse scale datasets (such as 250-m).