The raster image (0.5 meter resolution) represents the 2019 results from the Habitat Evolution Mapping Project (HEMP) Decadal Update (2019 & 2021). Nineteen biotic and abiotic classes (also referred to as ‘habitats’) were mapped using 8- band multispectral imagery collected by the Worldview-2 satellite sensor on June 8th, 2019. There are 12 total biotic classifications representing dominant tidal marsh vegetation alliances or associations. These include 7 salt marsh vegetation classifications (Cordgrass, Cordgrass /- Pickleweed, Pickleweed, Pickleweed /- Jaumea, Saltgrass, Pickleweed /- Gumplant, Alkali Heath); 3 brackish marsh vegetation classifications (Alkali Bulrush, Spearscale, and Pepperweed - although Pepperweed, and Spearscale to a lesser degree, are considered invasive species); and 2 freshwater marsh vegetation classifications (Freshwater Bulrush, Cattails). There are also 2 upland vegetation classifications: “Ruderal” (representing a number of dominant ruderal species like radish and mustard), and “Alkali Grasses”. Abiotic classifications include: “bare earth”, “water”, “wrack”, “mudflat”, and “mudflats with biofilm”.We manually edited the raster in targeted areas with known mapping issues. This habitat dataset achieved 85% overall accuracy at the Alliance level and 80% overall accuracy at the Association level.
<br/><br/>
The goal of the update is to better inform the South Bay Salt Pond Restoration Project (SBSP), it’s partners and stakeholders, about the status of marsh and mudflat habitats within the study area and to assess changes to these habitats since the first study a decade ago. The preliminary datasets produced from the HEMP2 study, and presented in this report, provide the SBSP Project Management Team (PMT) important resources for quantifying and visualizing the current distribution and extent of tidally influenced salt, brackish and freshwater marshes, as well as tidal mudflats, within the study area (See Figure 1). In combination with the results from the first phase of HEMP (Fulfrost, B., Thomson, D., 2012), they also allow the PMT and its’ partners and stakeholders to better understand how tidal marshes have evolved over the last decade (2009-2019). Area of Interest: Baylands south of San Mateo Bridge in San Francisco BayTime Period of Data: June 8th, 2019Format: ESRI file geodatabase raster (.gdb) Resolution: 0.5 metersCoordinate System: UTM, Zone 10 North, WGS84All habitats and mudflats were derived using semi-automated classification of high-resolution satellite imagery. On June 8th, 2019, we obtained a multispectral satellite image from the Worldview-2 sensor captured just below Mean Lower Low Water (MLLW), providing full exposure of tidal mudflats, and limiting water within marshes. For mapping habitats, we repeated the methods used between 2009 and 2011 because they achieved accurate results, are easily replicable, and widely used. For mudflats we used methods developed in our 2016 pilot study (Fulfrost, B., 2017) using imagery from the Worldview-3 sensor. Both our methods for mapping vegetative (and non-vegetative) habitats and our separate process for mudflats, which rely on identifying and distinguishing habitats based on their unique spectral responses, are optimized when using multispectral imagery like that found in the Worldview-2 (or Worldview-3) sensor. We began with the development and description of a set of ecologically relevant habitat types, which was completed in 2009-2011. Our methodology for mapping the vegetative (and related abiotic) habitat types then consisted largely of six steps. First, we focused on the identifying days and timing of satellite image acquisition at Mean Lower Low Water (MLLW), QA/QC of delivered imagery, and preprocessing to prepare the imagery for analysis, including orthorectification and pansharpening. Second, we conducted GPS based ground truthing for
validation and later calibration of the model as well as for building our initial training sites. Third, we developed an initial spectral model (i.e. training sites for each habitat type) and ran a supervised classification of imagery using this model. Fourth, we reviewed model output both in GIS and in the field in order to calibrate model results. Fifth, our model review would lead to improvements to the spectral model and changes to our training sites, rerunning of the supervised classifications, resulting in new and improved model output. Sixth, we would repeat Steps #4 and #5 until the model output was well calibrated, resulting in our final habitat model. We then manually edited the raster in targeted areas with known mapping issues. Our last step was a field-based validation of the final model output resulting in the final habitat datasets summarized in the final report. Contact Bran (
[email protected]) to obtain a copy of the final report.