🛰️ Remote Sensing Applications

Vegetation, Wildfire & Land Cover Analysis

This project uses satellite imagery to answer three real-world environmental questions:

• Is this crop field healthy? — tracked using NDVI over time
• How bad was this wildfire? — mapped using burn severity index (dNBR)
• What's on this land? — classified using Landsat 8 spectral bands

A SQL database is used to store the results after raster analysis is complete in ArcGIS Pro, where the outputs are exported as attribute tables and then loaded for further analysis. — making it possible to filter, aggregate, and summarize findings without reopening the GIS software. For example, querying which pixels had negative NDVI, how much area fell into each burn severity class, or how different land cover types compare in NIR reflectance.

All analysis was done in ArcGIS Pro and ERDAS Imagine, with results queried using SQL.

---

🗄️ SQL — Querying the Results

Raster attribute tables exported from ArcGIS Pro are loaded into a relational database. SQL is then used to summarize and filter findings across all three parts of the project.

Part A — Track average NDVI across the crop season:

SELECT
    image_date,
    ROUND(AVG(ndvi), 3)  AS avg_ndvi,
    ROUND(MIN(ndvi), 3)  AS min_ndvi,
    ROUND(MAX(ndvi), 3)  AS max_ndvi
FROM ndvi_timeseries
GROUP BY image_date
ORDER BY image_date;

Part B — See how much area burned at each severity level:

SELECT
    severity,
    COUNT(*)                               AS pixel_count,
    ROUND(AVG(dnbr), 3)                   AS avg_dnbr,
    ROUND(100.0 * COUNT(*) /
        SUM(COUNT(*)) OVER (), 2)         AS pct_of_total
FROM burn_severity
GROUP BY severity
ORDER BY avg_dnbr DESC;

Part C — Compare how different land cover types reflect NIR light:

SELECT
    class_name,
    ROUND(AVG(band5), 3)  AS avg_nir
FROM land_cover
GROUP BY class_name
ORDER BY avg_nir DESC;

---

📁 What's in This Repo

remote-sensing-analysis/
├── PartA/        → Soybean field vegetation analysis (SPOT-4 imagery)
├── PartB/        → King Fire burn severity mapping (Landsat 8)
├── PartC/        → Palm Desert land cover classification (Landsat 8)
├── Remote_Sensing_Applications_Data.pdf   → Full written report
├── LICENSE
└── README.md

---

🌿 Part A — Is the Crop Healthy? (NDVI Analysis)

What I did: Loaded 6 satellite images of a soybean field taken between August and October 1998. Computed NDVI for each image to track how the vegetation changed over the growing season.

What is NDVI? NDVI (Normalized Difference Vegetation Index) measures how green and healthy vegetation is. It uses the difference between Near-Infrared (NIR) and Red light — healthy plants absorb red light and reflect NIR strongly, so a high NDVI = healthy, low or negative NDVI = dead or bare.

NDVI = (NIR − Red) / (NIR + Red)

NDVI Range	Meaning
0.6 – 1.0	Dense healthy vegetation
0.2 – 0.6	Moderate vegetation
0.0 – 0.2	Sparse vegetation
< 0.0	Bare soil, water, or dead plants

What I found:

• NDVI was high in August — crops were actively growing
• By late September NDVI went negative, confirming the crops had died off
• The NIR and Red band values directly reflected the change in chlorophyll activity

Files in PartA/:

File	What it is
AOI_soybeanfield.shp	The boundary of the study area (the field)
AOI_soybeanfield.dbf	Data table attached to that boundary
AOI_soybeanfield.prj	The coordinate system the map uses
AOI_soybeanfield.cpg	Text encoding file (needed for the .dbf to read correctly)
spot-0804.img to spot-1026.img	The 6 satellite images, one per date

A shapefile is actually 4 files working together — .shp (shape), .dbf (data), .prj (projection), .cpg (encoding). All four are required.

---

🔥 Part B — How Bad Was the Fire? (Burn Severity Mapping)

What I did: Compared satellite images of the King Fire taken before (2013) and after (2015) to measure how severely the land was burned, using a metric called dNBR.

What is dNBR? NBR (Normalized Burn Ratio) measures vegetation and moisture using SWIR and NIR bands. By subtracting the post-fire NBR from the pre-fire NBR, you get dNBR — the higher the value, the more severe the burn.

NBR  = (NIR − SWIR) / (NIR + SWIR)
dNBR = NBR_prefire − NBR_postfire

dNBR Range	Burn Severity
< −0.10	Enhanced Regrowth
−0.10 to 0.09	Unburned
0.10 to 0.26	Low
0.27 to 0.43	Moderate
0.44 to 0.65	High
> 0.66	Very High

What I found:

• The core of the fire perimeter showed moderate burn severity
• Areas near water bodies and low-fuel terrain came back as unburned
• Post-fire SWIR reflectance increased significantly — a clear indicator of vegetation loss

Files in PartB/:

File	What it is
KingFire20130730	Pre-fire Landsat image (July 2013)
KingFire20150805	Post-fire Landsat image (August 2015)
NBR_PreFire.tfw	World file — positions the pre-fire NBR raster on the map
NBR_PostFire.tfw	World file — positions the post-fire NBR raster on the map
dNBR.tfw	World file — positions the final burn severity raster on the map

A .tfw world file is a small text file that tells GIS software where on Earth a raster image is located and what scale it's at.

---

🏜️ Part C — What's on This Land? (Land Cover Classification)

What I did: Used Landsat 8 imagery over Palm Desert, California to identify 9 different types of land cover by analyzing how each surface reflects light across 6 spectral bands.

How does it work? Every surface (water, trees, roads, buildings) reflects light differently across the electromagnetic spectrum. By comparing reflectance values across bands, you can distinguish land cover types — this is called spectral profiling.

Landsat 8 Bands Used:

Band	What it captures	Why it's useful
Band 2 — Blue	Short visible light	Detects water, atmospheric scatter
Band 3 — Green	Mid visible light	Vegetation vigor
Band 4 — Red	Long visible light	Chlorophyll absorption
Band 5 — NIR	Near-infrared	Vegetation health (used in NDVI)
Band 6 — SWIR-1	Short-wave infrared	Moisture content, burn detection
Band 7 — SWIR-2	Short-wave infrared	Burn severity (used in NBR)

Land Cover Classes Identified: Bare Soil · Pond · Salton Sea · Cropland 1 · Cropland 2 · Fallow · Trees · Road · Residential

Files in PartC/:

File	What it is
LC08_..._MTL.txt	Landsat 8 metadata — sensor info, acquisition date, calibration
LC08_..._ANG.txt	Sun angle coefficients used to correct reflectance values
palmdesert_stack234567.aux	All 6 bands stacked into one file for analysis
lc08_..._b2–b7.aux	Each band as a separate file

---

⬇️ Data Sources

Large .tif raster files are not included in this repo due to file size. Download the originals here:

• Landsat 8 imagery — USGS Earth Explorer
• SPOT-4 imagery — CNES / Airbus Defence
• King Fire boundary — CAL FIRE

---

🛠️ Tools Used

Tool	Purpose
ArcGIS Pro	Raster processing, NDVI/NBR computation, classification
ERDAS Imagine	Image processing and band analysis
SQL	Querying and summarizing exported raster attribute tables
Landsat 8 (OLI)	Multispectral imagery for Parts B and C
SPOT-4	Time-series imagery for Part A

---

📄 Full Report

See Remote_Sensing_Applications_Data.pdf for all maps, charts, and written analysis.

---