1. Land cover designation derives initially from the results of a supervised classification of Landsat TM imagery of 2000 and 2001. Agricultural land uses were identifed by subsequent expert examination and reclassification of the TM results, based on ca. 2000 USGS Digital Ortho-Quadrangle images, taxlot records, and county agricultural production data.
Agricultural Land Use/Land Cover
This document describes the process used to improve the accuracy of a classified satellite image depicting land use/land cover for the Willamette River Basin in the year 2000. The goals of this process were to two-fold: 1) to correct mis-classifications of agricultural land cover so that the data better conformed to county-level acreage statistics, and 2) to better represent the spatial extent of the agricultural and non-agricultural land. Time constraints hampered the full attainment of these goals, however, the general distribution of land cover throughout the basin, especially in the primary agricultural regions, has been improved.
Agricultural crop acreages from the satellite data were computed for each county and compared against acreage estimates from Oregon State University Agricultural Statistics Service 2000 acreage report, Oregon Department of Agricultural Statistics 2000 acreage reports, and the 2002 USDA agricultural Census. These values helped to determine the degree and type of mis-classifications occurring within the data set.
The initial satellite classification was spatially aligned to the grid wrbcm3i obtained from <http://oregonstate.edu/dept/pnw-erc>. Nearest-neighbor operations were then used to nibble out single pixels of non-natural vegetation. Next, non-agricultural classes (1-62, 86, 87, 89, 98) were set to a single value to reduce the visual complexity of the grid. Finally, ESRI's ArcINFO grid editing operations were used to alter the cell values in the grid, based on visual interpretation and/or ancillary data that suggested the type of agricultural land cover present. At the end of the project, nearest- neighbor operations were again used to remove regions of agricultural cells of three pixels or less for all classes except pasture. For pasture, only single pixels were replaced. These values represent the smallest number of cells used during editing to represent crop acreage.
The following data sets were used to help determine the spatial extent and land cover/land use occurring within the agricultural land of the Willamette River Basin: " 2000 Digital Orthophoto Quads (DOQs) raster data " University of Oregon ISE county tax lot vector data " Oregon Department of Agriculture Nursery and Christmas Tree growers databases " 1994/1995 DOQs raster data " Public Land Ownership from the State GIS Service Center vector data
The first two data sets provided the most generally useful information. DOQs provided the spatial extent of the agricultural fields and often afforded clues to the crops grown thereon. The tax lot data contains owner name and address that, by searching the Internet and Oregon Department of Agricultural databases, identified locations for many vineyard, nursery, and fresh fruit and vegetable operations.
The list below describes the types of crops associated with each agricultural crop class and the nature of the confusion that occurred, and may remain, with the LULC 2000 grid.
66 Hybrid Poplar: Hybrid poplar was over-predicted in all counties, especially Lane, Multnomah and Washington. Confusion was primarily between closed hardwood and closed mixed forest classes. Added fields of hybrid poplar were located based on GPS data or were locations known to the analyst.
67 Grass Seed (Annual rye grass, perennial rye grass, tall fescue, other grass and legume seeds): Grass seed acreage was under-predicted in all counties. This is the dominant crop in Linn, Marion, Polk, Benton, and Yamhill Counties, accounting for roughly half of all crop acreage in the Willamette Valley. Grass seed is confused with grain (71), field crops (90), and row crops (68).
68 Row Crops (Sweet corn, beans, peas, other fresh vegetables and truck crops): Row crop acreage was over-predicted in all counties except Multnomah, particularly Lane, Linn, Marion, and Polk counties. This class is confused with many classes, including natural vegetation along watercourses and roadways, grass seed (67) and hay (83). The satellite classification was generally successful in identify fields of row crops, thus the main effort was directed toward locating and changing mis-classified pixels.
71 Grains (Wheat, barley, other grains): Grains were over-predicted in Clackamas, Lane, and Linn counties, and under predicted in Washington county. Grains were generally confused with grass seed (67) and hay (83), and in Washington, perhaps also with turfgrass/parks (91)
72 Nursery Crops (In ground or container crops and horticultural test plots): Nursery crops were not included in the original satellite classification. However, many were discernable on the DOQs so they were included in the updated classification when possible. Nursery crops that have not been re-classified often have a characteristic mixture of the classes row crop (68) and bare/fallow (88).
73 Berries & Vineyards (Blueberries, caneberries, and wine grapes): The berry/vineyards class was over-predicted in all counties, especially Benton, Linn, and Lane counties. This class was often confused with rural residential land, pasture (85), and, natural vegetation (86,87). Ownership data helped to identify locations of berry and vineyard acreages, as did their characteristic appearance on the DOQs.
74 Double Cropping (grains in spring and row crops in late summer): To identify double cropping requires a time series of data, which was not available for the DOQs, nor are there statistics on these fields. Thus, fields that were predominately classified as double cropped retained that classification.
75 Hops: Continuous fields of hops were well classified, and easy to identify by tone and pattern on the DOQs. This is one of the most accurate class depictions.
76 Mint (peppermint for oil, spearmint for oil): Mint acreage was over-predicted for all counties. Mint was confused with row crop (68), grass seed (67), and turfgrass/park (91). Most of the mint acreage has been identified.
78 Sugar Beet Seed Like hops: This crop was well classified, and was easy to identify from the DOQs. There were fields in the northern valley that had a similar appearance to sugar beet seed, but were classified as field crop. Since this is an aggregate class for this crop, that designation remained.
83 Hay (Grass hay and alfalfa hay): Hay was somewhat over-predicted in Marion, Multnomah, and Washington counties, and under-predicted in all other counties. This was perhaps the most difficult class to interpret, as hay has no striking visual appearance. It is confused with pasture (85), grain (71) and row crop (68). Hay classification was easier in Columbia county where roughly 85% of the acreage is in hay.
85 Pasture (improved and unimproved pastureland): There were no statistics on improved and unimproved pasture, which was discerned primarily through spatial context and texture. This class appears to be confused with hay (83) and natural vegetation (86,87).
88 Bare/Fallow: Fallow fields were well identified in the satellite classification, thought some are mis-classified as rural residential unknown, young orchards (92), or nursery crops (72).
90 Field Crops (Potatoes, peppermint, sugar beets, other field crops): Field crops were over-predicted in all counties, especially Polk, Yamhill and Clackamas counties. This class is confused with natural vegetation (86,87), row crop (68), and grass seed (67).
91 Turfgrass/Park (turf grass, golf courses): Since this class contains non-crop acreage, statistics were not available. The satellite classification was successful in classifying golf courses. This class is confused primarily with mint (76), row crops (68), and natural vegetation (86,87).
92 Orchard (Hazelnuts, cherries, applies, peaches, and other tree and nut fruit): Orchards were over-predicted in all counties, particularly Columbia and Multnomah counties. In these two counties, orchards were confused primarily with forest classes. For the remainder of the counties, this class was moderately well classified in the original grid, with most of the editing work focused on clarifying the extent of the orchard fields.
93 Christmas Trees: The acreage of Christmas trees was moderately well predicted by the satellite classification, which captured many of the Christmas trees farms, however there was also many mis-classifications occurring in rural residential, bare/fallow(88), and forested classes.
95 Woodlot: The acreage of woodlot, defined as land that has been taken out of agricultural production and planted with Douglas Fir or other timber, is unknown. Thus, fields where the majority of pixels were classified as woodlot and that occurred within an agricultural setting retained that classification. There were a limited number of these fields, the majority of pixels were confused with forested classes.
Classification of Non-agricultural Rural Land Cover: In the course of evaluating the agricultural land cover class, any cells that appeared on the digital orthophotoquads as non-agricultural land (primarily forestland, shrub and grasslands, and rural structures) were coded with a single value. Later, this value was used as a mask to segment out and classify these non-agricultural cells in the raster data layer. The objective of this classification was not to visually interpret the land cover, as was done with the agricultural class, but rather to simply use a variety of recent vegetation classification, leveraging the fact that in many instances the non-agricultural land cover would change little over short periods of time. The following datasets were used in this assignment procedure:
WRB00: the initial 2000 land use/land cover satellite classification EC90, a circa 1990 land use/land cover map partially derived from a 1992 satellite classification (WRB90). This dataset is located at: <http://www.fsl.orst.edu/pnwerc/wrb/access.html>
WRB90: a 1992 35-class satellite classification, located at: <http://www.fsl.orst.edu/larse/wrb/wrb.html>.
In the case of WRB00 and WRB90, the grids were spatially aligned to the grid wrbcm3i obtained from <http://oregonstate.edu/dept/pnw-erc> prior to any further processing. The process used was as follows: Cells that were coded non-agricultural and whose class in WRB00 was non-agricultural were coded with the WRB00 class. Of the remaining non-classified cells, any cells that were coded as non-agricultural and whose class in EC90 was non-agricultural were coded with the EC90 class. Next, any remaining non-classified cells coded as non-agricultural that had a non-agricultural code in WRB90 were assigned that code. Finally, any remaining cells were coded to a non-agricultural land cover class as follows: Grass seed, grain, hay, and pasture were recoded to natural grassland (86). Row crop, berry & vineyard, double cropping, hops, mint, sugar beet seed, field crop, and turfgrass/park were recoded to natural shrub (87). Orchard was recoded as closed hardwood (53). Christmas trees was recoded as forest semi-closed mixed (52) Woodlot was recoded as forest closed mixed (54).
At each step, small (one to two pixel) cells of any agricultural class remaining after the reclassification were nibbled out using GRID nearest neighbor routines. This procedure retains wherever possible the initial LULC 2000 satellite classification, supplementing it when necessary with relatively recent (8 year old) data. The final adjustment to the natural vegetation data was a manual modification of land cover classes within a 120-meter buffer of 5th through 7th order (Strahler) streams, derived from the RIVREACH_2 dataset located at <http://www.fsl.orst.edu/pnwerc/wrb/access.html>. The objective of this procedure was to identify large areas of shrub, grassland, or agricultural land within the 120-meter buffer area and either verify that that land cover was correct, or assign a new land cover class that better expressed the appearance of the land cover on the digital orthophotoquads. In this particular case, the only replacement classes used were the classes forest closed hardwood (53) and forest closed mixed (54), depending on visual interpretation and the classes of neighboring cells.
2. Elements from the Oregon Natural Heritage Program GAP 2004 dataset were used to augment the identification of Oak (98), Natural shrub (87), and Forest closed mixed (54).
Based on comments of external reviewers (see Data_Quality_Information), elements from the Oregon Natural Heritage Program GAP 2004 dataset were used to augment the identification of Oak (98), Natural grassland (86), and Forest closed mixed (54).
Reclassing of Oak (class 98) to Forest closed mixed (class 54): Reviewers familiar with the northern counties commented that some LULC 2000 class 98 (oak) was more likely to be mixed forest than oak. LULC 2000 cells in class 98 within Clackamas, Washington, Multnomah and Columbia counties were reclassed based on the following criteria:
IF, in ec90, the cell was in upland forest semi-closed mixed (class 52) OR forest closed hardwood (class 53) OR forest closed mixed (class 54) OR upland forest semi-closed hardwood (class 62),
AND the cell was in GAP 2004 Broadleaf Mesic Seral Forest (class 9),
THEN the cell was reclassed to Forest closed mixed (class 54).
Incorporating GAP 2004 oak classes: Based on reviewer comments, four LULC 2000 classes were selected as locations for incorporating GAP 2004 oak classes. Cells in LULC 2000 forest closed hardwood (class 53), forest closed mixed (class 54), natural grassland (class 86) and natural shrub (class 87) were reclassed to oak (class 98) IF these cells were in GAP 2004 oak woodland (class 47) or GAP 2004 Western Oregon Upland Prairie and Oak Savanna.
Incorporating GAP 2004 Wet Prairie: LULC 2000 cells in GAP 2004 Willamette Valley Wet Prairie (class 67) were reclassed to Natural Grassland (class 86). Cells edited as part of agricultural classification (see Agricultural Land Use /Land Cover and Classification of Non-agricultural Rural Land Cover) were excluded from this reclass.
GAP 2004 Reference: Hak, J.C., C. Tobalske, J.S. Kagan, E.P. Gaines, K.A. Walsh, and E.J. Scheuering. 2004. Current and Historic Ecological System Mapping and Vertebrate Species Modeling for Oregon. Unpublished data and manuscript, Oregon Natural Heritage Information Center, Institute for Natural Resources, OSU, Portland, OR.
3. Urban and rural residential land uses were identified primarily from digital taxlot parcel data provided by nine of the ten counties in the Basin. These use classifications are based on interpretation of a standardized property class coding system established state wide, but modified by individual counties. Examination of DOQ images for examples of each property class designation were used to define the corresponding map class, as described here. From this a standard set of attribute fields, common to all jurisdictions' taxlot coverages contained in our database, was established as defined in this table (PDF alternatives: 1, 2) Note that some jurisdictions did not provide data for some of the fields.
4. Delineation of Willamette River channel and principal tributary confluences was performed by revising the EC90 data on the basis of both color DOQs for urban areas, and black and white DOQs for the rural landscape.
The AC2000 coverage is essentially an updated version of the 1995 channel. The basic methodology was to drape the 1995 vector over the 2000 digital orthophotoquads. The 1995 line was then closely inspected, and changes made to conform to the 2000 DOQ's.
The DOQ's were supplied in MrSid format (from State GIS Service Center via UofO/ISE in Oregon Lambert projection. We used 40 DOQ's (identified by standard Ohio code) which included all DOQ's touched by the slices_v2.1 coverage, plus the Biocomplexity study areas, plus a buffer in the areas of narrow river channel.
Edits to the coverage were made in the old Arc/Info Arc/Edit module. All MrSid DOQ's were converted to grid format using Arc/Catalog conversions. Note that this conversion does NOT restore the full level of detail in the original DOQ's (which, incidentally are tiff's in UTM). The MrSid compression program works by removing some fine scale information in the original DOQ's. However, we were unable to obtain the original uncompressed tiff files in time to complete this project.
We reprojected ac1995_tribx to Oregon Lambert and draped it over DOQ grids. Only changes from AC1995 to 2000 were digitized: large portions of the channel were not changed. We worked at the same scale as for the 1995 channel. Where it was available (Portland Metro only), we also used to 2000 fine scale color DOQ's on Terraserver (<http://terraserver.microsoft.com/>) to inform our decisions about channel feature location. In addition, the 2000 USACE Willamette River Book aerial photos (which are images only, and are not registered) were used as additional information.
The 2000 DOQ's were generally of much better quality (even with MrSid loss of information) than the DOP's we used for the 1995 channel generation. This is partly due to the patched-together nature of the 1995 photos, which were combined from a number of sources. The 2000 DOQ's had better contrast and more uniform levels of detail than the 1995 photos. Consequently, we attempted to avoid "over-digitzing" the 2000 channel. The spatial extent of the 2000 DOQ's was also much broader than that for 1995.
In addition, the 1995 and 2000 photos did not line-up well, particularly in the lower portion of the mainstem Willamette, which we know has not changed due to extensive bank protection. If both banks in the AC1995 coverage were off-set from the 2000 DOQ, we assumed the difference was due to registration error of one DOQ set or the other. If one bank was the same in both 1995 and 2000 and if the other bank showed some difference between the two dates, we assumed the later difference was actual channel change, and digitized it as such.
After digitizing, the coverage was rebuilt to polygons, and all new or changed polygon features were coded as needed.
The database for AC2000 is the same as that for AC1995 except:
Field "SECTION" has been removed. This is a hold-over from early (ca. early 1990's) work, and is no longer relevant.
New field "A1995" added; codes are 1= present in 1995 and 2000; 0 = absent in 1995, present in 2000. This is to identify features that were added, such as extensions of tributaries. Some features will be new (e.g., new islands and side channels, new gravel pits), and others will be new due to better imagery (such as remnant river elements).
Initial work was perfomed by Linda Ashkenas. The first version was reviewed by Rose Wallick, and her comments and suggestions were incorporated into the final version of AC2000.
5. The transportaion classes (18-21) were copied into the current grid map from EC90. Subsequently, cells initially classed 7 TM derived commercial-industrial that were coincident with roads in the TeleAtlas database for Oregon were converted to class 21 Minor roads.