WEBVTT 00:00:04.916 --> 00:00:05.583 Welcome to 00:00:05.583 --> 00:00:10.958 the EROS User Experience webinar series, where we talk to staff at EROS 00:00:10.958 --> 00:00:16.041 to learn more about the data, tools and services coming out of the USGS 00:00:16.083 --> 00:00:20.041 Earth Resources Observation Science or EROS Center. 00:00:20.583 --> 00:00:23.708 Today's webinar is titled RCMAP Long-term 00:00:23.708 --> 00:00:27.125 and Near-Real time Rangeland Vegetation Monitoring. 00:00:27.208 --> 00:00:31.208 I'm your host, Danielle Golon, the remote sensing user services 00:00:31.208 --> 00:00:34.333 lead here at EROS in Sioux Falls, South Dakota. 00:00:34.375 --> 00:00:36.416 The time is currently 12 p.m. central. 00:00:36.416 --> 00:00:38.125 So let's get started. 00:00:38.125 --> 00:00:42.000 First, a few logistics to ensure the best audio experience. 00:00:42.041 --> 00:00:44.333 All participants have been muted. 00:00:44.333 --> 00:00:47.291 If you have any questions or comments during the webinar, please 00:00:47.291 --> 00:00:50.833 use the Q&A feature, which can be found under the more settings 00:00:50.833 --> 00:00:54.750 if you do not see it currently or if this feature does not work for you. 00:00:54.791 --> 00:00:58.000 Please feel free to enter your questions or feedback into the chat, 00:00:58.000 --> 00:01:01.166 and we will address the questions and feedback at the end of the webinar. 00:01:01.375 --> 00:01:05.375 If you cannot use the Q&A feature or the chat due to your organization 00:01:05.375 --> 00:01:10.541 settings, please feel free to email your questions to custserv@USGS.gov. 00:01:10.958 --> 00:01:20.875 That is custserv@usgs.gov and we will answer them there. 00:01:21.458 --> 00:01:23.416 Today's webinar is being recorded. 00:01:23.416 --> 00:01:24.916 The recording will be available 00:01:24.916 --> 00:01:28.875 later on the USGS EROS User Experience webinar webpage. 00:01:28.916 --> 00:01:32.916 We'll post a link to that webpage in the chat later on, and today's webinar 00:01:32.916 --> 00:01:36.708 will consist of a presentation and then a question and answer session. 00:01:37.250 --> 00:01:39.458 Today's speaker is Matthew Rigge. 00:01:39.458 --> 00:01:41.458 Matthew is an ecologist with the U.S. 00:01:41.458 --> 00:01:44.083 Geological Survey here at EROS. 00:01:44.083 --> 00:01:46.833 His work focuses on quantifying the current 00:01:46.833 --> 00:01:50.875 and historical condition of rangelands across the western United States. 00:01:51.208 --> 00:01:54.625 He accomplishes this by producing and analyzing fractional 00:01:54.625 --> 00:01:58.375 vegetation cover maps derived from remote sensing and AI. 00:01:59.000 --> 00:02:03.375 These data sets support investigations into the drivers of change in rangeland 00:02:03.375 --> 00:02:08.416 ecosystems, including long term trends, interannual weather variability, 00:02:08.583 --> 00:02:11.583 and the effects of land management practices. 00:02:11.625 --> 00:02:14.625 We also have several EROS staff members from User Services 00:02:14.625 --> 00:02:18.458 and the RCMAP team here at EROS on the line, to help 00:02:18.458 --> 00:02:22.375 answer any questions you may have after Matthew has finished his presentation. 00:02:22.416 --> 00:02:24.958 Again, please feel free to add your questions or feedback 00:02:24.958 --> 00:02:28.958 throughout the webinar using that Q&A feature in teams or the webinar chat. 00:02:29.375 --> 00:02:32.083 We'll try to answer all questions within the time allotted, 00:02:32.083 --> 00:02:35.333 but if we're not able to address your question during the final Q&A 00:02:35.333 --> 00:02:38.416 portion of the webinar, we will follow up with you offline. 00:02:38.750 --> 00:02:41.458 And if there's a future webinar topic you would like us to cover. 00:02:41.458 --> 00:02:43.833 Please feel free to suggest that as well. 00:02:43.833 --> 00:02:47.875 With that, it's my pleasure to introduce today's speaker, Matthew Rigge. 00:02:47.958 --> 00:02:49.291 Take it away, Matthew. 00:02:49.291 --> 00:02:52.375 Thanks, everyone for taking your lunch hour 00:02:52.375 --> 00:02:57.250 or they're about to learn more about RCMAP, the Rangeland Condition 00:02:57.250 --> 00:03:00.333 Monitoring Assessment and Projection project. 00:03:00.375 --> 00:03:05.958 I'm going to talk today about our long term monitoring data and near real time. 00:03:05.958 --> 00:03:09.958 This work is supported by both the BLM, the Bureau of Land Management 00:03:09.958 --> 00:03:13.250 and the USGS National Land Imaging Program. 00:03:13.333 --> 00:03:18.458 Also would like to acknowledge support from our KBR contract here at EROS. 00:03:18.458 --> 00:03:23.958 Brett, Dev, Logan, Trenton, Jordan, Audra and Steve with RCMAP. 00:03:24.000 --> 00:03:28.375 Our goal is to capture both gradual and abrupt 00:03:28.375 --> 00:03:32.583 changes in rangeland condition over the Landsat time series. 00:03:32.916 --> 00:03:35.916 So that's 1985 until present. 00:03:36.125 --> 00:03:41.083 A lot of that change is related to interannual weather variation, 00:03:41.250 --> 00:03:46.833 but also disturbance, land management practices, all sorts of things 00:03:46.833 --> 00:03:51.875 going on in the landscape that are causing change worthy of detection. 00:03:51.916 --> 00:03:56.875 The overall goal of the project is that monitoring changes to vegetation 00:03:56.875 --> 00:04:00.166 composition in the past, evaluating 00:04:00.375 --> 00:04:03.375 the effect of management practices 00:04:03.833 --> 00:04:06.333 helps us to better target future 00:04:06.333 --> 00:04:09.833 improvements and better understand how the landscape is changing, 00:04:09.958 --> 00:04:13.833 ultimately improving wildlife habitat and landscape health. 00:04:14.083 --> 00:04:18.458 So that's one of the real emphasis of this project, is to make the data 00:04:18.833 --> 00:04:22.500 as tangible to real world applications as possible. 00:04:22.500 --> 00:04:26.958 So we do this by combining a series of field observations 00:04:26.958 --> 00:04:32.541 with machine learning and AI Landsat data, a bunch of other ancillary data 00:04:32.625 --> 00:04:36.333 sets like topography, vegetation indices, etc. 00:04:36.625 --> 00:04:40.708 to make a series of ten cover maps 00:04:41.041 --> 00:04:44.958 on an annual basis across the western US. 00:04:45.250 --> 00:04:50.000 So we have percent cover maps for bare ground litter or basis 00:04:50.041 --> 00:04:53.541 perennial, herbaceous annual or basis shrub, 00:04:54.125 --> 00:04:56.666 sagebrush, non sagebrush, 00:04:56.666 --> 00:04:59.833 shrub tree and the height of shrubs. 00:04:59.875 --> 00:05:04.000 So this is an example of what the data would look like for a pixel 00:05:04.041 --> 00:05:08.375 on the landscape were operating at the Landsat scale, which is 30 meter 00:05:08.666 --> 00:05:12.791 spatial resolution and gives us a pretty robust understanding 00:05:12.791 --> 00:05:18.125 of how that pixel has changed through time and helps us to sort of understand 00:05:18.291 --> 00:05:21.291 what the driving factors of that change are. 00:05:21.291 --> 00:05:24.333 Provide an example here in Montana, 00:05:24.375 --> 00:05:29.833 the Bridger Fire, which occurred in 2020, showing the benefit of our sea 00:05:29.875 --> 00:05:34.416 map data compared to some other ways of understanding change. 00:05:34.416 --> 00:05:37.833 So we have on the top left a Landsat 00:05:37.875 --> 00:05:41.250 image pre fire and top right post-fire. 00:05:41.708 --> 00:05:44.250 We can clearly see the burn impact. 00:05:44.250 --> 00:05:49.666 These darker brown areas within the burn perimeter in red indicating burn. 00:05:49.708 --> 00:05:54.666 The image does provide some context into how the landscape has changed. 00:05:55.041 --> 00:05:59.666 But looking at the sea map tree cover from the same years 19 and 21, 00:06:00.250 --> 00:06:03.541 we can understand more in depth 00:06:04.000 --> 00:06:07.125 what the magnitude of that change was. 00:06:07.125 --> 00:06:10.541 Also, compared to data like NDVI. 00:06:11.375 --> 00:06:14.375 That same fire scar here in Montana, 00:06:14.791 --> 00:06:17.791 the fire indicated with that dashed red line, 00:06:18.375 --> 00:06:21.916 we can see that the year after fire in 21, 00:06:21.958 --> 00:06:26.083 the NDVI was lower than pre fire, 00:06:26.250 --> 00:06:28.833 but it quickly recovers and it really doesn't provide 00:06:28.833 --> 00:06:33.250 a whole lot of information on how that landscape changed post-fire. 00:06:33.500 --> 00:06:36.500 Compare that again to our see map 00:06:37.041 --> 00:06:39.375 fire indicated in that dashed red line. 00:06:39.375 --> 00:06:44.958 We can see tree cover has gone down a lot post-fire 00:06:46.041 --> 00:06:49.083 and or cover has gone up. 00:06:49.291 --> 00:06:55.083 So we can understand how the vegetation community has redistributed post-fire. 00:06:55.083 --> 00:06:58.916 In this case, with herbicides increasing. 00:07:01.375 --> 00:07:02.333 Training data 00:07:02.333 --> 00:07:06.666 is really important to produce a model like this. 00:07:06.708 --> 00:07:12.416 One of the unique things about our map is our series of high resolution 00:07:12.416 --> 00:07:16.666 satellite image sites, where we sent out field 00:07:16.666 --> 00:07:22.125 crews to the ground and observe the cover of various patches of vegetation 00:07:22.666 --> 00:07:26.333 on different types of geology, all the different 00:07:26.333 --> 00:07:30.125 permutations of things that could exist on these image footprints, 00:07:30.125 --> 00:07:34.208 and then use those data to predict 00:07:34.208 --> 00:07:37.291 a two meter scale, a high resolution scale 00:07:37.541 --> 00:07:40.541 map of all the vegetation components I listed before, 00:07:41.125 --> 00:07:43.958 and then scale that up to 30m 00:07:43.958 --> 00:07:47.125 to serve as a key element of our training data. 00:07:47.166 --> 00:07:50.625 So those sites are indicated by all of these red dots here, 00:07:50.625 --> 00:07:55.208 which are approximately 15 by 15 kilometer image footprint. 00:07:55.250 --> 00:08:00.458 Those would be worldview two, worldview three primarily satellite images. 00:08:00.583 --> 00:08:06.625 So the benefit of that approach is that it gives us a huge N 00:08:06.750 --> 00:08:12.541 to operate with nearly 69,000,030 meter scale training points to operate with. 00:08:12.875 --> 00:08:18.541 With a model like this, we're using a neural net model, really data hungry. 00:08:18.541 --> 00:08:21.833 So having a high end is really important here. 00:08:21.916 --> 00:08:25.750 We also pull in all of the Bureau 00:08:25.750 --> 00:08:30.125 of Land Management aim the analysis inventory and monitoring data 00:08:30.166 --> 00:08:33.291 set and their LMF data as well. 00:08:33.875 --> 00:08:36.625 So these data are the gold 00:08:36.625 --> 00:08:39.916 standard monitoring data collected by the BLM. 00:08:40.125 --> 00:08:45.750 Most of those plots are randomly located across the domain in the west western 00:08:45.750 --> 00:08:50.958 part of the US, designed to represent essentially a single Landsat pixel 00:08:51.583 --> 00:08:54.583 measured with boots on the ground as well. 00:08:54.916 --> 00:08:58.833 So that's definitely a key data set we've brought in here as well, 00:08:59.333 --> 00:09:03.666 pulling in data from land fire, their public reference database, other 00:09:03.791 --> 00:09:09.125 sort of ad hoc additional data sets as we can get Ahold of it from the G. 00:09:09.125 --> 00:09:09.416 Biff. 00:09:09.416 --> 00:09:13.875 As an example, the Global Biodiversity Information Facility, 00:09:14.375 --> 00:09:18.916 some UAV classified imagery 00:09:19.125 --> 00:09:22.500 from Aaron Johnston in western South Dakota. 00:09:22.583 --> 00:09:26.458 We will take nearly any fractional component 00:09:26.458 --> 00:09:30.458 cover field observation as training data. 00:09:30.458 --> 00:09:33.833 So if you do have any data available that might fit the bill, 00:09:34.125 --> 00:09:35.958 please feel free to reach out altogether. 00:09:35.958 --> 00:09:40.375 We've got nearly 70 million training points to choose from, covering 00:09:40.666 --> 00:09:47.250 the broad temporal extent and capturing a lot of the spatial variability. 00:09:47.291 --> 00:09:51.958 We do note that there is higher density of training data in public lands, 00:09:51.958 --> 00:09:56.083 especially in BLM country, as opposed to like the Great Plains, 00:09:56.083 --> 00:09:59.083 as an example, where there's simply less public land. 00:09:59.125 --> 00:10:03.666 So in RCMAP, we produce two key series of data. 00:10:04.500 --> 00:10:08.708 One of them being the RCMAP set of annual products, 00:10:08.708 --> 00:10:12.000 those ten cover components I mentioned before, 00:10:12.750 --> 00:10:17.333 and the second being a series of RCMAP e.g. 00:10:17.625 --> 00:10:21.250 that is exotic annual grass near real time 00:10:21.375 --> 00:10:24.416 weekly and also annual products. 00:10:24.416 --> 00:10:29.416 So the annual products span the full Landsat archive 00:10:29.666 --> 00:10:34.125 going back to 1985, stretching through 2025. 00:10:34.291 --> 00:10:38.125 Currently, we have plans in place to extend that through 26 00:10:38.166 --> 00:10:42.666 and annual releases updating additional years at the end. 00:10:43.041 --> 00:10:44.000 Moving forward. 00:10:44.000 --> 00:10:48.750 Ten Cover Components I mentioned before the key input data 00:10:48.791 --> 00:10:54.375 here are Landsat ARD derived percentile composite. 00:10:54.750 --> 00:10:58.291 So we look across all the clear observations 00:10:58.791 --> 00:11:01.708 within each calendar year. 00:11:01.708 --> 00:11:04.708 Find the 10th, 50th, and 90th percentile 00:11:05.041 --> 00:11:09.625 of all the clear observations in that pixel and derived composites, 00:11:09.708 --> 00:11:14.416 the idea there being that those percentiles help us to capture 00:11:14.458 --> 00:11:18.708 the range of phenology conditions going on within that year. 00:11:18.791 --> 00:11:22.583 Also, pulling in spectral indices, topography features, 00:11:23.000 --> 00:11:25.791 and some texture metrics as well. 00:11:25.791 --> 00:11:31.208 Already talked about the training data for RCMAP moving over to the EAG side. 00:11:31.375 --> 00:11:32.416 The spring. 00:11:32.416 --> 00:11:37.666 We completed some major updates to that series of data. 00:11:37.708 --> 00:11:40.708 The previous PM of that project, Steve Boyd. 00:11:40.708 --> 00:11:45.083 We've been investigating updates to that release that this spring. 00:11:45.750 --> 00:11:48.958 So major updates to the methodology. 00:11:49.125 --> 00:11:52.541 Got a paper in the works describing those updates, 00:11:52.625 --> 00:11:55.625 and those are weekly products 00:11:55.750 --> 00:12:01.375 designed to be released within a week of the latest satellite 00:12:01.375 --> 00:12:06.625 observation going into the model produce for 2016 through 2026 currently. 00:12:06.625 --> 00:12:10.750 So we're in the midst of releasing 2026 data at the moment. 00:12:11.416 --> 00:12:14.416 And this data set we produce 00:12:14.625 --> 00:12:19.375 component cover similar as the RCMAP products fractional cover 00:12:19.708 --> 00:12:22.916 focusing on the herbs component, 00:12:23.125 --> 00:12:25.958 especially invasive grasses. 00:12:25.958 --> 00:12:28.958 So we have cover of total urban 00:12:28.958 --> 00:12:32.500 green and sinister cover, 00:12:32.833 --> 00:12:37.458 cheatgrass and medusa head, which are two of the key 00:12:37.500 --> 00:12:41.208 invasive grasses impacting a lot of the western United States. 00:12:41.875 --> 00:12:43.166 Field brome. 00:12:43.166 --> 00:12:47.083 Sandburg's, a native perennial grass that has similar phonology 00:12:47.166 --> 00:12:49.875 to many of the exotics, which is Sandburg's bluegrass, 00:12:49.875 --> 00:12:56.416 and the total cover of a list of 16 species of exotic annual grasses that are 00:12:57.416 --> 00:12:58.708 problematic. 00:12:58.708 --> 00:13:03.750 And with the data we're pulling in instead of Landsat data, 00:13:04.041 --> 00:13:09.041 we use harmonized Landsat Sentinel data, which gives us better 00:13:09.416 --> 00:13:13.750 temporal resolution, allowing us to better capture 00:13:14.208 --> 00:13:19.458 the logical patterns across the landscape, which is especially important 00:13:19.458 --> 00:13:24.625 for understanding the differences between a site with perennial grasses 00:13:24.625 --> 00:13:29.125 versus invasive grasses, where the phenology tends to be insights 00:13:29.125 --> 00:13:32.541 with a lot of invasive grasses otherwise similar 00:13:32.708 --> 00:13:35.708 inputs to the annual products. 00:13:36.000 --> 00:13:38.041 Similar training as well. 00:13:38.041 --> 00:13:39.208 I mentioned this before. 00:13:39.208 --> 00:13:45.083 Both of these products use neural network models for classification. 00:13:45.291 --> 00:13:49.958 Both of these data sets fit within a broader mapping hierarchy. 00:13:51.375 --> 00:13:52.541 A lot going on here. 00:13:52.541 --> 00:13:54.250 I'll give the basics. 00:13:54.250 --> 00:13:58.833 So in RCMAP we've got five primary components 00:13:58.875 --> 00:14:02.791 or basis shrub, tree, bare ground and litter. 00:14:03.125 --> 00:14:09.125 The sum of those five will equal 100% cover in each and every pixel 00:14:09.291 --> 00:14:14.250 that is enforced through post-processing as needed. 00:14:14.625 --> 00:14:19.625 We produce a series of secondary products for some of the primary components 00:14:20.083 --> 00:14:23.791 that add relevant floristic detail. 00:14:23.791 --> 00:14:27.791 For example, for sage grouse applications, understanding 00:14:28.083 --> 00:14:33.250 what portion of the shrub component is sagebrush is highly relevant. 00:14:34.500 --> 00:14:35.958 Those nest 00:14:35.958 --> 00:14:38.958 within their respective primary component. 00:14:39.208 --> 00:14:43.708 So on the total or basis side, we separate broadly into annuals and perennials, 00:14:44.500 --> 00:14:48.833 and these weekly and annual map EAG products 00:14:48.833 --> 00:14:52.708 all fit within the total or basis hierarchy. 00:14:52.875 --> 00:14:56.875 The overall scheme here is to capture 00:14:56.958 --> 00:15:02.375 the diversity of the landscape free of contradictions. 00:15:02.375 --> 00:15:05.375 And in this hierarchy, 00:15:06.583 --> 00:15:08.375 looking at our annual components, 00:15:08.375 --> 00:15:12.708 this is our most recent data release from the spring. 00:15:12.750 --> 00:15:15.750 Our version eight. 00:15:15.916 --> 00:15:18.541 These animations are showing how 00:15:18.541 --> 00:15:22.416 the component cover has changed from 1985 00:15:22.958 --> 00:15:26.458 until 2025 on an annual time step, 00:15:28.375 --> 00:15:30.583 showing that they are quite dynamic, 00:15:30.583 --> 00:15:34.291 especially components like bare ground and or basis. 00:15:35.375 --> 00:15:38.125 A lot to absorb at this level, 00:15:38.125 --> 00:15:41.541 which is why it's nice to condense them 00:15:42.041 --> 00:15:45.666 these temporal patterns into trends. 00:15:46.125 --> 00:15:49.125 And to me, time is the most interesting 00:15:49.166 --> 00:15:52.166 thing going on here with RCMAP. 00:15:52.250 --> 00:15:55.250 Understanding how the landscape has changed through time 00:15:55.375 --> 00:15:58.458 is really the take home message for all of this. 00:15:59.166 --> 00:16:02.416 So what I've done here is summarized 00:16:03.000 --> 00:16:07.041 the 1985 to 2025 trend per 00:16:07.041 --> 00:16:10.875 component per level three ecoregion, 00:16:12.083 --> 00:16:13.541 where ecoregions in 00:16:13.541 --> 00:16:17.333 blue have a net decreasing trend 00:16:17.583 --> 00:16:20.666 in that component through time, and the opposite 00:16:20.666 --> 00:16:23.666 for ecoregions in red. 00:16:24.375 --> 00:16:28.458 Some of the big stories here are related to 00:16:28.458 --> 00:16:32.375 annual basis, which would include invasive grasses. 00:16:32.916 --> 00:16:35.500 That problem is expanding 00:16:35.500 --> 00:16:38.500 across much of the West and the Great Basin. 00:16:38.958 --> 00:16:42.833 Some new areas of invasion of annual grasses 00:16:44.166 --> 00:16:47.166 into Montana and Wyoming as well. 00:16:47.500 --> 00:16:50.500 And that increase in annual 00:16:50.500 --> 00:16:54.791 invasive grasses is driving an increase in the fire regime, 00:16:55.666 --> 00:17:00.041 leading to more burned area, which results in a declining 00:17:00.083 --> 00:17:04.250 trend in sagebrush since 85 across 00:17:04.250 --> 00:17:07.250 most of the West. 00:17:07.375 --> 00:17:10.250 And also on the flip side, 00:17:10.250 --> 00:17:14.333 an increasing trend in herb cover overall 00:17:15.791 --> 00:17:19.708 as the grasses expand in burned areas. 00:17:19.750 --> 00:17:22.750 A lot of that increase in is related 00:17:22.958 --> 00:17:26.458 to increases in the annual portion of that. 00:17:26.458 --> 00:17:28.625 So not so great news. 00:17:28.625 --> 00:17:31.041 And another side effect of all of this 00:17:31.041 --> 00:17:34.166 is a declining pattern in bare ground. 00:17:34.500 --> 00:17:38.458 So the sagebrush biome in particular typically 00:17:38.458 --> 00:17:44.333 had a lot of bare ground between the inner spaces of shrubs. 00:17:44.375 --> 00:17:47.375 After invasion and following fire. 00:17:47.375 --> 00:17:51.500 Those spaces get filled in with invasive grasses, often 00:17:51.958 --> 00:17:55.375 leading to a declining pattern in bare ground. 00:17:56.375 --> 00:17:59.375 Another interesting pattern to note here 00:17:59.791 --> 00:18:02.791 in the Great Plains, this is really well documented. 00:18:03.666 --> 00:18:06.875 There's an increasing pattern in tree cover, 00:18:07.958 --> 00:18:12.625 especially in the southern portion of the Great Plains, as trees, 00:18:12.666 --> 00:18:17.708 especially eastern red cedar, are expanding due to fire suppression 00:18:17.708 --> 00:18:20.708 and other management impacts. 00:18:21.541 --> 00:18:24.541 How is Arcee map data been used in the past? 00:18:24.666 --> 00:18:27.083 These are examples from the literature. 00:18:27.083 --> 00:18:32.541 Lots of diverse applications of the data in wildlife management, 00:18:32.708 --> 00:18:37.291 climate analysis, work on energy development, all sorts of things. 00:18:37.291 --> 00:18:40.291 I think my favorite is probably a study on lampreys 00:18:40.291 --> 00:18:43.500 in the Goose Lake Basin in Oregon. 00:18:43.666 --> 00:18:47.875 There's a broad array of applications for these sort of data. 00:18:48.333 --> 00:18:50.875 I'm going to focus on three today. 00:18:50.875 --> 00:18:55.375 So mentioned the tree cover expansion in the Great Plains. 00:18:55.458 --> 00:18:58.000 This is an example of exactly that pattern. 00:18:58.000 --> 00:19:01.750 So in southwest Nebraska here 00:19:03.375 --> 00:19:06.375 I've got the tree cover slope 00:19:06.541 --> 00:19:09.541 between 85 and present 00:19:10.666 --> 00:19:11.250 here. 00:19:11.250 --> 00:19:15.000 So increases in red and decreases in blue 00:19:15.000 --> 00:19:18.291 and or basis cover same pattern 00:19:18.875 --> 00:19:21.875 blue showing decreased red showing increase. 00:19:22.791 --> 00:19:24.791 So in this area 00:19:24.791 --> 00:19:28.458 tree cover has been expanding since 85. 00:19:28.833 --> 00:19:32.125 We can see that's the general pattern here in red. 00:19:32.291 --> 00:19:37.125 And as one component goes up something else has to go down. 00:19:37.458 --> 00:19:40.583 As I mentioned before everything it adds up to 100. 00:19:40.666 --> 00:19:42.625 So it's a zero some game. 00:19:42.625 --> 00:19:47.166 And in this case what's going down is herb cover grass cover. 00:19:47.458 --> 00:19:52.125 The exception to that being this area that burned in 2002, 00:19:52.416 --> 00:19:57.541 we can clearly see that impact on the tree cover slope, where that is 00:19:57.541 --> 00:20:02.416 an exception to the general rule in this area of increase, 00:20:03.125 --> 00:20:07.291 and similarly an increasing trend in our basis. 00:20:07.291 --> 00:20:11.166 They're looking at a wildlife management application 00:20:11.208 --> 00:20:17.125 in the northern portion of Montana stretching into southern Canada. 00:20:17.416 --> 00:20:20.416 We did expand our study area a couple years ago 00:20:20.625 --> 00:20:26.500 to capture the sagebrush biome extent into Alberta and Saskatchewan as well. 00:20:26.541 --> 00:20:29.291 So Julie Heinrichs et al. 00:20:29.291 --> 00:20:32.833 In this case, used our data to understand 00:20:32.833 --> 00:20:36.875 long term changes in sagebrush connectivity. 00:20:37.000 --> 00:20:42.000 Put those data into a sage grouse resource use analysis model. 00:20:42.000 --> 00:20:47.541 Considering that as as one of several factors in that model, ultimately leading 00:20:48.000 --> 00:20:50.833 to an understanding of sage grouse 00:20:50.833 --> 00:20:53.833 mortality risk in this landscape. 00:20:53.916 --> 00:20:59.000 Another wildlife focused case study here, the U.S. 00:20:59.000 --> 00:21:04.375 Fish and Wildlife Service leads an effort called the Sagebrush Conservation Design. 00:21:04.625 --> 00:21:07.916 This effort aims to 00:21:09.041 --> 00:21:11.625 categorize the sagebrush biome 00:21:11.625 --> 00:21:14.916 into three priority classes core 00:21:15.666 --> 00:21:19.750 sort of the best of the best remaining sagebrush habitat, 00:21:19.875 --> 00:21:23.583 where the sagebrush cover is intact, 00:21:23.833 --> 00:21:26.833 not very badly fragmented, 00:21:27.041 --> 00:21:30.041 and little cover of invasive grasses. 00:21:30.750 --> 00:21:35.375 Growth Opportunity area is sort of a moderate degree of intactness. 00:21:35.375 --> 00:21:39.166 Maybe some invasives presence in that landscape 00:21:39.416 --> 00:21:43.458 and other rangeland areas where the sagebrush habitat 00:21:43.500 --> 00:21:47.166 has been badly degraded by invasives, 00:21:47.166 --> 00:21:50.333 fire, fragmentation, etc.. 00:21:50.375 --> 00:21:55.666 So the key map data both the sagebrush cover or basics 00:21:55.833 --> 00:22:00.500 cover a bare ground cover all feed into this classification, 00:22:00.541 --> 00:22:06.375 and this helps guide landscape scale management actions. 00:22:06.875 --> 00:22:09.583 Recognizing that we have limited 00:22:09.583 --> 00:22:14.708 conservation dollars available, we want to apply those dollars 00:22:14.708 --> 00:22:17.833 to the places that are most likely 00:22:17.833 --> 00:22:21.291 to see the greatest net benefit from them. 00:22:22.875 --> 00:22:25.875 Another important thing to note here is 00:22:26.791 --> 00:22:30.291 in the most recent version of the sagebrush conservation design, 00:22:30.291 --> 00:22:34.541 which is currently in development, the authors have decided 00:22:34.541 --> 00:22:39.250 to pull in data from the rangeland analysis platform as well. 00:22:39.333 --> 00:22:45.041 The wrap products are similar in many respects to what RCMAP does, 00:22:45.083 --> 00:22:50.916 and in this case, the authors of the Sagebrush Conservation Design 00:22:51.083 --> 00:22:54.916 found a benefit from doing 00:22:54.916 --> 00:22:59.041 a consensus of both our products and the wrap products. 00:22:59.041 --> 00:23:02.416 So that's one thing I really want to highlight here, is it's 00:23:02.416 --> 00:23:06.625 not a choice of either or necessarily which data set you use. 00:23:06.625 --> 00:23:10.916 There is a benefit of, in this case, a geometric 00:23:10.958 --> 00:23:14.875 average of both the Wrap and RCMAP products. 00:23:14.875 --> 00:23:17.500 I found the same thing in this before. 00:23:17.500 --> 00:23:22.416 Definitely less air in a consensus product than a single product by itself. 00:23:22.416 --> 00:23:23.625 Speaking of error, 00:23:24.875 --> 00:23:26.166 that's one thing 00:23:26.166 --> 00:23:30.791 we focus a lot on in RCMAP is accuracy assessment. 00:23:30.958 --> 00:23:34.208 So I mentioned the BLM Aim data previously. 00:23:34.791 --> 00:23:39.375 We withhold a 10% set 00:23:39.416 --> 00:23:43.833 of all of that Aim data for the purposes of validation, 00:23:44.041 --> 00:23:48.541 which is about 6600 points in the current iteration of RCMAP, 00:23:49.375 --> 00:23:52.166 the rangeland analysis platform folks withhold 00:23:52.166 --> 00:23:56.416 that same set of data for cross product comparisons. 00:23:56.625 --> 00:24:00.666 Any other project who wants to participate could do so as well. 00:24:00.750 --> 00:24:03.833 So in this graphic I am showing 00:24:03.958 --> 00:24:07.291 the accuracy stats the arms 00:24:07.375 --> 00:24:10.833 which is the root mean squared error on the left and the correlation, 00:24:10.916 --> 00:24:15.250 the R value on the right of our component 00:24:15.250 --> 00:24:18.916 cover relative to those withheld points. 00:24:19.125 --> 00:24:24.500 So the model did not see these points whatsoever in development or version eight 00:24:24.791 --> 00:24:29.291 is the most recently released product in May this year. 00:24:30.208 --> 00:24:34.791 Version seven was the prior year, six before that, and so on. So 00:24:35.750 --> 00:24:36.208 looking 00:24:36.208 --> 00:24:41.500 at our relative degree of enhancements through time over the last five years, 00:24:42.333 --> 00:24:45.333 so we can see in both our M.Sc. 00:24:45.666 --> 00:24:48.083 Where lower values would be better 00:24:48.083 --> 00:24:52.375 and in correlation were higher values would be better. 00:24:52.833 --> 00:24:55.375 We are definitely trending the right direction. 00:24:55.375 --> 00:24:59.791 There are some components that are just easier to produce than others. 00:24:59.916 --> 00:25:00.500 Their ground. 00:25:00.500 --> 00:25:05.958 As an example, trees an example tend to be more accurate overall, 00:25:06.166 --> 00:25:10.375 just a more clear spectral signal with those components 00:25:10.375 --> 00:25:14.750 as compared to something like shrub or litter as an example. 00:25:14.833 --> 00:25:21.000 So we spent a lot of time improving our products between each iteration, 00:25:21.375 --> 00:25:25.333 pulling in new data sets, new modeling techniques, 00:25:25.583 --> 00:25:28.375 new training data, etc. 00:25:28.375 --> 00:25:32.500 each new iteration of RCMAP is a completely new time series. 00:25:32.500 --> 00:25:36.250 So for example, when we do the V8 predictions, 00:25:36.375 --> 00:25:39.583 we would redo not just 2025. 00:25:39.833 --> 00:25:42.833 We would go back all the way to 1985 because 00:25:43.375 --> 00:25:47.708 new training data in 2025 from a new site 00:25:48.000 --> 00:25:51.833 would have impacts going all the way back through the time series. 00:25:51.916 --> 00:25:56.333 Probably the most common question I get from users is 00:25:56.375 --> 00:26:01.541 what is the appropriate spatial scale to apply your product? 00:26:01.625 --> 00:26:05.541 And we've always known that bigger is better. 00:26:06.166 --> 00:26:08.083 What do I mean by that? 00:26:08.083 --> 00:26:12.958 Is the mean of a 1000 acre pasture likely 00:26:12.958 --> 00:26:18.125 to be closer to truth than a single random pixel within that pasture? 00:26:18.166 --> 00:26:20.666 Yes, but that's not very clear guidance. 00:26:20.666 --> 00:26:25.250 So we undertook an analysis to get a better quantitative 00:26:25.250 --> 00:26:29.250 idea of what the appropriate scale would be. 00:26:29.583 --> 00:26:32.583 We did this by scaling our 00:26:32.833 --> 00:26:35.458 high resolution training sites 00:26:35.458 --> 00:26:38.416 to a variety of resolutions starting at 30m, 00:26:38.416 --> 00:26:42.625 which is the native resolution of our products all the way up to 1500 00:26:42.625 --> 00:26:47.791 meter resolution, and compared the accuracy of our products 00:26:47.791 --> 00:26:52.708 relative to that training data, which is approximately truth. 00:26:53.291 --> 00:26:58.166 And not surprisingly, we found that aggregating 00:26:58.208 --> 00:27:02.208 the map predictions to a 90 meter resolution 00:27:02.250 --> 00:27:07.708 improves the correlation and RMSE relative to the native 30 meter resolution. 00:27:07.708 --> 00:27:13.916 And that affects sort of plateaus as you go beyond about 210 meter resolution. 00:27:13.916 --> 00:27:18.375 Just to show that if you can get by with data 00:27:18.416 --> 00:27:23.875 at a coarser resolution than 30m, you can produce a more accurate product. 00:27:23.875 --> 00:27:26.875 All of this varies by landscape complexity, 00:27:26.916 --> 00:27:32.375 how heterogeneous landscape is, what component you're looking at, etc. 00:27:32.416 --> 00:27:34.833 but even at a 90 meter scale, 00:27:34.833 --> 00:27:38.708 the accuracy is quite a bit better than a 30 meter scale. 00:27:38.708 --> 00:27:39.958 And I should also note that 00:27:39.958 --> 00:27:44.458 that Aim validation I showed before is at a 30 meter scale. 00:27:44.458 --> 00:27:48.041 So we are comparing single pixel truth 00:27:48.125 --> 00:27:51.208 tag to our modeled predictions. 00:27:51.250 --> 00:27:54.666 Some other important things to consider in terms of accuracy. 00:27:54.708 --> 00:27:59.541 First, more common values are predicted more accurately. 00:27:59.541 --> 00:28:04.750 If we look at the frequency of a cover value 00:28:04.833 --> 00:28:07.958 per component combination versus 00:28:08.000 --> 00:28:11.708 the RMS of that cover combination. 00:28:11.708 --> 00:28:15.375 In our model predictions, we see a clear pattern here 00:28:15.375 --> 00:28:19.291 that the more frequent that cover been is, 00:28:19.708 --> 00:28:23.208 the less error there is in that prediction. 00:28:23.208 --> 00:28:26.041 So something really common, like a 00:28:27.458 --> 00:28:30.375 10% stage brush cover 00:28:30.375 --> 00:28:35.750 value, will all else equal be more accurate than something 00:28:35.750 --> 00:28:40.416 less common, like a 99% cover sagebrush bin? 00:28:40.458 --> 00:28:42.458 That would be very, very uncommon. 00:28:42.458 --> 00:28:45.583 Another important thing to think about is the distinction 00:28:45.583 --> 00:28:48.916 between absolute error and relative error. 00:28:49.000 --> 00:28:51.375 Those can be at odds at times. 00:28:51.375 --> 00:28:54.375 For example, looking at bare ground, the peak 00:28:54.416 --> 00:28:58.375 RMSE is at about 80% cover, but the peak 00:28:58.375 --> 00:29:02.333 relative error is at about to 10% cover. 00:29:02.333 --> 00:29:05.875 So considering the ratio of error 00:29:06.375 --> 00:29:09.916 of absolute error to the actual predicted value, 00:29:10.041 --> 00:29:14.041 aka relative error can be important as well. 00:29:14.208 --> 00:29:18.416 Also, should note that our models, like most models, 00:29:18.708 --> 00:29:21.458 suffer from regression to the mean. 00:29:21.458 --> 00:29:22.958 What does that look like? 00:29:22.958 --> 00:29:27.375 It looks like inflation of predicted values at the low end, 00:29:27.541 --> 00:29:30.750 and a deflation of values at the high end. 00:29:30.833 --> 00:29:33.833 So the range of values can be truncated 00:29:33.833 --> 00:29:37.166 relative to training data as an example. 00:29:37.208 --> 00:29:41.291 So we have enough withheld Aim data at this point 00:29:41.333 --> 00:29:45.166 to look at spatial patterns in accuracy. 00:29:45.750 --> 00:29:51.083 In this case I'm looking at independent correlation per level three ecoregion. 00:29:51.125 --> 00:29:57.333 Some patterns to note here are the more common that component 00:29:57.333 --> 00:30:02.333 is per ecoregion, the more accurate that prediction tends to be. 00:30:02.541 --> 00:30:06.416 So as an example, sagebrush cover is very common 00:30:06.416 --> 00:30:11.541 in the central Great Plains, much less common in the Mojave Desert, 00:30:12.083 --> 00:30:14.875 and our accuracy of sagebrush is higher 00:30:14.875 --> 00:30:18.291 in the central Great Basin then in the Mojave Desert. 00:30:18.666 --> 00:30:24.208 That's the general pattern that dominates the spatial patterns we see here. 00:30:24.250 --> 00:30:30.125 Other things like the background, mineralogy and geology of the site 00:30:30.166 --> 00:30:34.791 do complicate our predictions as well, especially in places with lots of bare 00:30:34.833 --> 00:30:39.416 ground like portions of the Mojave Desert, Great Basin, etc. 00:30:39.458 --> 00:30:44.166 in a lot of cases, the primary thing you see with Landsat imagery 00:30:44.166 --> 00:30:50.291 is that background mineralogy, which can, like I said, complicate predictions. 00:30:50.458 --> 00:30:50.833 Okay. 00:30:50.833 --> 00:30:55.875 Moving over to the other major data set that we produce 00:30:56.291 --> 00:30:59.291 are exotic annual grass product 00:30:59.666 --> 00:31:03.125 already talked about the invasive grass 00:31:03.541 --> 00:31:06.458 acceleration of the fire cycle, 00:31:06.458 --> 00:31:09.500 being a key driver of habitat loss 00:31:10.041 --> 00:31:14.416 and degradation across a lot of the western portion of the US. 00:31:15.708 --> 00:31:18.708 So again, we're using harmonized Landsat Sentinel, 00:31:18.833 --> 00:31:23.041 the BLM Aim data, our high resolution training sites 00:31:23.125 --> 00:31:27.750 to model weekly cover of basis, 00:31:28.375 --> 00:31:33.375 invasive grasses and some species specific maps like cheatgrass 00:31:33.375 --> 00:31:38.541 and medusa head near real time classification of cover conditions 00:31:38.625 --> 00:31:44.250 helps in rapid decision making related to grazing and fire. 00:31:44.250 --> 00:31:46.375 So that's one of our key focuses here. 00:31:46.375 --> 00:31:50.708 Again, in releasing these data as quickly as possible 00:31:51.125 --> 00:31:55.416 for near real time application mentioned this already. 00:31:55.458 --> 00:32:00.458 We did a major overhaul of our mapping methodology in the most recent 00:32:00.458 --> 00:32:01.333 data release. 00:32:01.333 --> 00:32:06.375 The key problem in the prior version of the data was associating 00:32:06.875 --> 00:32:09.625 the yearly Aim observations 00:32:09.625 --> 00:32:14.208 with a weekly ndVi curve, resulting in a one 00:32:14.208 --> 00:32:18.750 to many relationship instead of a 1 to 1 relationship as we do now. 00:32:19.375 --> 00:32:22.375 Therefore, the old product represented 00:32:22.458 --> 00:32:27.333 a forecast of peak cover, a forecast that changed 00:32:27.750 --> 00:32:30.500 as we got better and better information 00:32:30.500 --> 00:32:33.500 towards the peak of each year. 00:32:33.541 --> 00:32:38.000 In addition to how we've trained the model, we've expanded 00:32:38.000 --> 00:32:41.333 the study area for these products, expanded the mapping window. 00:32:41.375 --> 00:32:44.916 Previously, the maps were released March through June. 00:32:45.000 --> 00:32:47.833 Now we're going March through September. 00:32:47.833 --> 00:32:51.666 We've added additional components, made the data more consistent, 00:32:51.750 --> 00:32:54.083 improved input variables to the model. 00:32:54.083 --> 00:32:58.416 We are now participating in that 10% withholding aim validation 00:32:58.416 --> 00:33:03.375 I talked about before, and offered the data in a lot of new places 00:33:03.375 --> 00:33:08.500 in terms of user access like Google Earth Engine, MRC, etc. 00:33:09.416 --> 00:33:11.750 all of these changes have 00:33:11.750 --> 00:33:15.250 dramatically improved our validation results. 00:33:15.333 --> 00:33:21.166 So the top panel here is showing the accuracy of our e g products 00:33:21.166 --> 00:33:25.375 relative to that same set of 10% aim withheld points 00:33:25.375 --> 00:33:29.958 I talked about before, compared to the accuracy of our prior set of data 00:33:30.958 --> 00:33:32.166 total basis, 00:33:32.166 --> 00:33:35.666 we did not have that product before, so it can't make that comparison. 00:33:36.083 --> 00:33:41.291 But for all the other components, we are significantly better than previously. 00:33:41.666 --> 00:33:45.958 Also, the prior version of the data, in this case, it's 00:33:45.958 --> 00:33:49.375 not a fully fair comparison because we did 00:33:49.375 --> 00:33:53.250 not withhold these points from training the old version. 00:33:53.250 --> 00:33:57.375 So it's a cross-validation on the bottom compared to independent validation. 00:33:57.375 --> 00:33:58.875 And yet we're still doing better. 00:33:58.875 --> 00:34:03.041 So really good news there in terms of improved product accuracy 00:34:03.125 --> 00:34:04.958 based on how the model is trained. 00:34:04.958 --> 00:34:09.750 Our model will initially predict green cover of all of our components, 00:34:09.750 --> 00:34:14.666 and we derive a total and thinnest portion of 00:34:15.666 --> 00:34:19.375 cover through a post-processing approach that calculates 00:34:19.625 --> 00:34:23.791 maximum year to date green cover as total, 00:34:23.791 --> 00:34:30.041 and then cover would be the difference between weekly total and green cover. 00:34:30.083 --> 00:34:33.083 Here's an example of what that phenology looks like. 00:34:33.750 --> 00:34:36.416 We're total and green are synonymous. 00:34:36.416 --> 00:34:41.791 Until we see a decrease in green cover total would stay flat. 00:34:41.791 --> 00:34:44.750 In that scenario, senescent would increase. 00:34:44.750 --> 00:34:48.500 And then eventually, once vegetation is growth 00:34:48.500 --> 00:34:51.875 is done that year, all the cover would go to senescence. 00:34:52.958 --> 00:34:54.750 I think this crossing point here, 00:34:54.750 --> 00:34:59.375 where green and nest meat and eventually nest is higher 00:34:59.375 --> 00:35:04.000 than green, would have important fire implications. 00:35:06.708 --> 00:35:08.958 So looking at the weekly time 00:35:08.958 --> 00:35:11.958 series here from 2025 00:35:12.000 --> 00:35:15.666 for cheatgrass on the left and total e.g. 00:35:15.833 --> 00:35:18.875 on the right, we can see a rapid increase 00:35:18.875 --> 00:35:22.375 early in the spring and then a plot tailing effect. 00:35:22.750 --> 00:35:27.125 So we're looking here at total cover, which would not decrease 00:35:28.083 --> 00:35:31.083 throughout the course of the year. 00:35:32.500 --> 00:35:32.875 Moving 00:35:32.875 --> 00:35:35.875 to a different set here again from 25, 00:35:36.125 --> 00:35:39.125 we've got green total and senescent. 00:35:39.666 --> 00:35:42.666 Green is going to increase 00:35:43.083 --> 00:35:46.083 starting in the south and at low elevation. 00:35:46.291 --> 00:35:49.250 And the green will work its way further north 00:35:49.250 --> 00:35:52.625 and to higher elevation as the season progresses. 00:35:53.541 --> 00:35:58.500 Total or will accumulate and not decline. 00:35:59.041 --> 00:36:03.500 And the difference between total and green will go to senescent. 00:36:04.375 --> 00:36:06.541 The brown or dead 00:36:06.541 --> 00:36:10.125 plant material that was previously alive that year. 00:36:11.625 --> 00:36:13.833 Looking at the 00:36:13.833 --> 00:36:18.208 patterns across some select ecoregions of our study area. 00:36:18.250 --> 00:36:20.291 This is the green component. 00:36:20.291 --> 00:36:23.791 So again we would expect an increase and then decrease. 00:36:24.666 --> 00:36:28.041 So these patterns follow the expected 00:36:28.041 --> 00:36:33.500 phenology here in terms of when the peak green cover is 00:36:33.541 --> 00:36:37.791 and the timing of the start and end of season. 00:36:38.000 --> 00:36:42.416 Interesting also to compare across years. 00:36:42.833 --> 00:36:45.416 So in this case I've plotted the study 00:36:45.416 --> 00:36:48.583 area wide means by year 00:36:48.791 --> 00:36:52.375 for six of our EEG components showing. 00:36:52.541 --> 00:36:54.916 And then the lines are color coded by year. 00:36:54.916 --> 00:36:57.916 So focusing on cheatgrass as an example here 00:36:59.208 --> 00:37:01.416 we can see there's a lot of variability 00:37:01.416 --> 00:37:05.791 between years in terms of when green up occurs. 00:37:06.041 --> 00:37:12.041 And the relative amount of cheatgrass is pretty variable, especially compared to 00:37:12.041 --> 00:37:17.125 something like total or basis, where the the degree of variability is much lower. 00:37:17.916 --> 00:37:22.208 All of this underscores what we see in the literature that cheatgrass 00:37:22.500 --> 00:37:27.125 and invasive grass cover in general, tends to be much more dynamic temporarily 00:37:27.375 --> 00:37:30.375 than perennials are that e.g., data 00:37:30.500 --> 00:37:33.625 reflect fire nearly immediately. 00:37:33.875 --> 00:37:35.250 So this is an example. 00:37:35.250 --> 00:37:38.250 In California, the cornucopia fire. 00:37:38.583 --> 00:37:40.000 And that occurred 00:37:41.083 --> 00:37:42.750 about week 28. 00:37:42.750 --> 00:37:45.750 Here we can see as soon as that fire occurred 00:37:45.750 --> 00:37:50.125 we well, actually even prior to that fire occurring, green was going down 00:37:50.375 --> 00:37:53.625 making that eventual fire more likely. 00:37:54.083 --> 00:37:57.291 And then after fire, we lose all our green cover. 00:37:57.541 --> 00:37:59.583 And that goes to senescence. 00:37:59.583 --> 00:38:05.250 That's one caveat in these products, really, the fire would consume that 00:38:05.708 --> 00:38:09.375 to cover instead of it going up, it should go down as well. 00:38:09.375 --> 00:38:12.375 That's something we're working on in future generations 00:38:12.458 --> 00:38:15.875 in terms of improving that post-fire representation. 00:38:15.916 --> 00:38:17.208 Change accuracy. 00:38:17.208 --> 00:38:22.916 So all the accuracy statistics I've shown so far are comparing the accuracy 00:38:22.916 --> 00:38:26.666 of our map, a single date of our map 00:38:26.666 --> 00:38:30.583 to a single date of validation data. 00:38:30.625 --> 00:38:35.750 The BLM aim database does contain lots of examples of plots 00:38:35.750 --> 00:38:41.916 that have been observed multiple times, both across years and even within a year. 00:38:41.916 --> 00:38:45.791 So here's an example of an aim point in Nevada 00:38:45.833 --> 00:38:50.000 that's been observed six times starting in 2013. 00:38:50.666 --> 00:38:53.416 We've got plot photos here showing the conditions 00:38:53.416 --> 00:38:56.416 of what that plot looked like through time. 00:38:56.500 --> 00:38:59.958 And I've got the Aim data plotted in these orange dots, 00:39:00.125 --> 00:39:04.083 the RCMAP components at that same location 00:39:05.041 --> 00:39:06.833 with the blue line. 00:39:06.833 --> 00:39:11.250 So we can see in both annular basis and in bare ground. 00:39:11.875 --> 00:39:15.666 The sea map components do a nice job of capturing 00:39:15.750 --> 00:39:19.791 that temporal pattern observed in AME data. 00:39:20.375 --> 00:39:24.958 This is an example of within year change from AME data. 00:39:25.416 --> 00:39:28.208 This is in the Mojave Desert in Nevada. 00:39:28.208 --> 00:39:31.208 Looking at the year 2020, 00:39:31.375 --> 00:39:35.375 this site has been observed five times within 2022, 00:39:35.416 --> 00:39:39.541 rather in June and September of 22. 00:39:39.750 --> 00:39:43.875 Quite a bit of difference even in that shorter time period. 00:39:45.208 --> 00:39:49.166 A little hard to see, but there's a patch of cover here 00:39:49.166 --> 00:39:53.125 that really is less dense in June, so they must 00:39:53.125 --> 00:39:58.583 have had a pretty good monsoon season this year to get more cover. 00:39:59.000 --> 00:40:01.416 So in this case we can see the aim. 00:40:01.416 --> 00:40:06.541 Observations of urban cover have increased from here 00:40:06.541 --> 00:40:10.625 to here, and our predictions have done the same. 00:40:10.625 --> 00:40:14.750 They have increased nearly synonymously with AME data. 00:40:14.875 --> 00:40:17.958 In this example, if we plot these out through time, 00:40:18.000 --> 00:40:23.833 all the course bonding AME observations versus the course week of our 00:40:24.875 --> 00:40:25.833 cover, we 00:40:25.833 --> 00:40:28.833 see a pretty nice correlation there. 00:40:29.041 --> 00:40:31.208 Moving over to Oregon, 00:40:31.208 --> 00:40:33.750 focusing on phenology here. 00:40:33.750 --> 00:40:38.333 This is an Aim site that was observed in May and June of 2025. 00:40:38.958 --> 00:40:43.125 You can see the phenology is much more progressed on the right. 00:40:43.583 --> 00:40:48.375 We've lost most of that green cover in this month span, 00:40:48.916 --> 00:40:51.500 and our data correspond 00:40:51.500 --> 00:40:54.791 pretty well with that in terms of less green cover 00:40:54.916 --> 00:40:57.916 in June as compared to May. 00:40:58.583 --> 00:41:00.875 One thing to note here is that 00:41:00.875 --> 00:41:04.166 satellite products tend to lag reality 00:41:04.166 --> 00:41:10.125 by about two weeks, so probably more fair to move this green curve a little bit 00:41:10.125 --> 00:41:13.916 to the left in this comparison, which would better align the products. 00:41:13.916 --> 00:41:18.541 And if we plot all of these out, there's like 8000 examples of interannual 00:41:18.583 --> 00:41:22.958 change in AME data and about a couple 00:41:23.000 --> 00:41:25.958 thousand examples within your change. 00:41:25.958 --> 00:41:29.791 So focusing on the cover component from RCMAP, 00:41:30.166 --> 00:41:34.375 we can see that aim change on the 00:41:34.375 --> 00:41:37.875 x axis versus map change on the y axis. 00:41:38.541 --> 00:41:41.666 We do a pretty good job of capturing 00:41:41.666 --> 00:41:45.791 that Interannual change in the mapping product 00:41:46.833 --> 00:41:49.833 also should highlight, as I mentioned before, 00:41:49.916 --> 00:41:54.541 the map change slope is less than ame, 00:41:55.041 --> 00:41:59.041 meaning that the map change tends to be less 00:41:59.041 --> 00:42:02.041 than the real world observed change. 00:42:02.916 --> 00:42:06.541 A pretty nice correlation with the interannual 00:42:06.541 --> 00:42:10.583 within season change as well with e.g. 00:42:10.666 --> 00:42:11.375 cover. 00:42:11.375 --> 00:42:17.458 We've done some work with the weekly EEG products to understand 00:42:17.666 --> 00:42:22.916 how the landscape condition currently differs from the long term mean. 00:42:22.916 --> 00:42:24.666 So what we've done here. 00:42:24.666 --> 00:42:27.208 This is an example from 2026. 00:42:27.208 --> 00:42:29.458 Week 23. 00:42:29.458 --> 00:42:34.333 We took the mean of all prior week 23 in our products. 00:42:34.333 --> 00:42:40.250 So the mean of the basis cover from 2016 through 2025. 00:42:40.250 --> 00:42:46.041 And then compare that from our 2026 week 00:42:46.083 --> 00:42:49.708 23 map and mapped out the spatial patterns here. 00:42:49.708 --> 00:42:53.083 So blue meaning we in week 23 of 00:42:53.083 --> 00:42:58.083 26 had more cover than the long term mean. 00:42:58.833 --> 00:43:03.291 And red being we have less verbatim cover than the long term mean. 00:43:03.958 --> 00:43:07.166 So this would be about June of this year. 00:43:07.875 --> 00:43:10.916 I did some field work in this part of the world around Gillette, 00:43:11.000 --> 00:43:15.833 Wyoming about that time, and indeed it checks out in the field. 00:43:15.833 --> 00:43:20.791 It was very, very dry in this part of the world at that time. 00:43:22.416 --> 00:43:23.875 Another interesting thing to 00:43:23.875 --> 00:43:28.958 point out here is all these blue dots in Oregon, Washington, etc. 00:43:29.000 --> 00:43:31.666 are burned areas. 00:43:31.666 --> 00:43:35.333 So places where there was tree cover at some point 00:43:35.333 --> 00:43:38.833 in that reference period between 16 and 25. 00:43:39.041 --> 00:43:41.208 And now we're a post burn. 00:43:41.208 --> 00:43:44.208 So or cover has increased 00:43:44.375 --> 00:43:47.208 as tree cover was reduced with fire 00:43:47.208 --> 00:43:51.458 leading to that positive deviation in herbaceous cover. 00:43:52.000 --> 00:43:55.125 So on the left we can see what that looks like temporarily. 00:43:56.083 --> 00:43:59.875 This is from 2026 focusing on total or basis. 00:44:00.291 --> 00:44:03.625 I've got the long term mean per week plotted in blue 00:44:03.666 --> 00:44:07.166 with standard deviation in that shade. 00:44:07.833 --> 00:44:12.791 We started the year above normal, probably as a result of a warm 00:44:12.791 --> 00:44:16.375 winter across a lot of the West, and as the season progressed, 00:44:16.916 --> 00:44:21.625 we eventually went below the long term mean by June 00:44:21.666 --> 00:44:25.625 our data both the RCMAP and RCMAP EAG. 00:44:25.750 --> 00:44:30.583 The weekly products are available on the MRC Rangeland Viewer, 00:44:30.625 --> 00:44:35.666 where you can download custom extents of the products, do time series 00:44:35.666 --> 00:44:40.208 plotting by a series of different polygon types, so plot long 00:44:40.250 --> 00:44:44.125 term trends by watershed by a lot meant, etc. 00:44:44.250 --> 00:44:47.375 the data are also available for static download on MRL. 00:44:47.375 --> 00:44:50.666 See we've got w w CS services 00:44:50.750 --> 00:44:54.041 on MRC, also on ArcGIS online. 00:44:54.125 --> 00:44:57.041 The data are also in the Google Earth 00:44:57.041 --> 00:45:00.833 Community Catalog, specifically the awesome Community catalog 00:45:00.875 --> 00:45:06.166 there on the Desert Research Institute Climate engine there on Science Base. 00:45:06.166 --> 00:45:10.083 We try to make these as broadly accessible as possible 00:45:10.125 --> 00:45:11.875 and lots of different venues. 00:45:11.875 --> 00:45:17.125 So we are in the midst of releasing 2026 data for the E.g. 00:45:17.333 --> 00:45:19.083 side of the project. 00:45:19.083 --> 00:45:20.375 We'll continue that through. 00:45:20.375 --> 00:45:23.208 Next week is actually our last release in the background. 00:45:23.208 --> 00:45:26.875 We've been releasing the 2016 through 2025, E.g. 00:45:26.916 --> 00:45:27.375 data. 00:45:27.375 --> 00:45:30.750 We've got a paper summarizing the updated e.g. 00:45:30.833 --> 00:45:36.291 methods in works, another paper summarizing the entire RCMAP project, 00:45:36.333 --> 00:45:39.375 and a third one looking at regional accuracy 00:45:39.375 --> 00:45:44.125 patterns of RCMAP will produce a new generation of e.g. 00:45:44.166 --> 00:45:46.291 products starting early next year. 00:45:46.291 --> 00:45:50.708 Same with a new generation of RCMAP early next year. 00:45:50.750 --> 00:45:53.958 We are looking into expanding the number 00:45:53.958 --> 00:45:57.000 of unique species we're mapping on the E.g. 00:45:57.041 --> 00:46:01.916 side, with some more emphasis on some of the species in the southwest 00:46:02.000 --> 00:46:05.750 like red brome, buffel grass, etc. 00:46:05.833 --> 00:46:08.916 we're also exploring additional modes 00:46:08.916 --> 00:46:14.041 of data input, so specifically hyperspectral data, 00:46:14.458 --> 00:46:18.333 SAR data, etc., further improving our products. 00:46:18.333 --> 00:46:21.875 And then lastly, this is still in its infancy, 00:46:22.000 --> 00:46:25.833 but we are looking into better incorporation 00:46:25.833 --> 00:46:30.833 of RCMAP data into both land fire and NLCD. 00:46:30.958 --> 00:46:34.500 And with that I am happy to take any questions. 00:46:34.500 --> 00:46:38.291 We've got contact information noted here, website link, 00:46:38.333 --> 00:46:42.500 viewer link and please reach out with any questions. 00:46:42.541 --> 00:46:43.916 Perfect. Thanks, Matthew. 00:46:43.916 --> 00:46:44.375 And Matthew 00:46:44.375 --> 00:46:45.583 is going to leave this slide up 00:46:45.583 --> 00:46:47.458 for the rest of the webinar, just so you can write this down 00:46:47.458 --> 00:46:49.208 if you need any of this contact information. 00:46:49.208 --> 00:46:53.208 But for those listening in, I have added a few final links to the chat 00:46:53.333 --> 00:46:55.500 which include more information about RCMAP 00:46:55.500 --> 00:46:58.083 as well as links to sign up for a few of our listeners, 00:46:58.083 --> 00:47:02.208 a link to watch the recorded webinars, and a few social media channels 00:47:02.208 --> 00:47:03.500 to follow up. You're interested. 00:47:03.500 --> 00:47:06.958 I do want to thank Matthew for a really great presentation at this time. 00:47:06.958 --> 00:47:11.625 Please add any final questions you have about the data or tools covered today, or 00:47:11.625 --> 00:47:15.208 perhaps any future data or tutorial ideas you'd like to see with RCMAP? 00:47:15.208 --> 00:47:18.958 Please feel free to add those to the Q&A feature and teams or the chat now 00:47:19.041 --> 00:47:22.333 you can also email them into serve at USGS gov. 00:47:22.416 --> 00:47:25.000 If the chat and QA feature does not work for you, 00:47:25.000 --> 00:47:28.500 but with that, we will transition over to the Q&A portion of the webinar. 00:47:28.583 --> 00:47:32.416 The first question we have is from Scott. 00:47:32.541 --> 00:47:38.333 How does RCMAP reconcile low temporal or spatial support for a given state? 00:47:38.375 --> 00:47:42.583 For example, states east of Colorado or sampled substantially less frequently 00:47:42.583 --> 00:47:43.666 or spatially? 00:47:43.666 --> 00:47:47.083 Even with 70 million training points, there are magnitudes 00:47:47.083 --> 00:47:50.875 more unmonitored pixels and potentially unobserved vegetation. 00:47:50.875 --> 00:47:52.541 Community distributions. 00:47:52.541 --> 00:47:54.083 Yeah, that's a good question. 00:47:54.083 --> 00:47:55.625 The density of training data 00:47:55.625 --> 00:47:59.416 is definitely lower in the Great Plains than it is further west. 00:47:59.416 --> 00:48:03.625 So we do have a number of our high resolution 00:48:03.625 --> 00:48:06.458 satellite sites in the Great Plains. 00:48:06.458 --> 00:48:09.458 We've added a few more in Texas, 00:48:09.583 --> 00:48:12.625 South Dakota, North Dakota, Nebraska, 00:48:12.750 --> 00:48:17.625 and we are investigating always incorporating new lines of data there. 00:48:17.625 --> 00:48:20.875 There are some land fire observations in the Great Plains. 00:48:20.875 --> 00:48:25.000 We're looking into getting some data from the National Park Service there. 00:48:25.041 --> 00:48:27.625 I am a data I believe it's called. 00:48:27.625 --> 00:48:31.500 Yes, that is a bit of a weakness that the training 00:48:31.500 --> 00:48:34.541 data density is lower in that part of the world. 00:48:34.541 --> 00:48:38.333 Also, in general, the training data is less dense 00:48:38.375 --> 00:48:42.166 earlier in the time series relative to later in the time series. 00:48:42.166 --> 00:48:45.166 The BLM Aim data come online in about 2008. 00:48:45.250 --> 00:48:48.666 Our high res site data come online in about 2013. 00:48:48.750 --> 00:48:51.875 We do have some data going back through the whole archive 00:48:51.916 --> 00:48:54.916 through 85, but it is less dense there as well. 00:48:54.916 --> 00:48:57.416 So that is something to note. 00:48:57.416 --> 00:48:58.750 All right. Thanks, Matthew. 00:48:58.750 --> 00:49:00.875 Our next question comes from Megan. 00:49:00.875 --> 00:49:02.041 She had two questions. 00:49:02.041 --> 00:49:05.000 The first was if we can share where the webinar will be posted. 00:49:05.000 --> 00:49:06.833 So I did add that link to the chat. 00:49:06.833 --> 00:49:07.916 You know, she had several partners 00:49:07.916 --> 00:49:11.125 that were not able to attend but are very interested in our sea map. 00:49:11.125 --> 00:49:12.291 So excited to hear that. 00:49:12.291 --> 00:49:15.166 And then Megan also had a question for you, Matthew. 00:49:15.166 --> 00:49:18.708 Was there any exploration of changes in non or excuse me, 00:49:18.875 --> 00:49:20.083 sorry if I'm mispronouncing that. 00:49:20.083 --> 00:49:22.583 Shrub cover and height in the woody encroachment. 00:49:22.583 --> 00:49:24.708 Partners in northeast have expressed interest 00:49:24.708 --> 00:49:28.125 in both encroachment of trees and shrubs, and climate dependency 00:49:28.125 --> 00:49:32.416 and treatment efficacy will likely vary across the functional groups. 00:49:32.500 --> 00:49:37.541 Yeah, they are asking about Artemisia, which is the sagebrush genus. 00:49:37.708 --> 00:49:41.166 And the question is have we explored non 00:49:41.166 --> 00:49:45.208 artemisia changes in the Great Plains. 00:49:45.208 --> 00:49:45.791 Was that it. 00:49:45.791 --> 00:49:49.625 Was there any exploration of the changes in non artemisia shrub 00:49:49.625 --> 00:49:51.958 cover and height in the woody encroachment. 00:49:51.958 --> 00:49:54.958 Yes. So and Megan note in the chat in Nebraska. 00:49:55.000 --> 00:49:56.041 Okay. Yes. 00:49:56.041 --> 00:50:01.666 So that would be captured in our shrub cover 00:50:01.791 --> 00:50:06.750 trend in places where there are no sagebrush present. 00:50:06.875 --> 00:50:10.750 The shrub cover trend would be that precisely. 00:50:10.750 --> 00:50:13.875 So we can see the woody expansion in the Great Plains. 00:50:14.000 --> 00:50:19.500 You can kind of see this pattern where it starts off as an increase in shrubs. 00:50:19.875 --> 00:50:23.291 What those shrubs really are are juvenile trees. 00:50:23.333 --> 00:50:28.375 Trees less than three meters tall would be called shrubs in our product. 00:50:28.375 --> 00:50:32.291 And then eventually, once the tree individuals who pass that height 00:50:32.333 --> 00:50:36.500 threshold, then that becomes a tree cover trend. 00:50:36.791 --> 00:50:39.833 So all of that being said, in the Great Plains, 00:50:40.125 --> 00:50:43.708 the first early warning sign of this woody 00:50:43.708 --> 00:50:48.458 expansion would typically be increasing trend in shrubs, 00:50:48.583 --> 00:50:51.583 which would be, again, those juvenile trees. 00:50:51.583 --> 00:50:52.958 All right. Thank you, Matthew. 00:50:52.958 --> 00:50:55.833 The next question we have and I will note it is 1:00, 00:50:55.833 --> 00:50:58.208 but we're going to extend the webinar for about five more minutes 00:50:58.208 --> 00:50:59.958 while we can get through all these questions. 00:50:59.958 --> 00:51:02.916 So if you need to hop off. Thank you all for joining us. 00:51:02.916 --> 00:51:05.000 But our next question is from Karen. 00:51:05.000 --> 00:51:07.333 Have you been able to map yet? 00:51:07.333 --> 00:51:10.875 Yeah. It's not a it's another is another. 00:51:10.875 --> 00:51:11.208 Yeah. 00:51:11.208 --> 00:51:14.583 The botanists like to make pronunciation difficult. 00:51:14.625 --> 00:51:17.708 Vince Nata is another important invasive grass. 00:51:17.708 --> 00:51:20.375 We have looked into that. 00:51:20.375 --> 00:51:25.166 The challenge we're facing there is it is just not very well 00:51:25.166 --> 00:51:29.708 represented in the BLM Aim database at this point. 00:51:29.875 --> 00:51:33.500 I think there was just a couple hundred maybe observations 00:51:33.500 --> 00:51:36.166 total within that Aim database. We are trying. 00:51:36.166 --> 00:51:39.583 I mentioned the red brome and buffel 00:51:39.583 --> 00:51:43.041 grass being ones that we're going to look into as well. 00:51:43.083 --> 00:51:48.500 A lot of our invasive grass species are really Great Basin centric. 00:51:48.625 --> 00:51:51.875 The cheatgrass medusa head, but also the field brome 00:51:51.875 --> 00:51:55.083 component includes Japanese brome, 00:51:55.083 --> 00:51:59.875 which is a pretty common invasive grass across a lot of the Great Plains. 00:51:59.958 --> 00:52:03.541 So definitely looking into expanding that list, 00:52:03.666 --> 00:52:06.625 I would say the less common 00:52:06.625 --> 00:52:10.583 that species is in the training data, the more difficulty 00:52:10.625 --> 00:52:14.333 we will have mapping it, but definitely looking into it. 00:52:14.375 --> 00:52:15.125 All right. Excellent. 00:52:15.125 --> 00:52:16.875 Our next question is 00:52:16.875 --> 00:52:20.875 can this product or model be available to use outside of the US? 00:52:21.208 --> 00:52:24.125 I want to use it for another one that I'm going 00:52:24.125 --> 00:52:27.375 to have to pronounce Pentagon Harlem mapping. 00:52:27.458 --> 00:52:29.500 Okay. That one I'm not familiar with. 00:52:29.500 --> 00:52:33.125 And no, it's not currently available outside of the US, 00:52:33.208 --> 00:52:38.083 aside from that little portion of Canada that's within the sagebrush biome, 00:52:38.208 --> 00:52:43.083 we are considering expanding the study area to cover all of Conus. 00:52:43.125 --> 00:52:43.833 At least. 00:52:43.833 --> 00:52:47.500 I'm not sure if that would cover the users study area of interest. 00:52:47.500 --> 00:52:52.750 There are some global fractional cover products available based on Motus. 00:52:52.750 --> 00:52:57.458 If I recall correctly, it wouldn't be as detailed as what we do. 00:52:57.500 --> 00:53:01.500 It would just be shrub or basis, maybe bare ground, 00:53:01.541 --> 00:53:04.541 I think maybe three components and maybe an NPV 00:53:04.583 --> 00:53:07.583 and non photosynthetic veg component as well. 00:53:07.583 --> 00:53:08.500 And of course 00:53:08.500 --> 00:53:13.791 or facial resolution to my knowledge that would be available at a global scale. 00:53:13.791 --> 00:53:18.625 But we currently don't have any plans to expand RCMAP at a global level. 00:53:18.666 --> 00:53:19.916 Okay. Thank you Matthew. 00:53:19.916 --> 00:53:23.625 And our last question is broadly, how do RCMAP data 00:53:23.666 --> 00:53:27.541 compare to RAP data or Rangeland Analysis Platform? 00:53:27.625 --> 00:53:29.625 Apologies if you covered this and I missed it. 00:53:29.625 --> 00:53:31.708 Yeah that's a really great question. 00:53:31.708 --> 00:53:33.500 And it's a big topic. 00:53:33.500 --> 00:53:38.125 So the rangeland analysis platform data, the RAP data produced by the ARS 00:53:38.458 --> 00:53:42.708 and University of Montana and others is similar 00:53:42.708 --> 00:53:46.625 in many respects to our data. 00:53:46.875 --> 00:53:49.625 The key difference is our training data. 00:53:49.625 --> 00:53:54.041 We've got those high resolution sites that are not used in the rangeland 00:53:54.041 --> 00:53:55.291 analysis platform. 00:53:55.291 --> 00:53:58.291 Our post-processing methods differ. 00:53:58.291 --> 00:54:00.500 Some of our modeling methods differ. 00:54:00.500 --> 00:54:04.041 All of this difference is good news for users. 00:54:04.041 --> 00:54:08.958 It's good that the products are different in that you get more statistical power 00:54:09.166 --> 00:54:12.250 in having a different modeling framework. 00:54:12.250 --> 00:54:15.333 So generally speaking, the two data 00:54:15.375 --> 00:54:20.500 sets are pretty similar, especially the bigger and temporal scale 00:54:20.541 --> 00:54:23.541 you look at, the more similar they will become. 00:54:23.833 --> 00:54:27.583 There are some important differences that I don't have time to go into. 00:54:27.625 --> 00:54:32.875 We do have a paper in the works with rap people as coauthors 00:54:32.875 --> 00:54:35.875 talking about some of these differences, so look out for that. 00:54:35.875 --> 00:54:40.166 Did also mention earlier that there is benefit from looking 00:54:40.166 --> 00:54:44.500 at a consensus of both Rap data and or C map data. 00:54:44.500 --> 00:54:48.583 So for example, if shrub cover was of interest, a mean 00:54:48.958 --> 00:54:52.416 or some other statistic, considering the data from both RAP 00:54:52.416 --> 00:54:56.750 and RCMAP could be of benefit and often is of benefit. 00:54:56.791 --> 00:54:57.958 Excellent. Thank you Matthew. 00:54:57.958 --> 00:55:01.458 We are at 105 so we are going to conclude the webinar. 00:55:01.500 --> 00:55:04.708 Clarinda, I do see that you had a question about some open spaces. 00:55:04.708 --> 00:55:07.625 So I'll make sure we touch base with you offline on that one. 00:55:07.625 --> 00:55:10.333 But I did want to thank everyone for attending the webinar. 00:55:10.333 --> 00:55:14.458 If you have any additional questions later on about see map, the RCMAP Viewer, 00:55:14.458 --> 00:55:18.416 or any data, tools or services the USGS EROS provides, 00:55:18.500 --> 00:55:21.791 you can always email our team at custserv@USGS.gov. 00:55:22.041 --> 00:55:23.083 I'd like to once again 00:55:23.083 --> 00:55:26.333 thank our speaker, Matthew for his time and great presentation 00:55:26.416 --> 00:55:28.375 and our team in the background that helped answer 00:55:28.375 --> 00:55:30.833 some of those great questions you all had today. 00:55:30.833 --> 00:55:34.291 If you'd like to be notified of when the next webinar will occur, please 00:55:34.291 --> 00:55:36.166 sign up for our EROS User Services 00:55:36.166 --> 00:55:39.875 listserv by emailing that custserv@USGS.gov email. 00:55:40.166 --> 00:55:42.875 Again, thank you all for taking the time to join us today. 00:55:42.875 --> 00:55:44.750 We hope you have a wonderful rest of the day. 00:55:44.750 --> 00:55:46.208 We'll see you next time. Thank you.