San Francisco Bay: Change Detection and Mapping
Use of Landsat Thematic Mapper Multitemporal Images to Detect and Map Changes in the San Francisco Bay Area
Ecosystems within coastal regions are undergoing dramatic changes and are impacted by human activities from economic, land use, and resource development. Habitats for wildlife and fish species in many coastal areas are being threatened by some of these human activities, as well as possible sea-level rise. In addition, biodiversity and wildlife resources in these coastal areas can be altered by changes in global climate. To understand and map important ecosystem parameters, such as suspended sediment loads and surficial changes in wetlands along coastal areas, vegetation types and density, landforms and their changes, as well as urban growth patterns, we must employ our capabilities to detect and map surface variability temporally, spatially, and spectrally. It is critical to have tools, procedures, and data to not only study ecosystems, but to be able to monitor them over time and space as population expands and/or climate changes occur. Satellite multispectral digital image (Chavez and MacKinnon, 1994) data are particularly helpful for monitoring and detecting surface variability. In particular, these satellite data allow an area to be studied from a regional perspective over time.
The images shown here are portions of two Landsat Thematic Mapper (TM) digital images recorded on September 7, 1984 and September 16, 1993 from an altitude of approximately 700 km (438 miles). Landsat TM images have an approximate 30 meter resolution and cover an area that is about 185 by 185 km (115 by 115 mi). The images shown cover the San Francisco Bay and Delta region and the color composites were generated using Landsat TM spectral bands 2 (green), 4 (near-infrared), and 5 (mid-infrared) as blue, green, and red, respectively. The digital data were processed using the USGSMIPSsoftware package to correct for both geometric and radiometric distortions and enhance the overall contrast and local detail. These images are part of a set of data being used to study regional changes and help develop tools and procedures to monitor ecosystems. The images were geometrically registered to each other so that the same pixel within both images represents the same location on the ground and they had both radiometric calibration and corrections applied (Chavez, 1989). The City of San Francisco and the Golden Gate Bridge can be seen to the left of center; extreme South Bay is towards the bottom center; San Pablo and Suisun Bays are to the northwest of center; the Sacramento and San Joaquin Rivers are shown in the north/northeast portion of the image. Highways 101 and 880, along with other main roads, can also be seen in the digital images. Also, note the clouds along the coast in the 1993 image.
The geometrically registered and radiometrically calibrated multitemporal image pair was used as input to our change-detection procedure to generate digital change images from the visible, near-infrared, and mid-infrared spectral bands (Chavez and MacKinnon, 1994). The change image results for the three spectral bands were then used to make the color composite shown here. The three spectral bands will detect many of the same temporal changes, however, there will be some changes that are detected better in one band versus the others or some changes that will be detected by only one or two of the three spectral bands (e.g., sediment load differences are seen better in the visible band and, in some cases, vegetation differences are seen better in the near- and mid-infrared bands). In general, some of the colors seen in the change-image composite correspond to the following:
Magenta - These pixels/locations were more vegetated in 1984 than in 1993. Examples of this are the loss of wetlands along the Bay or where there is less natural vegetation/grass in the foothills in 1993 due to differences in the amount of rainfall. Also, the vegetation type can be the same on both dates but either the density or level of maturity of the vegetation on the two dates is different (i.e., more dense/mature in 1984 than 1993).
Green - Green is the opposite of magenta (i.e., more vegetated in 1993 than in 1984). This mostly occurs in agricultural areas and is related to crop rotation, crop type changes, or crop density/maturity differences between the two dates.
Blue - Most areas that are blue represent urban growth between 1984 and 1993. These are areas where either new residential or industrial development occurred between 1984 and 1993. Examples include the blues seen on the east side of extreme South Bay which represents mostly new industrial buildings, while the blues north of Carquinez Strait, in Livermore, and elsewhere are areas with new residential development between the two dates.
White - White areas show where dramatic changes in surface brightness/reflectance occurred; these areas were much brighter in 1993 than in 1984. Examples of this are the clouds in the lower left portion of the image, and areas that were burned in 1984 and have recovered by 1993. Other examples are seen north of Mallard Island and an area due east from the image center/east side of the foothills.
Black - The black areas also had a dramatic change in surface brightness/reflectance, but were darker rather than brighter in 1993 versus 1984. Examples of this are areas burned in 1993 or were natural landscape areas in 1984 that were converted to agricultural lands and recently tilled.
South Bay Area
Carquinez Strait Area
Honker Bay Area
Included here are three different subareas at 1X, 2X, and 4X zoom. The subareas shown from left to right are: the east side of extreme South Bay, just north of Carquinez Strait, and just northeast of Honker Bay, respectively. The images demonstrate the level of detail contained in the Landsat TM thirty-meter resolution images, and the digital change image generated using the 1984 and 1993 data. The 1984, 1993, and change image subareas were digitally mosaicked next to each other for comparison and display purposes. These image products show the level of information that can be extracted from the data.