pixltsnorm

Pixel-based Linear Time Series Normalizer

pixltsnorm is a small, focused Python library that:

• Bridges or harmonizes numeric time-series data (e.g., reflectance, NDVI, etc.) across multiple sensors or sources.
• Fits simple linear transformations (y = slope*x + intercept) to map one sensor’s scale onto another.
• Chains transformations to handle indirect overlaps (sensor0 → sensor1 → …).
• Filters outliers using threshold-based filtering before fitting linear models.

Although originally inspired by NDVI normalization across different Landsat sensors, pixltsnorm is domain-agnostic. You can use it for any numeric time-series that needs linear alignment.

---

Features

1. Outlier Filtering

   • Removes large disagreements in overlapping time-series pairs, based on a simple threshold for |A - B|.

2. Local (Pixel-Level) Linear Bridging

   • Regress one sensor’s measurements onto another’s (e.g., a single pixel’s time series).
   • Produces an easy-to-apply transform function for new data.

3. Global Bridging

   • Follows the approach of Roy et al. (2016): gather all overlapping values across the entire dataset, fit one “universal” slope/intercept.
   • Useful if you need scene-wide or region-wide continuity between two or more sensors (e.g., L5 → L7 → L8).

4. Chaining

   • Allows any number of sensors to be combined in sequence, producing a single transform from the first sensor to the last.

5. Lightweight

   • Minimal dependencies: numpy, scikit-learn, and optionally pandas.

6. Earth Engine Submodule

   • A dedicated earth_engine subpackage provides GEE-specific helpers (e.g., for Landsat) that you can incorporate in your Earth Engine workflows.

---

Basic Usage

Harmonize Two Sensors (Pixel-Level Example)

import numpy as np
from pixltsnorm.harmonize import harmonize_series

# Suppose sensorA_values and sensorB_values have overlapping data
sensorA = np.array([0.0, 0.2, 0.8, 0.9])
sensorB = np.array([0.1, 0.25, 0.7, 1.0])

results = harmonize_series(sensorA, sensorB, outlier_threshold=0.2)
print("Slope:", results['coef'])
print("Intercept:", results['intercept'])

# Transform new data from sensorA scale -> sensorB scale
transform_func = results['transform']
new_data = np.array([0.3, 0.4, 0.5])
mapped = transform_func(new_data)
print("Mapped values:", mapped)

Chaining Multiple Sensors

from pixltsnorm.harmonize import chain_harmonization
import numpy as np

# Suppose we have 4 different sensors that partially overlap:
sensor0 = np.random.rand(10)
sensor1 = np.random.rand(10)
sensor2 = np.random.rand(10)
sensor3 = np.random.rand(10)

chain_result = chain_harmonization([sensor0, sensor1, sensor2, sensor3])
print("Pairwise transforms:", chain_result['pairwise'])
print("Overall slope (sensor0->sensor3):", chain_result['final_slope'])
print("Overall intercept (sensor0->sensor3):", chain_result['final_intercept'])

# Apply sensor0 -> sensor3 transform
sensor0_on_sensor3_scale = (chain_result['final_slope'] * sensor0
                            + chain_result['final_intercept'])
print("sensor0 mapped onto sensor3 scale:", sensor0_on_sensor3_scale)

Global Bridging

import pandas as pd
from pixltsnorm.global_harmonize import chain_global_bridging

# Suppose we have three DataFrames: df_l5, df_l7, df_l8
# Each has row=pixels, columns=dates (plus 'lon','lat').
# The approach merges all overlapping values across the region/time:
result = chain_global_bridging(df_l5, df_l7, df_l8, outlier_thresholds=(0.2, 0.2))

# We get a single slope/intercept for L5->L7, L7->L8, plus the chain L5->L8
print("Global bridging L5->L7 =>", result["L5->L7"]["coef"], result["L5->L7"]["intercept"])
print("Global bridging L7->L8 =>", result["L7->L8"]["coef"], result["L7->L8"]["intercept"])
print("Chained L5->L8 =>", result["L5->L8"]["coef"], result["L5->L8"]["intercept"])

Earth Engine Submodule

from pixltsnorm.earth_engine import create_reduce_region_function, addNDVI, cloudMaskL457

# Use these GEE-based helpers inside your Earth Engine scripts

Please see the docs and example notebooks for more examples.

---

Installation

1. Clone or download this repository.
2. (Optional) Create and activate a virtual environment.
3. Install in editable mode:

pip install -e .

Then you can do:

import pixltsnorm

and access the library’s functionality.

---

Acknowledgements

• Joseph Emile Honour Percival performed the initial research in 2021 during his PhD at Kyoto University, where the pixel-level time-series normalization idea was first applied to multi-sensor analysis.
• The global bridging logic is inspired by Roy et al. (2016), which outlines regression-based continuity for Landsat sensors across large areas.

Roy, David P., V. Kovalskyy, H. K. Zhang, Eric F. Vermote, L. Yan, S. S. Kumar, and A. Egorov. "Characterization of Landsat-7 to Landsat-8 reflective wavelength and normalized difference vegetation index continuity." Remote sensing of Environment 185 (2016): 57-70.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.