Image scaling

Uncomment the following line to install the latest version of geeagri if needed.

In [ ]:

# !pip install -U geeagri

# !pip install -U geeagri

Import libraries¶

In [ ]:

import ee
import geemap
from geeagri.preprocessing import (
    MeanCentering,
    StandardScaler,
    MinMaxScaler,
    RobustScaler,
)

import ee import geemap from geeagri.preprocessing import ( MeanCentering, StandardScaler, MinMaxScaler, RobustScaler, )

Initialize a Map object¶

Authenticate and initialize Earth Engine. If it doesn't work, specify a project name

In [ ]:

# ee.Authenticate()
# ee.Initialize(project='your-project-id')

Map = geemap.Map()
Map

# ee.Authenticate() # ee.Initialize(project='your-project-id') Map = geemap.Map() Map

Import region of interest¶

In [ ]:

bbox = [-120.398369, 37.006574, -120.208168, 37.101924]
region = ee.Geometry.BBox(*bbox)
region_style = {"color": "red", "width": 1}
Map.addLayer(region, region_style, "Region")
Map.centerObject(region, 12)

bbox = [-120.398369, 37.006574, -120.208168, 37.101924] region = ee.Geometry.BBox(*bbox) region_style = {"color": "red", "width": 1} Map.addLayer(region, region_style, "Region") Map.centerObject(region, 12)

Load Sentinel-2 data and create a median composite¶

In [ ]:

image = (
    ee.ImageCollection("COPERNICUS/S2_SR_HARMONIZED")
    .filterBounds(region)
    .filterDate("2024-01-01", "2025-01-01")
    .filterMetadata("CLOUDY_PIXEL_PERCENTAGE", "less_than", 10)
    .select(["B8", "B4", "B3"])
    .median()
    .multiply(0.0001)
    .clip(region)
)

image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 0.3}

Map.addLayer(image, image_vis, "Sentinel 2 FCC")

image = ( ee.ImageCollection("COPERNICUS/S2_SR_HARMONIZED") .filterBounds(region) .filterDate("2024-01-01", "2025-01-01") .filterMetadata("CLOUDY_PIXEL_PERCENTAGE", "less_than", 10) .select(["B8", "B4", "B3"]) .median() .multiply(0.0001) .clip(region) ) image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 0.3} Map.addLayer(image, image_vis, "Sentinel 2 FCC")

MeanCentering¶

Subtracts the mean from each band to center values around zero:

$$ X_\text{centered} = X - \mu $$

Where:

• $X$: original pixel value
• $\mu$: mean of the band computed over the given region

In [ ]:

scaler = MeanCentering(image, region)
centered_image = scaler.transform()

# Adjust visualization range around 0 to reflect centered values
centered_image_vis = {"bands": ["B8", "B4", "B3"], "min": -0.09, "max": 0.09}
Map.addLayer(centered_image, centered_image_vis, "MeanCentering")

scaler = MeanCentering(image, region) centered_image = scaler.transform() # Adjust visualization range around 0 to reflect centered values centered_image_vis = {"bands": ["B8", "B4", "B3"], "min": -0.09, "max": 0.09} Map.addLayer(centered_image, centered_image_vis, "MeanCentering")

MinMaxScaler¶

Normalizes each band using the min and max values over the region:

$$ X_\text{scaled} = \frac{X - \min}{\max - \min}, \quad X_\text{scaled} \in [0, 1] $$

Where:

• $\min$, $\max$: band-wise minimum and maximum values over the region.

In [ ]:

scaler = MinMaxScaler(image, region)
scaled_image = scaler.transform()

scaled_image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 1}
Map.addLayer(scaled_image, scaled_image_vis, "MinMaxScaler")

scaler = MinMaxScaler(image, region) scaled_image = scaler.transform() scaled_image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 1} Map.addLayer(scaled_image, scaled_image_vis, "MinMaxScaler")

StandardScaler¶

Standardizes each band using z-score normalization (mean and standard deviation):

$$ X_\text{standardized} = \frac{X - \mu}{\sigma} $$

Where:

• $X$: original pixel value
• $\mu$: mean of the band over the specified region
• $\sigma$: standard deviation of the band over the specified region

In [ ]:

scaler = StandardScaler(image, region)
standardized_image = scaler.transform()

# You may want to visualize within a reasonable range
standardized_image_vis = {"bands": ["B8", "B4", "B3"], "min": -2, "max": 2}
Map.addLayer(standardized_image, standardized_image_vis, "StandardScaler")

scaler = StandardScaler(image, region) standardized_image = scaler.transform() # You may want to visualize within a reasonable range standardized_image_vis = {"bands": ["B8", "B4", "B3"], "min": -2, "max": 2} Map.addLayer(standardized_image, standardized_image_vis, "StandardScaler")

RobustScaler¶

Scales each band using robust percentiles (e.g., 1st and 99th) to reduce the influence of outliers:

$$ X_\text{scaled} = \frac{X - P_\text{lower}}{P_\text{upper} - P_\text{lower}}, \quad X_\text{scaled} \in [0, 1] $$

Where:

• $X$: original pixel value
• $P_{\\text{lower}}$: lower percentile value (e.g., 25th percentile)
• $P_{\\text{upper}}$: upper percentile value (e.g., 75th percentile)

In [ ]:

scaler = RobustScaler(image, region, lower=1, upper=99)
scaled_image = scaler.transform()

scaled_image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 1}
Map.addLayer(scaled_image, scaled_image_vis, "RobustScaler")

scaler = RobustScaler(image, region, lower=1, upper=99) scaled_image = scaler.transform() scaled_image_vis = {"bands": ["B8", "B4", "B3"], "min": 0, "max": 1} Map.addLayer(scaled_image, scaled_image_vis, "RobustScaler")

💡 Tips for Efficient Processing

• Increase the scale parameter to reduce computation time. Calculating statistics (e.g., mean, percentiles) over high-resolution images (like Sentinel-2 at 10 m) can be time-consuming or cause timeouts. Try scale=100 or higher when possible.

• Use simpler geometries. Avoid complex shapes like administrative boundaries or large regions. Instead, use simplified or bounding box geometries.

• You can simplify any geometry by converting it to a bounding rectangle using:

  simple_region = complex_region.bounds()
  

  This can drastically speed up reduceRegion operations while still capturing spatial context.