habitat suitability modelling: raster analysis · image manipulation methods and geoprocessing...

1
Image manipulation methods and geoprocessing operations can be used to prepare raster format images for inclusion in a GIS model designed to identify areas of likely wildlife habitat and connecting corridors. These methods could include the use of remote sensing data and image classification to identify probable landcover classes, querying and extraction of applicable data within these images, conversion of vector data into raster for inclusion in analysis, and use of patch analysis to compile standardized areas containing mean values of the assembled inputs. Floerkea proserpinacoides (common name ‘false mermaidweed’) is a broadly-distributed but rare annual spring ephemeral, occurring in disjoined populations throughout North America, including Nova Scotia. Floerkea’s presence in Nova Scotia was first recorded in 1948, when a population (still extant) was found in the Coldbrook area. Its preferred habitat is riverbanks, swamps, and other riparian areas, although it is also sometimes seen in non-wetlands. Landsat 8 Imagery Results Introduction The image above is a true colour representation of the study area created from Landsat 8 imagery that has been pansharpened against the panchromatic band to improve visual clarity, and clipped to the area’s boundaries. Land Cover Classification Image 1: F. proserpinacoides Data Preparation Patch Analysis In addition to the multi-band Landsat imagery, I selected the following vector datasets to be used as inputs for the model: Nova Scotia Topographic Database: roads, watercourses, and waterbodies. Nova Scotia Department of Natural Resources: forestry and surficial geology. All data was cropped to the study area boundaries. For each vector dataset, it was necessary to isolate a data subset containing attribute values that I’ve hazarded will be meaningful in establishing a habitat model. These subsets were then converted to raster format, allowing them to be processed using ArcGIS’ Map Algebra functionality and other raster-specific analysis tools. To this end, the roads data was queried into separate ‘paved’ and ‘unpaved’ sets, my rationale being that, although it’s conceivable that a Floerka population would exist adjacent to a road, potential corridors of propagation would likely be impeded by motorized vehicle traffic. The following commands were executed in ArcGIS Pro to perform this task (note that in the NSTDB datasets, the identifying feature code for paved roads contains a ‘1’): Floerkea populations have an association with sugar maple stands, so areas where maple was the dominant species were similarly extracted from the forestery data: For this model, I have made the assumption that Floerkea will be more likely to grow in areas geologically rich in organic material, so only polygons of this type were extracted from the surficial geology dataset. Shown below is my process for selecting these areas from the original ‘surfGeo’ vector dataset and exporting them as a new raster image called ‘organicSoil’. Land cover classification is a process that assigns each cell of an image a class or category pertaining to land use or vegetation. The process consists of: • Obtaining and analyzing input data. • Producing graphic signatures for analysis. • Performing the classification. • Evaluating the results and revising the process if necessary. The accuracy of this technique is subject to the spatial resolution of the imagery and the extent of field work (ground truthing) data collected and analyzed for the purpose. Supervised classification is the technique used in this model. It requires that an analyst select sample pixels within the image, to be defined as known members of a given class. This creates a training area. For example, an area known to be a hay field will be enclosed in a polygon. This captures its pixel values, which represent the unique spectral characteristics of the area. The graphic signature of a hay field will be distinct from that of a lake or rock quarry. The processing software then uses these training area values to automatically classify all other pixels. This is in contrast to unsupervised classification, whereby software determines which pixels are statistically likely to represent the same sort of spectral response. The application defines the groupings, which are then interpreted by people. The image below shows the training area schema I created for use in this model, hoping to differentiate between coniferous forest, deciduous forest, developed areas, barren land, wetlands, agricultural fields, and open water. These sample areas were verified by airphoto interpretation and my own knowledge of Nova Scotian terrain. The training area schema can be saved as GIS data for future use. ArcGIS Pro’s Class Probability geoprocessing tool was used to perform the classification. This tool takes a raster file and a training set as inputs, producing a multiband raster stack containing one band for each class. Pixel values of these bands indicate the probability, expressed as a percentage, that the pixel is a member of the given class. A low probability value may be indicative of a mixed class—for example, a pixel representative of a mixed deciduous/coniferous forest stand could score approximately 50 in the ‘hardwood’ layer and 50 in the ‘softwood’ layer. For habitat modelling work, these graduated values can provide us with significantly more meaningful results than a simple Boolean yes/no array. In ArcGIS Pro, the Create Signatures function is used to apply the training area sample data to the entire raster. The output is a small ASCII format file containing tables that provide a reference between the previously-defined sample areas and the spectral characteristics of the study area extent. Create Signatures can be accessed as a geoprocessing tool or as a function within the ArcGIS Raster Calculator. ExtractByAttributes('classProb/Band_4', 'VALUE > 75') Image 2: Key map with study area. In the second image, darker tones indicate a higher likelyhood of deciduous forest stands as predicted by Class Probability. Image 3: Landsat 8 imagery for study area. River line features were converted to raster first and then merged with a raster containing identified open water area (created using the classification process described below), to be used in definition of riparian zones. This was accomplished using the Cell Statistics tool. SelectLayerByAttribute_management('surfGeo','NEW_SELECTION', '"UNIT" LIKE \'Organic Deposits\'') FeatureToRaster_conversion(’surfGeo’, 'OBJECTID', ‘organicSoil’) Data Querying, Data Conversion Creation of Training Areas Supervised Classification Graphic Signatures Patch analysis applies a consistent geometric tessellation to spatial data, allowing for analyis to occur with minimal sampling bias. I used ArcGIS Pro’s Generate Tessellation tool to produce a 30 × 30 hectare grid of hexagons for this model. I then converted this vector output to raster format, after assigning each hexagon feature a unique ID using the ArcGIS Field Calculator tool. This unique ID became the ‘Value’ field for the newly-created raster, allowing me to identify and query individual hexagons even after the conversion to raster. GenerateTessellation_management('hex30', arcpy.env.extent, 'HEXAGON', '30 Hectares') All the necessary data has now been assembled, queried, and converted such that it is now ready for inclusion in habitat suitability and corridor indentification models. Image 9: Class Probability tool output, displayed here as an RGB composite where red = band 7 (wetlands), green = band 4 (deciduous), and blue = band 1 (open water) Preparation CellStatistics(['tLake', 'tRiver'], 'MAXIMUM', 'DATA') Landsat 8 provides land observation remote sensing data at a spatial resolution of 30 metres for the visual, near-infrared, and shortwave infrared wavelengths and 15 metres for panchromatic images. Its temporal resolution is 16 days. This data serves as a resource for many fields of research, including agriculture, planning, geology, surveillance, and educa- tion. Image scenes representing approximately 34, 000 square kilometres of land may be obtained from the United States Geological Survey website at no cost. Image 4: Surficial geology vector data with features of interest selected. Image 5: Selected data converted to raster. Image 6: Converting ‘river’ vector features to raster and merging its values with the ‘lakes’ raster. Image 7: Creation of training areas. To better interpret the image, I leveraged the Landsat near-infrared band to create an image composite that allowed me to visually discern differences in vegetation types that cannot normally be seen in the electromagnetic wavelengths visible to humans. Using this band combination, it is typical for coniferous canopies to appear darker, and deciduous canopies brighter. Image 8: Use of a false-colour image composite for visual interpretation of landcover. Image 10: Class Probability raster, band 4. Image 11: Values greater than 75% for band 4. Image 12: Output of Generate Tessellation. Image 13: Data converted to raster. An association between deciduous (in particular sugar maple) forest stands and the plant has been documented by some researchers. Floerkea proserpinacoides may be threatened by habitat loss due to development, off-road vehicle activity, invasive species, or any activities that contaminate or alter its associated aquifers. Because of the plant’s relatively brief lifecycle, vulnerable populations may potentially go unnoticed. ArcGIS Pro 2.2.0, including functionality provided by the Spatial Analyst extension, will be used to analyze a study area in rural Nova Scotia as a possible candidate for suitable habitat, which could be used to inform biodiversity conservation efforts in the area. Results of these analyses must be subject to verification by a biologist or other expert, ideally supported by ground-truthing or other empirical data. This analysis has been conducted as a test of modelling techniques only and has not been verified. arcpy.AddField_management('hex30', 'ID', 'SHORT') arcpy.CalculateField_management ('hex30', 'ID', '!OBJECTID!') arcpy.FeatureToRaster_conversion('hex30','ID', 'hex30R') For those output bands displaying a meaningful correlation to their assigned class, data was queried and extracted to a new raster. This allows for isolation of the relevant information, to be used in the steps to follow. It also provides us with the opportunity to query only data with a chosen degree of fidelity to its supposed class. Image 10 shows a single band of the Class Probability tool output, for the ‘deciduous’ class. Data values range from zero (lowest probability, black) to 100 (highest probability, white). Image 11 shows the resultant image after the following Raster Calculator command was executed to retain only pixels scoring 76% or higher: SelectLayerByAttribute_management('roadLine', 'NEW_SELECTION', '"FEAT_CODE" LIKE \'%1\') FeatureToRaster_conversion('roadLine', 'OBJECTID', 'roadPaved') SelectLayerByAttribute_management('roadLine','SWITCH_SELECTION') FeatureToRaster_conversion('roadLine', 'OBJECTID', 'roadUnpaved') SelectLayerByAttribute_management('forest', 'NEW_SELECTION', '"SPECIES" LIKE \'SM%\') FeatureToRaster_conversion('forest', 'OBJECTID', 'maple') Class Probability 1. Using ArcGIS Pro and Landsat 8 Imagery for Habitat Model Data Preparation Habitat Suitability Modelling: Raster Analysis This map is produced as a portion of the requirements of the Geographic Sciences Program at the Centre of Geographic Sciences, NSCC, Lawrencetown, Nova Scotia. The product is unedited, unverified and intended for educational purposes only. © 2018 COGS Produced by: Nicole White Date: October 2018 Sources: Floerkea proserpinacoides in Nova Scotia. Erskine, D. Schofield, W. 1948. Rhodora, Vol. 50. pg. 283 -- 284. https://www.biodiversitylibrary.org/page/617550#page/338/mode/1up Minnesota Department of Natural Resources Rare Species Guide. https://www.dnr.state.mn.us/rsg/index.html Image 1: USDA-NRCS PLANTS Database / Britton, N.L., and A. Brown. 1913. An illustrated flora of the northern United States, Canada and the British Possessions. 3 vols. Charles Scribner's Sons, New York. Vol. 2: 441. Landsat 8 imagery: The United States Geological Survey. Scene LC08_L1TP_008029_20180707_20180717_01_T1. July 17, 2018. https://earthexplorer.usgs.gov/ (October 1, 2018). Vector data for roads and waterbodies: Nova Scotia Topographic Database (NSTDB). 1:10000 Enhanced Topographic Database. September 10, 2018. https://geonova.novascotia.ca/ (October 1, 2018). Surficial geology vector data: Nova Scotia Department of Lands and Forestry. Surficial Geology Map of the Province of Nova Scotia. 1992 - 1993. https://bit.ly/2Nub0Co (October 1, 2018). Forest cover vector data: Nova Scotia Department of Lands and Forestry. Current Forest Data (Cycle 2 & 3 with additional updates). June 2015. https://bit.ly/2IfizuN (October 1, 2018). Base for keymap and illustrations: Esri. "Topographic" [basemap]."World Topographic Map". Sept. 26, 2018. https://bit.ly/1dRTvEn. (October 1, 2018).

Upload: others

Post on 27-May-2020

10 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Habitat Suitability Modelling: Raster Analysis · Image manipulation methods and geoprocessing operations can be used to prepare raster format images for inclusion in a GIS model

Image manipulation methods and geoprocessing operations can be used to prepare raster format images for inclusion in a GIS model designed to identify areas of likely wildlife habitat and connecting corridors.

These methods could include the use of remote sensing data and image classification to identify probable landcover classes, querying and extraction of applicable data within these images, conversion of vector data into raster for inclusion in analysis, and use of patch analysis to compile standardized areas containing mean values of the assembled inputs.

Floerkea proserpinacoides (common name ‘false mermaidweed’) is a broadly-distributed but rare annual spring ephemeral, occurring in disjoined populations throughout North America, including Nova Scotia. Floerkea’s presence in Nova Scotia was first recorded in 1948, when a population (still extant) was found in the Coldbrook area. Its preferred habitat is riverbanks, swamps, and other riparian areas, although it is also sometimes seen in non-wetlands.

Landsat 8 Imagery

Results

Introduction

The image above is a true colour representation of the study area created from Landsat 8 imagery that has been pansharpened against the panchromatic band to improve visual clarity, and clipped to the area’s boundaries.

Land Cover Classification

Image 1: F. proserpinacoides

Data Preparation

Patch Analysis

In addition to the multi-band Landsat imagery, I selected the following vector datasets to be used as inputs for the model:

• Nova Scotia Topographic Database: roads, watercourses, and waterbodies.• Nova Scotia Department of Natural Resources: forestry and surficial geology.

All data was cropped to the study area boundaries.

For each vector dataset, it was necessary to isolate a data subset containing attribute values that I’ve hazarded will be meaningful in establishing a habitat model. These subsets were then converted to raster format, allowing them to be processed using ArcGIS’ Map Algebra functionality and other raster-specific analysis tools.

To this end, the roads data was queried into separate ‘paved’ and ‘unpaved’ sets, my rationale being that, although it’s conceivable that a Floerka population would exist adjacent to a road, potential corridors of propagation would likely be impeded by motorized vehicle traffic. The following commands were executed in ArcGIS Pro to perform this task (note that in the NSTDB datasets, the identifying feature code for paved roads contains a ‘1’):

Floerkea populations have an association with sugar maple stands, so areas where maple was the dominant species were similarly extracted from the forestery data:

For this model, I have made the assumption that Floerkea will be more likely to grow in areas geologically rich in organic material, so only polygons of this type were extracted from the surficial geology dataset. Shown below is my process for selecting these areas from the original ‘surfGeo’ vector dataset and exporting them as a new raster image called ‘organicSoil’.

Land cover classification is a process that assigns each cell of an image a class or category pertaining to land use or vegetation. The process consists of:

• Obtaining and analyzing input data.• Producing graphic signatures for analysis.• Performing the classification.• Evaluating the results and revising the process if necessary.

The accuracy of this technique is subject to the spatial resolution of the imagery and the extent of field work (ground truthing) data collected and analyzed for the purpose.

Supervised classification is the technique used in this model. It requires that an analyst select sample pixels within the image, to be defined as known members of a given class. This creates a training area. For example, an area known to be a hay field will be enclosed in a polygon. This captures its pixel values, which represent the unique spectral characteristics of the area. The graphic signature of a hay field will be distinct from that of a lake or rock quarry. The processing software then uses these training area values to automatically classify all other pixels.This is in contrast to unsupervised classification, whereby software determines which pixels are statistically likely to represent the same sort of spectral response. The application defines the groupings, which are then interpreted by people.

The image below shows the training area schema I created for use in this model, hoping to differentiate between coniferous forest, deciduous forest, developed areas, barren land, wetlands, agricultural fields, and open water. These sample areas were verified by airphoto interpretation and my own knowledge of Nova Scotian terrain.The training area schema can be saved as GIS data for future use.

ArcGIS Pro’s Class Probability geoprocessing tool was used to perform the classification. This tool takes a raster file and a training set as inputs, producing a multiband raster stack containing one band for each class. Pixel values of these bands indicate the probability, expressed as a percentage, that the pixel is a member of the given class. A low probability value may be indicative of a mixed class—for example, a pixel representative of a mixed deciduous/coniferous forest stand could score approximately 50 in the ‘hardwood’ layer and 50 in the ‘softwood’ layer. For habitat modelling work, these graduated values can provide us with significantly more meaningful results than a simple Boolean yes/no array.

In ArcGIS Pro, the Create Signatures function is used to apply the training area sample data to the entire raster. The output is a small ASCII format file containing tables that provide a reference between the previously-defined sample areas and the spectral characteristics of the study area extent.Create Signatures can be accessed as a geoprocessing tool or as a function within the ArcGIS Raster Calculator.

ExtractByAttributes('classProb/Band_4', 'VALUE > 75')

Image 2: Key map with study area.

In the second image, darker tones indicate a higher likelyhood of deciduous forest stands as predicted by Class Probability.

Image 3: Landsat 8 imagery for study area.

River line features were converted to raster first and then merged with a raster containing identified open water area (created using the classification process described below), to be used in definition of riparian zones. This was accomplished using the Cell Statistics tool.

SelectLayerByAttribute_management('surfGeo','NEW_SELECTION','"UNIT" LIKE \'Organic Deposits\'')

FeatureToRaster_conversion(’surfGeo’, 'OBJECTID', ‘organicSoil’)

Data Querying, Data Conversion

Creation of Training Areas

Supervised Classification

Graphic Signatures

Patch analysis applies a consistent geometric tessellation to spatial data, allowing for analyis to occur with minimal sampling bias. I used ArcGIS Pro’s Generate Tessellation tool to produce a 30 × 30 hectare grid of hexagons for this model.

I then converted this vector output to raster format, after assigning each hexagon feature a unique ID using the ArcGIS Field Calculator tool. This unique ID became the ‘Value’ field for the newly-created raster, allowing me to identify and query individual hexagons even after the conversion to raster.

GenerateTessellation_management('hex30', arcpy.env.extent, 'HEXAGON','30 Hectares')

All the necessary data has now been assembled, queried, and converted such that it is now ready for inclusion in habitat suitability and corridor indentification models.

Image 9: Class Probability tool output, displayed here as an RGB composite where red = band 7 (wetlands), green = band 4 (deciduous), and blue = band 1 (open water)

Preparation

CellStatistics(['tLake', 'tRiver'], 'MAXIMUM', 'DATA')Landsat 8 provides land observation remote sensing data at a spatial resolution of 30 metres for the visual, near-infrared, and shortwave infrared wavelengths and 15 metres for panchromatic images. Its temporal resolution is 16 days. This data serves as a resource for many fields of research, including agriculture, planning, geology, surveillance, and educa-tion. Image scenes representing approximately 34, 000 square kilometres of land may be obtained from the United States Geological Survey website at no cost.

Image 4: Surficial geology vector data withfeatures of interest selected.

Image 5: Selected data converted to raster.

Image 6: Converting ‘river’ vector features to raster and merging its values with the ‘lakes’ raster.

Image 7: Creation of training areas.

To better interpret the image, I leveraged the Landsat near-infrared band to create an image composite that allowed me to visually discern differences in vegetation types that cannot normally be seen in the electromagnetic wavelengths visible to humans.

Using this band combination, it is typical for coniferous canopies to appear darker, and deciduous canopies brighter. Image 8: Use of a false-colour image composite for

visual interpretation of landcover.

Image 10: Class Probability raster, band 4. Image 11: Values greater than 75% for band 4.

Image 12: Output of Generate Tessellation. Image 13: Data converted to raster.

An association between deciduous (in particular sugar maple) forest stands and the plant has been documented by some researchers.

Floerkea proserpinacoides may be threatened by habitat loss due to development, off-road vehicle activity, invasive species, or any activities that contaminate or alter its associated aquifers.

Because of the plant’s relatively brief lifecycle, vulnerable populations may potentially go unnoticed.

ArcGIS Pro 2.2.0, including functionality provided by the Spatial Analyst extension, will be used to analyze a study area in rural Nova Scotia as a possible candidate for suitable habitat, which could be used to inform biodiversity conservation efforts in the area.

Results of these analyses must be subject to verification by a biologist or other expert, ideally supported by ground-truthing or other empirical data. This analysis has been conducted as a test of modelling techniques only and has not been verified.

arcpy.AddField_management('hex30', 'ID', 'SHORT')arcpy.CalculateField_management ('hex30', 'ID', '!OBJECTID!')arcpy.FeatureToRaster_conversion('hex30','ID', 'hex30R')

For those output bands displaying a meaningful correlation to their assigned class, data was queried and extracted to a new raster. This allows for isolation of the relevant information, to be used in the steps to follow. It also provides us with the opportunity to query only data with a chosen degree of fidelity to its supposed class.

Image 10 shows a single band of the Class Probability tool output, for the ‘deciduous’ class. Data values range from zero (lowest probability, black) to 100 (highest probability, white). Image 11 shows the resultant image after the following Raster Calculator command was executed to retain only pixels scoring 76% or higher:

SelectLayerByAttribute_management('roadLine', 'NEW_SELECTION','"FEAT_CODE" LIKE \'%1\')

FeatureToRaster_conversion('roadLine', 'OBJECTID', 'roadPaved')

SelectLayerByAttribute_management('roadLine','SWITCH_SELECTION')

FeatureToRaster_conversion('roadLine', 'OBJECTID', 'roadUnpaved')

SelectLayerByAttribute_management('forest', 'NEW_SELECTION', '"SPECIES" LIKE \'SM%\')

FeatureToRaster_conversion('forest', 'OBJECTID', 'maple')

Class Probability

1. Using ArcGIS Pro and Landsat 8 Imagery for Habitat Model Data Preparation

Habitat Suitability Modelling: Raster Analysis

This map is produced as a portion of the requirements of the Geographic Sciences Program

at the Centre of Geographic Sciences, NSCC, Lawrencetown, Nova Scotia.The product is unedited, unverified and intended

for educational purposes only.© 2018 COGS

Produced by: Nicole WhiteDate: October 2018

Sources:

Floerkea proserpinacoides in Nova Scotia. Erskine, D. Schofield, W. 1948. Rhodora, Vol. 50. pg. 283 -- 284. https://www.biodiversitylibrary.org/page/617550#page/338/mode/1upMinnesota Department of Natural Resources Rare Species Guide. https://www.dnr.state.mn.us/rsg/index.htmlImage 1: USDA-NRCS PLANTS Database / Britton, N.L., and A. Brown. 1913. An illustrated flora of the northern United States, Canada and the British Possessions. 3 vols. Charles Scribner's Sons, New York. Vol. 2: 441.

Landsat 8 imagery: The United States Geological Survey. Scene LC08_L1TP_008029_20180707_20180717_01_T1. July 17, 2018. https://earthexplorer.usgs.gov/ (October 1, 2018).

Vector data for roads and waterbodies: Nova Scotia Topographic Database (NSTDB). 1:10000 Enhanced Topographic Database. September 10, 2018. https://geonova.novascotia.ca/ (October 1, 2018).Surficial geology vector data: Nova Scotia Department of Lands and Forestry. Surficial Geology Map of the Province of Nova Scotia. 1992 - 1993. https://bit.ly/2Nub0Co (October 1, 2018).Forest cover vector data: Nova Scotia Department of Lands and Forestry. Current Forest Data (Cycle 2 & 3 with additional updates). June 2015. https://bit.ly/2IfizuN (October 1, 2018).

Base for keymap and illustrations: Esri. "Topographic" [basemap]."World Topographic Map". Sept. 26, 2018. https://bit.ly/1dRTvEn. (October 1, 2018).