Week 7: Lab

EDS 223: Geospatial Analysis & Remote Sensing

Author

Annie Adams

Published

November 10, 2026

The first image taken by Landsat 8 on March 18,2023,showing where the Great Plains and the Rocky Mountains meet in CO.

In this lab we’ll work with our first remotely sensed dataset! We’ll read the metadata file distributed alongside every Landsat scene to understand the image properties and extract the parameters needed for reflectance scaling, then create true and false color composite images.

1. Set up

library(terra)
library(tmap)
library(tidyverse)
library(kableExtra)

2. Data

We’ll work with a Landsat 8 OLI image collected over Santa Barbara, CA on October 15, 2026. Landsat 8 is part of the longest-running Earth observation program, operated jointly by NASA and USGS since 1972.

Downloading the data

We’ll download data directly from USGS Earth Explorer, the primary portal for Landsat imagery.

Step 1: Create a free USGS account

Go to earthexplorer.usgs.gov and click Login → Create New Account. You’ll need to verify your email before you can download data.

Step 2: Define your search area

Navigate to the Search Criteria tab. In the Address/Place search box, type Santa Barbara, CA and click Show.

Step 3: Select the California option (first row)

Step 4: Select a date range of 10/01/2026 to 11/01/2026

Step 5: Select Datasets

Click Landsat > Landsat Collection 2 Level-2 > Landsat 8-9 OLI/TIRS C2 L2. Then click Results.

Step 6: Download

After selecting the download arrow, click “Select Files” for the Level-2 Surface Reflectance Bands (13 files) option. Donwload the .tif files for bands 2-7, as well as the metadata (MTL.xml)

WarningFile sizes

You should have the following files:

LC08_L2SP_042036_20261015_20261021_02_T1_SR_B2.TIF   ← Blue
LC08_L2SP_042036_20261015_20261021_02_T1_SR_B3.TIF   ← Green
LC08_L2SP_042036_20261015_20261021_02_T1_SR_B4.TIF   ← Red
LC08_L2SP_042036_20261015_20261021_02_T1_SR_B5.TIF   ← Near Infrared
LC08_L2SP_042036_20261015_20261021_02_T1_SR_B6.TIF   ← SWIR 1
LC08_L2SP_042036_20261015_20261021_02_T1_SR_B7.TIF   ← SWIR 2
LC08_L2SP_042036_20261015_20261021_02_T1_MTL.txt     ← Metadata

Where LC08 = Landsat 8, 042036 = WRS-2 path/row, and the first date is the acquisition date.

3. Metadata exploration

USGS distributes an _MTL.txt file alongside every Landsat scene. Open the file directly in a text editor and use the guide below to find key fields. You’ll answer questions about what you find before moving on to the R analysis.

The MTL file is organized into named GROUP sections. The two most important for this lab are:

  • IMAGE_ATTRIBUTES — when and where the image was collected, cloud cover, and sun geometry
  • LEVEL2_SURFACE_REFLECTANCE_PARAMETERS — the scaling parameters needed to convert raw DN values to surface reflectance

Metadata field guide

IMAGE_ATTRIBUTES

Field What it means
SPACECRAFT_ID Which satellite collected the image
SENSOR_ID Which sensor (OLI_TIRS = Operational Land Imager + Thermal Infrared Sensor)
DATE_ACQUIRED Date the image was captured
SCENE_CENTER_TIME Time of acquisition in UTC
WRS_PATH / WRS_ROW WRS-2 grid location — the global addressing system for Landsat scenes. Path = descending orbit track, Row = latitude band
CLOUD_COVER_LAND Percentage of the land area covered by clouds
IMAGE_QUALITY_OLI Sensor quality score from 0 (worst) to 9 (best)
SUN_ELEVATION Angle of the sun above the horizon at acquisition (degrees). Higher = summer/tropics, lower = winter/poles. Affects shadow length and overall image brightness
SUN_AZIMUTH Compass direction of the sun at acquisition (degrees clockwise from north). Determines which side of objects shadows fall on
EARTH_SUN_DISTANCE Distance from Earth to Sun in astronomical units (AU) at time of acquisition. Earth is closest (~0.983 AU) in early January and farthest (~1.017 AU) in early July
SATURATION_BAND_X Whether any pixels in that band hit the sensor’s maximum recordable value. Y = at least one pixel was brighter than the sensor could record. Its DN is clipped to 65535 (max value for 16 bit) regardless of the true reflectance, so the actual value is unknown. This typically happens over very bright surfaces (snow, clouds, sun glint on water).

LEVEL2_SURFACE_REFLECTANCE_PARAMETERS

Field What it means
REFLECTANCE_MULT_BAND_X Scale factor — multiply raw DN by this value
REFLECTANCE_ADD_BAND_X Additive offset — add this after multiplying
QUANTIZE_CAL_MAX/MIN_BAND_X Valid DN range (1–65535; a value of 0 means no data)
REFLECTANCE_MAXIMUM/MINIMUM_BAND_X Valid reflectance range after scaling (slightly negative values can occur from atmospheric correction; values slightly above 1 are physically possible for some surfaces)

Band specifications (from USGS Landsat sensor documentation, not the MTL file)

Band Wavelength Useful for
Band 2 (blue) 0.45–0.51 µm Water depth, coastal features, urban areas
Band 3 (green) 0.53–0.59 µm Vegetation health (chlorophyll), sediment in water
Band 4 (red) 0.64–0.67 µm Iron oxides in soil; healthy vegetation absorbs red
Band 5 (NIR) 0.85–0.88 µm Vegetation biomass; water absorbs NIR strongly
Band 6 (SWIR1) 1.57–1.65 µm Soil/vegetation moisture, fire detection
Band 7 (SWIR2) 2.11–2.29 µm Mineral mapping, burned area detection
WarningQuestions

Open the _MTL.txt file in a text editor and use the guide above to answer the following:

  1. Which satellite and sensor collected this image?
  2. What date and time (UTC) was the image acquired?
  3. The EARTH_SUN_DISTANCE is ~0.997 AU. Is Earth closer to or farther from the Sun than average on this date? What does this mean for the amount of incoming solar radiation?
  4. Write out the complete scaling equation for Band 4 (Red) using the values you find in LEVEL2_SURFACE_REFLECTANCE_PARAMETERS.
  5. What is the CLOUD_COVER_LAND percentage? Would you consider this scene suitable for analysis? What cloud cover threshold would you set when searching for usable Landsat scenes, and why?
  6. Check SATURATION_BAND_4 and SATURATION_BAND_5. If a band is marked Y, what does that mean for the pixel values in that band? How could saturation affect a spectral index like NDVI that uses both bands?

Loading the data

data_dir <- here::here("course-materials/labs/data/landsat/")


band_files <- list.files(data_dir,
                         pattern = "\\.TIF$",
                         full.names = TRUE,
                         ignore.case = TRUE) |>
  sort()

landsat <- terra::rast(band_files)
names(landsat) <- c("blue", "green", "red", "NIR", "SWIR1", "SWIR2")

Now set the scaling parameters you found in the MTL file. We’ll use these in Section 4.

scale_factor <- 2.75e-05
add_offset   <- -0.2

Raster properties

Printing the raster object gives a summary of the spatial properties. Compare what you see here to the four image resolutions from lecture.

landsat
class       : SpatRaster 
size        : 7961, 7831, 6  (nrow, ncol, nlyr)
resolution  : 30, 30  (x, y)
extent      : 121185, 356115, 3713685, 3952515  (xmin, xmax, ymin, ymax)
coord. ref. : WGS 84 / UTM zone 11N (EPSG:32611) 
sources     : LC08_L2SP_042036_20261015_20261021_02_T1_SR_B2.TIF  
              LC08_L2SP_042036_20261015_20261021_02_T1_SR_B3.TIF  
              LC08_L2SP_042036_20261015_20261021_02_T1_SR_B4.TIF  
              ... and 3 more sources
names       : blue, green, red, NIR, SWIR1, SWIR2 
# confirm data type — determines radiometric resolution
datatype(landsat)
[1] "INT2U" "INT2U" "INT2U" "INT2U" "INT2U" "INT2U"
NoteConnecting raster properties to image resolutions
Resolution What it describes Where to find it
Spatial Pixel size on the ground Raster summary
Spectral Number of bands Raster summary
Radiometric Possible values per pixel datatype() — INT2U = 16-bit = 65,536 levels
Temporal Revisit time Landsat Mission Info
NoteBit depth in Landsat Collection 2

Landsat Collection 2 Level-2 data is stored as 16-bit unsigned integers (INT2U): 2^16 = 65,536 possible values per pixel. This is higher radiometric resolution than older 8-bit Landsat products, allowing finer discrimination of subtle reflectance differences.

Visualizing individual bands

Each band responds to a different portion of the electromagnetic spectrum. Before combining bands into composites, let’s visualize them individually as grayscale images where bright = high pixel value, dark = low pixel value.

Code 
plot(landsat, col = gray.colors(256))

WarningExplore the bands
  • In which bands does water appear darkest? Why?
  • In which band does vegetation appear brightest? Why?

4. Reflectance scaling

The downloaded TIF files contain scaled integers, not raw sensor output. Landsat Collection 2 Level-2 products are already atmospherically corrected to surface reflectance, but the values are stored as integers to reduce file size. To recover actual reflectance values (0–1), we apply the linear equation with the scale factor and additive offset values we found in the metadata:

\[\text{Reflectance} = \text{integer} \times \text{scale factor} + \text{additive offset}\]

landsat_scaled <- (landsat * scale_factor) + add_offset

|---------|---------|---------|---------|
=========================================
                                          

|---------|---------|---------|---------|
=========================================
                                          
landsat_scaled <- clamp(landsat_scaled, lower = 0, upper = 1)

|---------|---------|---------|---------|
=========================================
                                          
# verify: values should now be approximately 0–1
minmax(landsat_scaled)
       blue     green     red NIR SWIR1 SWIR2
min 0.00000 0.0000000 0.00000   0     0     0
max 0.74501 0.7985525 0.84093   1     1     1

5. True color imagery

A true color composite maps the red, green, and blue bands to the red, green, and blue display channels, approximating how the scene looks to the human eye.

Display channel Satellite band Wavelength
Red Band 4 (red) 0.64–0.67 µm
Green Band 3 (green) 0.53–0.59 µm
Blue Band 2 (blue) 0.45–0.51 µm 
tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("red", "green", "blue"), multivariate = TRUE),
                            col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("True color composite") +
  tm_scalebar(position = c("left", "bottom"))
SpatRaster object downsampled to 3189 by 3137 cells.

6. False color imagery

A false color composite assigns satellite bands beyond the visible spectrum (or in a non-standard order) to the display channels, letting us highlight features that are difficult to distinguish in true color.

Vegetation false color (NIR, Red, Green)

Placing the near infrared band in the red channel makes healthy vegetation appear vivid red — useful for assessing vegetation extent and health.

Display channel Satellite band Why
Red Band 5 (NIR) Vegetation reflects strongly in NIR → appears red
Green Band 4 (Red) Soil reflects red → appears green
Blue Band 3 (Green) Water reflects slightly in green → appears blue 
tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("NIR", "red", "green"), multivariate = TRUE),
         col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("False color: NIR–Red–Green") +
  tm_scalebar(position = c("left", "bottom"))
SpatRaster object downsampled to 3189 by 3137 cells.

SWIR false color (SWIR1, NIR, Green)

Placing the shortwave infrared band in the red channel highlights moisture. Water absorbs both SWIR and NIR strongly, so water bodies appear very dark.

Display channel Satellite band Why
Red Band 6 (SWIR1) Dry/burned surfaces reflect strongly → bright
Green Band 5 (NIR) Vegetated areas reflect NIR → appears green
Blue Band 3 (Green) Water absorbs all three → very dark 
tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("SWIR1", "NIR", "green"), multivariate = TRUE),
         col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("False color: SWIR1–NIR–Green") +
  tm_scalebar(position = c("left", "bottom"))
SpatRaster object downsampled to 3189 by 3137 cells.

Comparing all three composites

Code 
map_true <- tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("red", "green", "blue"), multivariate = TRUE),
         col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("True color")

map_nir <- tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("NIR", "red", "green"), multivariate = TRUE),
         col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("NIR false color")

map_swir <- tm_shape(landsat_scaled) +
  tm_rgb(col = tm_vars(c("SWIR1", "NIR", "green"), multivariate = TRUE),
         col.scale = tm_scale_rgb(max_color_value = 1)) +
  tm_title("SWIR false color")

tmap_arrange(map_true, map_nir, map_swir, nrow = 1)
SpatRaster object downsampled to 3189 by 3137 cells.
SpatRaster object downsampled to 3189 by 3137 cells.
SpatRaster object downsampled to 3189 by 3137 cells.

WarningInterpret the composites
  1. What land cover types are clearest in the true color image?
  2. Where do you see evidence of vegetation in the NIR false color image? Does its brightness match the Band 5 (NIR) grayscale image?
  3. How does the SWIR composite help distinguish open water from other dark features such as shadows or urban areas?