Image Classification

Supervised image classification for Black Water National Wildlife Refuge.

Problem Statement

The Black Water National Wildlife Refuge managers would like to develop a management plan for the refuge and are in need of land cover data. The available data is for aerial photograph from Black Water National Wildlife Refuge which has not been classified yet, hence it’s difficult to know what to develop where based on the land cover type available on that area. The purpose of this assignment is therefore to develop imagery classes based on the photograph provided and the training samples to be developed. The calculate area in meters for each class to inform the land development plan for the refuge.

Analysis procedure

To address the problem, I used ArcGIS Pro 3.0.2. I used a 4 band aerial photograph for Black Water National Wildlife Refuge named m_3807640cr.tif (file size about 22 MB). This data was provided by the course instructor, and had NAD 1983 UTM Zone 18N projection system. Tools used includes training sample tool, classify tool, export raster, build raster attribute table tool, add field tool, and calculate field.

I started the assignment with producing training sample polygons from the image which was initially symbolized with the following band combination; Red=band 4, Green= band 3, Blue=Band 2. This was meant to enhance the distinct land class boundaries. The training sample was done using training samples managers which is found under imagery classification tools. Each class was drawn with more than one training samples. All the common samples were compressed together. After completion, the training samples were saved in a folder which could easily be traced. Thereafter, under the imagery, then classification tools, I opened the classify tool which was used to classify the photograph, and I used Support Vector Machine under classified, and my training samples under training sample.

Secondly, I revised my training samples by adding few more samples on water, herbaceous, and developed areas. Then I reclassified the image using this revised training sample, holding all other variables constant.

I exported the raster’s from both products, built attribute tables in the exported raster using build raster attribute table from the geoprocessing pane. Then I added new field in each table separately and named it area. This new field was added from the geoprocessing pane as it was not possible to add directly. Then I calculated the area separately using the count field in each table. Since the image was in meters, and the cell size was 1*1, I calculated this new area field by multiplying the count field by 1 to find total area per class in meters.

Process Diagram

Results

The results indicated a significant difference between the land area covered by training sample 1 and 2 as shown of figure 3 and 4 below. Figure 3 results indicated high land area covered by all other classes except water and Hibiscus. These two classes were more pronounced in figure 4.

Application and Reflection

Problem description: Malawi as a landlocked country, relies heavily on agriculture to fuel its economy. The main challenge is on competing resource needs, as the same land is needed in other industry. There is need to conduct a land cover use classification to have an in-depth understanding of what land resources are found where, and what land is covered by what.

Data needed: Satellite imagery rater layers for Malawi. The data will be obtained from NASA through USAID's GeoCenter.

Analysis procedure: I will use the supervised classification in updating the land cover classification for Malawi. I will start with developing training samples for all the classes, picking at-least 5 samples per class. Then I will reclassify the satellite images obtained using the training samples. Finally, I will export the reclassified raster layer, and display it on a map to show which areas have what classes.