Mid-Atlantic RESAC
Field Data Collection Campaign

The Chesapeake bay watershed covers 64,000 mi² area, extending from the "finger lakes" of up-state New York to the North Carolina and Virginia borders. The watershed includes 37 physiographic provinces, and just as many different farming practices and land use management strategies. Choosing and implementing a land cover mapping strategy for an area of this size and diversity is a daunting task.

The mapping challenge

Most macroscale land cover maps produced from mosaiced Landsat scenes use an unsupervised or supervised classification methodology. Unsupervised classification requires significant manual interpretation and is therefore impractical for applications that aim for near real-time updating capabilities. Nor are they sufficiently flexible to tailor individual applications. Supervised classification requires a-priori knowledge of the landscape. This too can be difficult, as most examples of Landsat based classifications depend on either ancillary data for training and validation or in-situ field validation. In the case of the Chesapeake Bay, there are few spatially consistent data sets available for comparison. Of the few data sets that exist, most are either coarse in spatial resolution, out of date, or are error prone.

Obtaining field data conincident with Landsat image acquisitions can also be a challenge. Digital aerial photorgraphy (DOQQs) are an excellent source of reference data but repeat coverage only occurs about once in five years. Interannual variation in vegetation, urban development, and row crop rotations all vary at less than a 5-year time interval.

Field sampling methodology

With these limitations in mind, a field campaign was designed and executed in order to both characterize the mid-Atlantic landscape and to generate a data set for training and validation. The field campaign was conducted in the Summer of 1999. Two teams of students set out for one month of data collection. Each team was equipped with a laptop computer equipped with geographic information system (GIS) software and Landsat Thematic Mapper (TM) scenes acquired within a few weeks of the field campaign. Real-time differential geographic positioning systems (GPS) were used to determine precise locations within the Landsat TM imagery. Field teams were thus able to follow their route on the laptop while in the field, and driving between points. USGS digital line graphs (DLGs), which include roads, populated places and streams, were overlayed on the imagery to aid in navigation.

Each of the two field teams had a unique focus. The first team concentrated on forested land cover types and the second on agriculture. Of 12 land cover classes (using the Anderson Level II system), forests, pastures and row crops are the most difficult to discern from visual inspection of imagery alone. Built environments are more easily identifiable from remotely sensed imagery, with few exceptions.

Data collected for each forested sample include a survey of general species composition, tree heights, topography, and canopy closure. The agriculture team collected data on crop type, farming practices including irrigation and tilling, vegetation height, canopy closure, and soil moisture and color.

Field Data Collected

In four weeks, the two teams collected data at nearly 1300 sites representing approximately 75% of the watershed. These data have been compiled into a GIS. The number of sites per land cover class observed in the field is shown in the following table.

The field data have been incorporated into the land cover mapping effort. Land cover products derived in part from these data can be seen in the section of the mid-Atlantic RESAC web site.