Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross
Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross
Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross
Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross
Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross
Description: The initial objective for producing maps of changing forest cover was to obtain digital imagery completely covering the southern half of the Midland quad that went back in time as far as possible. All aerial photography used was taken in the summer with-on conditions. Recent digital imagery was obtained first, in the form of black and white digital orthophoto quarter quads (DOQQ's) from 1991 and color infrared (CIR) imagery in a digital format from 2002. The Kansas Data Access and Support Center (DASC) provided the 1991 DOQQ's at 1 m ground resolution in UTM NAD27 projection system. Two DOQQ's from 1991 were required to cover the southern half of the Midland Quad.
The historical data that were located contained between 9 to 11 black and white aerial photography prints for each of the following years: 1941, 1954, 1966 and 1976. These hardcopy photographs were archived at the Douglas, Jefferson and Leavenworth County Planning and/or Farm Service Agency offices. All photographs were in a 24 by 24-inch enlarged format, which were originally flown at a scale of 1:20,000 and collected on 10 by 10-inch negatives covering 2.84 ground miles on a side. The agencies that possessed these photos would not allow prints to leave their facilities, so a laptop computer. The agencies that possessed these photos would not allow these prints to leave their facilities, so a laptop computer and small format (8.5 by 14 in) desk scanner were used to capture the images in a digital format. Each 24 square-inch photo print was scanned into 4 images (or fewer if part of the image fell outside the study area), each spanning approximately 1 square ground mile (one PLSS section). Using a 300 dpi scanning resolution, the images had a ground pixel size of approximately 0.64 m, which allowed for down sampling at the time of rectification to match the 1 m resolution of the 1991 and 2002 data sets. This process resulted in 24 images per study year from 1941 to 1976, one for each of the 24 PLSS sections in the study area.
Data Preprocessing
Prior to classification, the next objective involved pre-processing the digital aerial photography. Image geo-rectification was performed in the ERDAS Imagine software package using the Image Geometric Correction tool. Seven to ten ground control points (GCP's) per image were used to perform a second order polynomial transformation with an RMS error of 2 m or less. When resampling the imagery to 1 m ground resolution, the cubic convolution resampling method was used to maintain the spatial character of the data and avoid the jagged results characteristic of nearest neighbor resampling. The more recent historical photographs, from 1976 and 1966, were geo-rectified using the 1991 DOQQ as the reference image, while the older historical photographs from 1954 and 1941 used the newly geo-rectified 1966 imagery for reference. This shift in the reference imagery for the rectification of the older photography was done because it became increasingly difficult to identify GCP objects, such as roads and buildings, in common with the 1991 DOQQ.
Once rectified, the yearly data sets were still made up of multiple individual images: 24 images for the 1941, 1954, 1966 and 1976 dates; and 2 images for 1991. Before classification, it was decided that mosaicking the images together into a complete dataset would be beneficial to reduce mismatch along the edges of images during classification. ERDAS Imagine's Mosaic tool was used to stitch all photos from each year together into a single raster dataset covering the study area. During mosaicking, images were cropped in overlapping areas, manual cut lines were drawn and feathering techniques were used to blend datasets together with the goal of reducing the visibility of seam lines. The mosaicking process resulted in six georeferenced aerial photography data sets from 1941 to 1991 with an approximate temporal interval of 10 years (Figure 2).
Attempting to classify the entire study area for each year would have been unfeasible for several reasons. Of primary concern was the contrast difference between and within the original photographs. For example, the spectral information describing a tree object in a bright portion of the mosaic could be confused with the spectral information of a non-forest grassland object in a darker section. Also, the large amount of time required to manually check and clean up the results of the classification for the entire quad would make it difficult to break up the workflow on this project into discrete manageable pieces. For these reasons, the final preprocessing step divided up the complete mosaics into the 24 PLSS sections contained in the mosaic (Figure 3). This process resulted in 144 section-sized images (six years of 24 sections each), which were the basis of the classification procedure.
Matthew D. Dunbar 's 2005 MasterThesis: Mapping, Analyzing, and Visualizing 60 Years of Forest Cover Change in Northeast Kansas.
In 2007, the resulting images were converted to NAD83 by Jorgina Ross