2012 ASPRS Track, Intertidal Coastal Vegetation Mapping, Joe Chuita

1. Nearshore Substrate Mapping withMulti-Spectral Aerial Imagery

2. California Sea Grant supports science-basedmanagement, conservation and enhancement ofCalifornias coastal and aquatic resources throughresearch, extension and education. Commissioned Ocean Imaging to create a baselinedataset of kelp canopy, shallow subtidal andintertidal bottom substrate, and estuarinegroundcover of Californias North Central Coast

3. Important for Marine Protected Areas (MPAs) Protect diversity and abundance of marine life Sustain, conserve, and protect marine life for Economic value To replete low numbers Improve recreational, educational, and researchopportunities Protect unique habitats

4. Details Used ArcGIS Maximum Likelihood Classification Algorithm DMSC (Digitial Multi-Spectral Camera) Multi-spectral data 1meter resolution OI camera Acquired additional ADS40 ( Airborne Digital Sensor) data at30cm resolution Varying tide levels in imagery Decision to overlay imagery to create final dataset 290km stretch of the mid-northern coast of California Field work Ground reference data, GPS points, photos Accuracy Assessment

5. Imagery GPS points Field photos UCSC biologists

6. Mosaicked imagery scenes by Quadrangle (22) Large areas to classify 13.2km average linear coastline Mis-classification of pixels Near Infrared channel - Terrestrial, Kelp, Red/Brown Algae, veg is red Sand, Rock white, grey, and tan colors Separated the mosaicked image into three tiers (Raster Clip) Terrestrial, above water line, water Digitized a shapefile to create tiers Minimize algorithm error Clip Raster Tool > Train and classify image Ocean Imaging 1 meter resolution data Some quads have only DMSC data Acquired ADS40 30cm resolution data

7. Coast LengthFUGRO quad (km)Pigeon Point 9San Gregorio14.2Half Moon Bay 15.4 Montara Mtn13.8 SF South14 Pt Bonita14.3 San Rafael2.3 Bolinas15.9DoublePt11.4Inverness4.8 DrakesBay35.4Tomales 18.7ValleyFord 5.6 Bodega 17.8DucansMills8.9ArchedRock11.6 FortRoss3 Plantation 18.9StewartsPoint 19.9Gualala 14.8SaundersReef 6.2 PointArena 14.2 290.1Avgs. 13.2

8. ADS40 30cm Data DMSC 1m Data

9. 16 Classes Whitewash Water Sandy Beach Mixed Red/Brown Algae Tidepool/Shadow Terrestrial Vegetation Unvegetated Rock Wrack vegetation that has washed up on shore - Kelp Blue-Green Algae Cobble Man-made Driftwood Unknown Vegetation, Green Algae, Eel Grass

10. Create shapefile to isolate data that needs to be edited Clip the data Reclassify the clipped raster to edit values Change all values to new value based on your shapefile digitizing whole beach which classified as rock Change certain values rock to beach, red/brown algae to wrack, leave others as classified (water, whitewater)

11. Mosaic to new raster Down-sample to 1 meter to be consistent Some quads only had 1 m DMSC data Numerous iterations of this process

12. Assess the accuracy of Ocean Imagings classification todetermine classification precision and how likely that whatyou see classified is what is on the ground Congalton Matrix Used field and office data GPS points Field photos Field notes Multi-spectral imagery Virtual Earth Used imagery, photos, notes, and Virtual Earth to createadditional accuracy points in ArcGIS

13. Used 114 accuracy points which were GPS derived or createdand cross verified with multiple sources Used ArcGIS tools to verify the accuracy Attribute table of the point file contains field class type of the ground reference data Used the Extract Values to Points tool to obtain the raster value of the final classification which keeps the field classification type in the attribute table as well The ground reference values of the point shp and the raster classification values are based on a corresponding number code In the attribute table, the ground reference and extracted class values were concatenated in a new field and the combinations were queried and inserted into the Excel matrix

14. Error Matrix Producers Accuracy The producer can determine how well a category was classified Dividing the total # of correctly classified records in a category by the total # of sample reference data records in that category Users Accuracy The end user can predict how likely a classified sample is to represent what is actually on the ground Dividing the total number of sample reference data records in a category by the number of correctly classified records in that category Overall Accuracy Sum of correct classifications divided by total sample reference points Our overall accuracy was 86.0% Assessing the Accuracy of Remotely Sensed Data: Principles andPractices by Russell G. Congalton and Kass Green

15. First time that the California coast has been mapped to thisextent at this spatial level Will have 2/3 of the California coast mapped by the end of2013 Questions

16. California Sea Grant. Retrieved September 17, 2012, fromhttp://www-csgc.ucsd.edu/index.html California Sea Grant North Central Coast Marine ProtectedArea Baseline Program RFP. Retrieved September 4, 2012,from http://www-csgc.ucsd.edu/FUNDING/APPLYING/NorthCentralCoastMPA2009-10.html#Background Congalton, R.G.; Green, K. 1999. Assessing the Accuracy ofRemotely Sensed Data: Principles and Practices Boca Raton,FL: Lewis Publishers.

2012 ASPRS Track, Intertidal Coastal Vegetation Mapping, Joe Chuita

Documents

data dmsc

meter resolution data

dmsc data numerous iterations

imagery decision

classified water

mosaicked imagery scenes

tiers raster clip terrestrial

new raster