Analyzing Impervious Surfaces using Machine Learning and ArcGIS Pro

Background: In a Louisville, KY neighborhood, identifying impervious surfaces is crucial to managing stormwater effectively. By utilizing ArcGIS Pro, a precise mapping of these surfaces was pursued to aid urban planning and hydrology efforts.

Skills Used: 

1. Spatial Analysis and Raster Analysis: Utilizing ArcGIS Pro to interpret geographic patterns, work with raster data, and perform image classification concerning impervious and pervious surfaces.
2. Machine Learning: Training, testing, and retraining models to classify satellite imagery accurately and iteratively enhancing the model based on initial results for higher accuracy.
3. Data Visualization and Map Symbology: Creating visual representations through map symbology and other visualization techniques to understand and present the distribution of impervious surfaces across the neighborhood.
4. Geoprocessing and Data Management: Employing geoprocessing tools like "Tabulate Area," joining and relating datasets, and modifying attribute tables to calculate and manage impervious surface area data per land parcel effectively
   

Solution: An image classification ML model was trained to accurately classify impervious and pervious surfaces. The results were then merged with parcel data, visualizing which land plots contained more impervious areas.

Potential Impact: With a detailed map of impervious surfaces per parcel, urban planners can now devise better stormwater management strategies, potentially leading to a fairer stormwater fee distribution among the neighborhood's landowners.

Introduction

Urban planners and hydrologists aim to manage rainwater in urban settings effectively. This management largely hinges on the understanding of impervious and pervious surfaces. Impervious surfaces, like rooftops, roads, and sidewalks, prevent water from soaking into the ground, while pervious surfaces, such as grass and open fields, allow water to penetrate the soil. The distinction between these surfaces aids in drainage planning, flood risk assessments, and the determination of stormwater fees. My exploration took place in Louisville, Kentucky, to unravel the extent of impervious surfaces within a neighborhood.

Analysis

The task involved delving into machine learning and spatial analysis capabilities within ArcGIS Pro to deduce the extent of impervious surfaces within a neighborhood in Louisville, Kentucky. The cornerstone of this endeavor was the classification and training of a machine learning model to distinguish between impervious and pervious surfaces based on the aerial imagery and geospatial data acquired from Louisville's open data portal.

The initial phase comprised the classification of various surfaces within the dataset. The classes identified in the model were as follows:

• Impervious Surfaces: Roofs, roads, driveways, and other man-made structures.
• Pervious Surfaces: Grass, bare earth, water, and other natural surfaces.

The Image Classification Wizard within ArcGIS Pro proved to be an indispensable tool for this phase. This tool facilitated the classification of imagery into distinct classes based on the spectral characteristics detected from the aerial imagery.

Post initial classification, it was imperative to train the machine learning model to recognize and correctly classify the various surfaces. The training entailed the creation of training samples that would serve as the model's reference in distinguishing between impervious and pervious surfaces. I ran the model to classify the surfaces within the neighborhood.

Despite the training, the first pass of the model exhibited several misclassifications, which underscored the necessity for further refinement. I manually adjusted the training samples using the polygon tool to correct the misclassifications and enhance the model's accuracy. The retraining of the model was paramount to ensure the final classifications were as accurate as possible.

Upon achieving satisfactory accuracy post-retraining, the following step was to merge the model's classification data with the parcel data within the neighborhood. This merger was executed using the Tabulate Area tool in the Geoprocessing pane, which facilitated the calculation of the area of impervious and pervious surfaces within each parcel. The parcel layer was then joined with the impervious area data to create a comprehensive dataset.

A visualization of parcel lots was created to elucidate the distribution of impervious surfaces within the neighborhood. The Symbology tools in ArcGIS Pro were utilized to symbolize the parcels, enabling a clear visual representation of areas with higher impervious surfaces. The Graduated Colors symbology was particularly useful in depicting the variations in impervious surface area across different parcels.

This analysis spotlighted the extent of impervious surfaces within the neighborhood. It showcased the robust capabilities of machine learning and geoprocessing tools within ArcGIS Pro in handling such spatial analysis projects.

Conclusion

While lakes may seem like pervious surfaces, they are considered less permeable than grass or bare earth as they don't absorb water into the ground.

Urban planners can glean insightful deductions from these analyses to mitigate impervious surfaces within communities. They could endorse using permeable materials for pavements or advocate for green roofing systems, among other sustainable urban planning practices.

This endeavor underscored how machine learning paired with geospatial analysis can provide simplified yet potent solutions to urban planning predicaments.