WEEK 1-3 NOTES WITS UNIVERSITY

GIS AND REMOTE SENSING

Week 1
Detailed Notes on Remote Sensing Image Preprocessing & Transformation
1. Introduction to Image Preprocessing
• Purpose: Prepares remote sensing data by correcting errors and standardizing
images for accurate interpretation and analysis. Image preprocessing steps are
essential before analysis, especially in environmental monitoring where
precision affects decisions.
• Real-World Applications:
o Urban Expansion Analysis: Geometric correction ensures images taken
over several years align perfectly to accurately assess changes in urban
sprawl.
o Agricultural Monitoring: Atmospheric correction removes interference
from clouds, haze, and sunlight variations to allow clearer assessment of
crop health across large fields.
2. Types of Image Preprocessing Techniques
• Atmospheric Correction:
o Description: Converts TOA reflectance (raw satellite reflectance) to
surface reflectance (true ground conditions).
o Examples:
▪ Agriculture: Removes cloud shadows to ensure clearer NDVI
calculation, essential for mapping crop health.
▪ Forestry: Using ATCOR in ERDAS to reveal forest canopy details for
accurate biomass assessment.
▪ Tools: FLAASH (ENVI) and Sen2Cor (for Sentinel-2 data) are widely
used.
• Radiometric Correction:
o Description: Adjusts pixel values from raw digital numbers to reflectance
values, improving data consistency.
o Example: In water quality studies, radiometric correction ensures that
spectral differences truly represent water characteristics rather than
sensor variations.
 o Real-World Use: For Landsat images used to track chlorophyll-a in water
bodies, this correction helps isolate the water's natural reflectance.
• Geometric Correction:
o Description: Aligns images spatially by correcting distortions from sensor
movement, Earth rotation, etc.
o Example:
▪ Disaster Management: Aligning images from different times to
track changes post-flood, such as in Mozambique’s floodplains.
▪ Tool: Georeferencing with high-accuracy DEM data aligns data
from satellites like Landsat or Sentinel.
3. Image Transformation and Indices
• Purpose: Uses mathematical combinations of spectral bands to highlight
certain features, enhancing the interpretability and specificity of data.
• Key Indices and Their Real-World Applications:
o Vegetation Indices:
▪ NDVI (Normalized Difference Vegetation Index): Calculates
vegetation health by comparing red and near-infrared reflectance.
▪ Example: NDVI maps of the Amazon rainforest reveal areas
of deforestation and ecosystem health, aiding conservation
efforts.
▪ Chlorophyll Vegetation Index (CVI): Tracks plant chlorophyll
levels, useful for agricultural management to optimize crop yields.
o Water Indices:
▪ MNDWI (Modified Normalized Difference Water Index):
Separates water from land, particularly in areas with soil or built-
up surfaces.
▪ Example: MNDWI aids in flood monitoring by distinguishing
floodwater from nearby land during seasonal rains in the
Nile Delta.
▪ NDWI (Normalized Difference Water Index): Useful for assessing
water content in vegetation or soil, providing insights into drought
conditions.
o Built-Up Indices:
▪ IBI (Index-Based Built-up Index): Highlights built-up areas for
urban planning and zoning.
 ▪ Example: Tracking urban expansion in cities like Nairobi,
this index helps city planners manage growth and
infrastructure needs.
o Burnt Area Indices:
▪ NBR (Normalized Burn Ratio): Used to assess the severity of
wildfires and monitor recovery.
▪ Example: Used in California after wildfires, NBR assists in
understanding burn severity and ecosystem recovery rates.

In-Depth Application Questions

1. Why is image preprocessing essential for remote sensing analyses?
Describe a scenario where unprocessed data could lead to inaccurate
conclusions.
o Example Answer: In assessing the vegetation health of an agricultural
region, failing to perform atmospheric correction could lead to cloud
cover misinterpreted as low NDVI values, causing an inaccurate report on
crop health.
2. Explain how NDVI and MNDWI are used in flood monitoring. Why are these
indices particularly valuable for rapid response?
o Example Answer: NDVI and MNDWI highlight vegetation and water
bodies, respectively. In flood scenarios, MNDWI separates floodwater
from the land, providing real-time flood extent, while NDVI assesses flood
impact on vegetation.

Detailed Notes on Introduction to Active Remote Sensing

1. Overview of Active Remote Sensing
• Definition: Active sensors emit their own energy to interact with a target and
capture the reflected or backscattered energy.
• Characteristics:
o Operates in all-weather and nighttime conditions, offering consistency
where passive sensing might fail.
o Particularly valuable in dense, cloudy regions or areas with variable
daylight.
• Applications:
 o Forestry and Biomass Estimation: In the Amazon rainforest, LiDAR aids
in biomass estimation by penetrating the forest canopy to measure tree
height and canopy structure.
o Urban Infrastructure: Radar-based data can monitor infrastructure
subsidence and stability over time, as seen in large construction projects
in Tokyo.
2. LiDAR (Light Detection and Ranging) Technology
• Functionality: Measures distances by timing laser pulses that reflect back to the
sensor. The time delay helps calculate surface characteristics.
• Real-World Applications:
o Urban Topography: Used in New York City to assess building heights and
street-level infrastructure, aiding emergency response and urban
planning.
o Coastal Mapping: Bathymetric LiDAR maps underwater features for
coastal erosion studies, especially in areas like Florida, prone to
hurricanes and storm surges.
• Advantages:
o 3D Capabilities: Provides detailed models for both land and urban
landscapes.
o Precision: Essential in detecting small surface changes in terrain.
o Versatility: Applicable in diverse settings, from cityscapes to dense
forests.
3. Radar Remote Sensing (Synthetic Aperture Radar - SAR)
• Mechanism: Uses radio waves to capture structural information on Earth’s
surface, including subsurface features in some cases.
• Key Acquisition Modes of SAR:
o Stripmap Mode: Moderate resolution, ideal for land cover mapping over
large regions.
▪ Example: Tracking agricultural changes in India, Stripmap mode
helps distinguish crop types and assess harvest patterns.
o Spotlight Mode: High resolution, focusing on specific areas for detailed
analysis.
▪ Example: Used to monitor structural integrity in buildings post-
earthquake, such as in Nepal after the 2015 earthquake.
 o ScanSAR Mode: Lower resolution, wide coverage suitable for large-area
assessments.
▪ Example: Monitoring glacial melt and ice extent in Greenland, this
mode provides an extensive but lower-resolution view of
environmental changes.
• Real-World Use of SAR Data:
o Environmental Monitoring: Sentinel-1 SAR tracks deforestation and
illegal mining in the Amazon by detecting land cover changes.
o Disaster Management: RADARSAT’s ScanSAR mode provides real-time
flood extent maps, essential for relief efforts in monsoon-prone regions
like Bangladesh.

Application-Based Questions for Mastery

1. Describe a scenario where LiDAR would be preferred over radar for a project.
How does LiDAR’s data precision contribute to project outcomes?
o Example Answer: In forest biomass estimation, LiDAR provides accurate
canopy height and structure, allowing more precise carbon stock
calculations—valuable in conservation programs to track deforestation
impacts.
2. How does SAR’s ScanSAR mode aid in disaster response? Give an example
of a recent natural disaster where such a mode could provide valuable
insights.
o Example Answer: During floods in Southeast Asia, ScanSAR captures
wide flood extents, helping agencies prioritize rescue operations by
highlighting the most affected areas, even under cloudy conditions.
3. Compare the advantages and limitations of using Sentinel-1’s SAR data for
urban monitoring versus environmental monitoring.
o Example Answer: In urban monitoring, SAR tracks subsidence and
structural changes, but high-resolution Spotlight Mode limits coverage.
For environmental monitoring, ScanSAR covers larger areas, yet with
lower detail, suitable for broader landscape changes like deforestation.
Week 2
Detailed Notes on Digital Image Classification
1. Introduction to Digital Image Classification
• Definition: Digital image classification uses algorithms to categorize pixels in
remote sensing images, transforming raw satellite data into thematic maps that
reveal environmental features.
• Applications:
o Land Cover Mapping: Used to classify urban areas, forests, water bodies,
and agricultural land, assisting in city planning and environmental
management.
o Biodiversity Monitoring: Classifies habitats in conservation areas to
track changes over time, as seen in mapping invasive plant species in
South Africa’s Drakensberg region.
2. Types of Image Classification Approaches
• Pixel-Based Classification:
o Characteristics: Classifies each pixel based on spectral values alone,
ideal for high-resolution imagery.
o Advantages: Simple and computationally efficient, often used in projects
with limited spatial context needs.
o Real-World Use: Effective for fine-scale mapping of urban areas, where
each pixel can represent a distinct feature (e.g., buildings, roads).
• Object-Based Classification (OBIC):
o Characteristics: Groups pixels into objects before classifying based on
spectral, spatial, and textural information.
o Advantages: Reduces “salt-and-pepper” noise by grouping similar pixels,
leading to clearer class boundaries.
o Real-World Use: Common in forestry applications, such as delineating
tree crowns in high-resolution images.
3. Classification Types: Supervised vs. Unsupervised
• Supervised Classification:
 o Description: Uses training samples (known reference points) to classify
the entire image based on predefined classes.
o Examples:
▪ Agricultural Assessment: Mapping different crop types by training
on field samples, as seen in classifying crop fields in the
Drakensberg exercise using Sentinel-2 data.
▪ Urban Planning: Detecting urban expansion using classes like
‘Residential’ and ‘Commercial’ based on ground-verified points.
o In Practice: Maximum Likelihood and SVM classifiers were used in your
practical to classify the Drakensberg area into classes such as bare
ground, shrubs, and settlements.
• Unsupervised Classification:
o Description: Automatically groups pixels into clusters without prior
knowledge, based on spectral similarity.
o Examples:
▪ Forest Health Monitoring: Identifying clusters of healthy vs.
stressed trees when training data is unavailable.
▪ Initial Mapping in Unknown Regions: Quickly categorizes land
cover in unexplored areas, such as rainforest mapping where
ground data is difficult to obtain.
o In Practice: ISODATA clustering was used in ENVI to segment the
Drakensberg region into five classes (e.g., grassland, shrubs) without prior
knowledge.
4. Pixel-Based vs. Object-Based Classification
• Pixel-Based Classification:
o Advantages: Simplicity, scalability, and high resolution, making it useful
for capturing small details in urban maps.
o Limitations: Prone to noise; lacks spatial context, leading to isolated
misclassified pixels (salt-and-pepper effect).
o Example: Classifying bare ground and settlement areas in Sentinel-2
imagery, where each pixel represents distinct land use.
• Object-Based Classification:
o Advantages: Incorporates spatial context, reducing noise, and improving
boundary clarity, especially effective in complex scenes like forests or
urban environments.
 o Limitations: Computationally intensive, sensitive to segmentation
quality.
o Example: Segmenting Drakensberg images to distinguish homogeneous
land features (like dense shrub areas), resulting in clearer classifications.
5. Common Traditional and Advanced Classifiers
• Traditional Classifiers:
o Maximum Likelihood Classifier (MLC): Assumes normal data
distribution, commonly used for multispectral data.
▪ Example: Used in the practical exercise to classify Drakensberg
land cover based on spectral reflectance values of vegetation and
built-up areas.
o Minimum Distance Classifier: Assigns pixels to the closest class center,
faster but may be less accurate.
▪ Real-World Use: Useful in simpler classifications, like
distinguishing between land and water bodies in coastal mapping.
• Advanced Machine Learning Classifiers:
o Support Vector Machine (SVM):
▪ Description: Identifies a hyperplane to separate classes
effectively in high-dimensional space.
▪ Example: Used for Drakensberg area classification in ENVI,
providing better differentiation of vegetation types due to high-
dimensional separation capabilities.
o Random Forest (RF):
▪ Description: Uses multiple decision trees to enhance
classification robustness and reduce overfitting.
▪ Example: Effective in classifying complex landscapes like mixed-
use urban areas, where vegetation and built-up areas intersect.
o Artificial Neural Networks (ANN):
▪ Description: Powerful for complex, non-linear data; often used in
deep learning applications.
▪ Example: Suitable for large-scale urban mapping projects where
road networks and buildings require precise boundary detection.
6. Steps in Supervised Classification Workflow
 1. Image Acquisition: Source imagery from reliable databases like USGS or Google
Earth Engine.
2. Image Preprocessing: Apply atmospheric correction and resampling to
enhance data quality.
3. Feature Extraction: Stack bands or indices relevant to the study area (e.g., NDVI
for vegetation).
4. Reference Sampling: Collect field samples or create training points in Google
Earth Pro, ensuring spatial consistency.
5. Classification:
o Use Maximum Likelihood for areas with distinct class distributions.
o Apply SVM when more complex, high-dimensional separations are
needed.
6. Accuracy Assessment: Compare 30% of reserved points using metrics like
Kappa and Overall Accuracy to gauge performance.
Real-World Application of Classification Steps in Class
• Drakensberg LULC Mapping: Using Sentinel-2, classification involved
supervised techniques like Maximum Likelihood and SVM. Reference points for
classes (e.g., grassland, settlements) were digitized in Google Earth Pro. The
outputs highlighted distinct land cover types in the region, valuable for assessing
land use impacts on this biodiversity hotspot.

Application Questions for Mastery

1. Why would an SVM classifier be more effective than MLC in classifying urban
landscapes with mixed features? Describe a practical scenario in which SVM
would provide more reliable results.
o Example Answer: SVM's capability to handle complex, high-dimensional
data is beneficial for mixed-use urban areas, where building materials,
vegetation, and road surfaces create diverse reflectance patterns. For
instance, in classifying high-density cities, SVM can better delineate
residential, commercial, and green spaces.
2. How can object-based classification improve accuracy in a heterogeneous
environment, such as a forested area with mixed vegetation and soil types?
o Example Answer: By grouping pixels with similar spatial and spectral
properties, object-based classification reduces noise and enhances
boundary clarity, making it suitable for forests with varied species. This
 approach is ideal for identifying distinct forest types in conservation
areas, allowing better biodiversity monitoring.
3. Compare the use of unsupervised and supervised classification techniques
in mapping inaccessible, large geographic areas. Under what conditions
would each approach be preferred?
o Example Answer: Unsupervised classification is useful for initial
mapping in unexplored areas, where prior knowledge is limited, such as in
rainforest or desert ecosystems. Supervised classification, on the other
hand, is preferred in more studied areas where accurate, predefined
categories (like land use) are critical, such as mapping agricultural zones.
Week 3
Key Concepts and Applications of Remote Sensing in Water Quality Assessment
1. Importance of Water Resources Monitoring
o Applications: Vital for tracking water availability, assessing environmental
health, supporting agriculture, and managing drinking water sources.
o Examples: Monitoring droughts, river flows, and reservoir levels in areas
prone to water scarcity, such as the Nigerien Sahel and California.
2. Challenges to Water Resources
o Key Threats: Climate variability, pollution, land degradation, and invasive
species impact water quality and quantity.
o Real-World Example: Industrial and agricultural pollutants entering rivers
and lakes globally, as seen in the pollution of the Ganges River in India.
3. Remote Sensing Techniques for Water Quality Assessment
o Mapping Water Bodies: Identifying and classifying water bodies using
satellite imagery, which is essential for tracking seasonal variations and
the impacts of human activities.
o Monitoring Water Quality Parameters: Remote sensing is used to estimate
parameters like turbidity, chlorophyll-a, and pH.
o Applications: Useful in tracking algal blooms in lakes and identifying
pollution events in rivers.
4. Types of Data and Datasets
o Common Datasets: Sentinel-2, Landsat 8, and GRACE mission data are
frequently used in water resource assessments.
o Examples: Sentinel-2 has been used in South Africa for Vaal Dam studies
and globally in coastal monitoring.
5. Remote Sensing of Water Quality Indicators
o Chlorophyll-a: Indicates algae presence, which can affect drinking water
quality and ecosystem health.
o Turbidity: Measures water clarity and can indicate sediment or pollution
levels.
o Applications: Used in assessing drinking water reservoirs and
understanding agricultural runoff impacts on water clarity.
 6. Remote Sensing Methods for Data Collection and Analysis
o Methods: Image classification, threshold-based approaches (using water
indices), and regression models.
o Example Models: Linear regression for basic analyses, with advanced
models like Random Forest for more complex relationships.
o Practical Example: Using Sentinel-2 data and regression modeling in
XLStat for analyzing Vaal Dam pH and temperature data.
7. Regression Models in Water Quality Assessment
o Modeling Approach: Extract pixel values corresponding to field GPS
points, then create regression models.
o Applications: Enables large-scale mapping of water parameters, aiding
policymakers in regions like South Africa's Vaal Dam for water quality
management.
8. GRACE Mission for Water Depth Monitoring
o Overview: Measures gravitational changes due to water mass
movements, useful for assessing large-scale water storage changes.
o Applications: Effective in tracking drought impacts and groundwater
changes in areas like California's Central Valley.

Questions for Review

Long Questions
1. Describe how remote sensing can be used to assess water quality. What are
the key indicators, and how do these contribute to understanding water
health?
2. Discuss the applications of remote sensing in monitoring chlorophyll-a and
turbidity. Provide specific examples of where these applications are critical.
3. Explain the regression model process for water quality monitoring using
Sentinel-2. How do field measurements and remote sensing data interact in
this approach?
4. How does the GRACE mission contribute to water resource monitoring, and
what are its limitations and advantages? Provide examples of its real-world
applications.
5. Analyze the limitations of using remote sensing for water quality monitoring
in reservoirs like the Vaal Dam. How might these limitations be mitigated in
future studies?
Short Questions
1. What are the primary challenges facing water resources globally?
2. List two indicators of water quality that can be monitored using remote
sensing.
3. What is the purpose of the GRACE mission in relation to water monitoring?
4. Name one advantage of using regression models in water quality
assessment with remote sensing.
5. How does chlorophyll-a concentration impact water quality?

Additional Real-World Examples for Application

1. Algal Bloom Tracking in Lake Erie
o Application: Using Landsat and Sentinel data to monitor seasonal algal
blooms, aiding in water management for drinking supplies.
2. Turbidity Monitoring in the Amazon River
o Description: Monitoring suspended sediment levels during rainy seasons
helps predict flooding and track erosion impacts.
3. Coastal Pollution Monitoring in the Gulf of Mexico
o Application: Remote sensing helps track oil spills and nutrient runoff from
agricultural lands, supporting cleanup and conservation efforts.
4. Groundwater Storage Assessment in the Middle East using GRACE
o Application: GRACE data has highlighted depleting groundwater levels in
regions like Iran, driving international water management discussions.