Topographic Normalization in Google Earth Engine: A Multi-Criteria Evaluation of Four Correction Methods Applied to Landsat 8 over Pico de Orizaba, Mexico
2026-11-12 –, Sala Geomática 7

Keywords: topographic correction, Landsat 8, Google Earth Engine, Pico de Orizaba, Gamma method, remote sensing, multi-criteria analysis

1. Introduction

In mountainous regions, topographic relief introduces systematic illumination biases in satellite-derived surface reflectance, independent of surface material properties, a phenomenon known as the topographic effect. This compromises land cover classification, vegetation monitoring, and change detection. Landsat 8 OLI is particularly susceptible in rugged terrain like the Trans-Mexican Volcanic Belt. While models like Cosine, C-Correction, Modified Minnaert, and Gamma exist, rigorous comparative evaluations using quantitative metrics within cloud platforms like Google Earth Engine (GEE) remain scarce. This study addresses this gap by implementing and comparing four algorithms over Pico de Orizaba, demonstrating that visual inspection alone is insufficient, and proposing a replicable, cloud-based evaluation framework.

2. Objectives

(1) Implement four algorithms (Cosine (Lambertian), C-Correction, Modified Minnaert, and Gamma) on a Landsat 8 scene over Pico de Orizaba using GEE.
(2) Evaluate performance using Standard Deviation (SD) and Coefficient of Variation (CV) on Band 5 (NIR, ~860 nm), selected for its sensitivity to vegetation and susceptibility to topographic effects.
(3) Identify the most effective method for reducing terrain-induced spectral heterogeneity in high-relief volcanic terrain and discuss implications for operational workflows.

3. Study Area

Pico de Orizaba (Citlaltépetl, 5,636 m) straddles Veracruz and Puebla. Its extreme elevation gradient (1,200 m to 5,636 m), complex aspect distribution, frequent deep shadowing, and diverse land cover (glaciers, subalpine grasslands, conifer forests, agriculture) make it an ideal benchmark site for evaluating topographic correction.

4. Data and Methods

A 2023 Landsat 8 OLI Collection 2 TOA scene (<5% cloud cover) was processed in GEE. Terrain parameters (slope, aspect) were derived from the SRTM DEM (30 m). Solar azimuth and elevation were extracted from image metadata.

The cosine of the local solar incidence angle (( \cos i )) was derived as:
[
\cos i = \cos \theta_z \cos \theta_s + \sin \theta_z \sin \theta_s \cos(\phi_a - \phi_s)
]
where ( \theta_z ) is the solar zenith angle, ( \theta_s ) is terrain slope, ( \phi_a ) is solar azimuth, and ( \phi_s ) is terrain aspect. Pixels with ( \cos i \leq 0.01 ) were masked to avoid over-correction in deep shadows.

The four algorithms were implemented as follows:

  • Cosine (Lambertian): Assumes isotropic reflection.
    [
    \rho_c = \rho \times \frac{\cos \theta_z}{\cos i}
    ]

  • C-Correction: Introduces an empirical factor ( c = a/b ) (intercept/slope from a linear regression between ( \cos i ) and band reflectance) to moderate over-correction.
    [
    \rho_c = \rho \times \frac{\cos \theta_z + c}{\cos i + c}
    ]

  • Modified Minnaert: Applies a constant ( k = 0.5 ) to model non-Lambertian behavior.
    [
    \rho_c = \rho \times \left( \frac{\cos \theta_z}{\cos i} \right)^k
    ]

  • Gamma: Weights direct and diffuse irradiance components for robustness in shadowed terrain.
    [
    \rho_c = \rho \times \frac{2\cos \theta_z}{\cos \theta_z + \cos i}
    ]

Statistical evaluation was performed on Band 5 (NIR) across the full scene at a 300 m analysis scale, computing mean (( \mu )), standard deviation (( \sigma )), and the Coefficient of Variation:
[
CV = \frac{\sigma}{\mu} \times 100\%
]

5. Results and Discussion

Quantitative evaluation yielded the following CV results for Band 5 NIR:

MethodCV (%)Change vs. Original (pp)
Original (uncorrected)40.96%–
Cosine (Lambertian)45.15%+4.19 (worse)
C-Correction41.05%+0.09
Modified Minnaert (k=0.5)40.87%-0.09
Gamma40.64%-0.32 (best)

The Cosine method paradoxically increased spectral heterogeneity, worsening CV by more than four percentage points. This aligns with literature showing that the Lambertian assumption fails in steep, high-aspect-variability terrain with extensive shadows, leading to severe over-correction on sun-facing slopes and under-correction in shadow zones. Critically, this proves that visual improvement does not necessarily translate into quantitative improvement, a key methodological finding for validation practices.

The Gamma method achieved the best quantitative performance, followed closely by Modified Minnaert and C-Correction. All three outperformed Cosine by partially compensating for diffuse irradiance in shadowed terrain that is the key physical mechanism absent from the Lambertian formulation. The marginal differences among Gamma, Minnaert, and C-Correction suggest that final method selection may also depend on secondary factors such as computational cost or the availability of auxiliary data (e.g., surface roughness for Minnaert optimization).

6. Conclusions

This study demonstrates that visual inspection alone is insufficient for validating topographic correction; quantitative metrics like the Coefficient of Variation are essential for objective algorithm discrimination. The Gamma method proved most effective for reducing terrain-induced spectral heterogeneity over Pico de Orizaba, while the Cosine (Lambertian) method actively degraded data quality, which is not recommended to be used in a high-relief settings.

The full GEE implementation ensures reproducible, scalable processing without local infrastructure requirements, making this workflow directly transferable to environmental monitoring, vegetation mapping, and change detection applications across Mexico's volcanic highlands. Future work will explore band-by-band optimization of correction parameters and integration with land cover classification pipelines.


AI Use Disclosure: This extended abstract was prepared with the assistance of Claude (Anthropic), an AI tool used for text editing and translation purposes, in compliance with iGISc 2026 submission guidelines.

Raúl Enrique Mora González es maestrando en Ciencias de la Información Geoespacial con orientación en Observación de la Tierra en el Centro de Investigación en Ciencias de Información Geoespacial (CentroGeo), en la Ciudad de México. Es licenciado en Geohistoria por la Escuela Nacional de Estudios Superiores Morelia de la Universidad Nacional Autónoma de México (UNAM), formación que le proporcionó una base sólida para el análisis del espacio y el tiempo como dimensiones constitutivas y complementarias del territorio.
Sus intereses y competencias abarcan el procesamiento digital de imágenes satelitales (Sentinel-1 SAR, Sentinel-2, Landsat 8/9), el análisis espacial y los Sistemas de Información Geográfica (SIG), el desarrollo de aplicaciones web geoespaciales, y la aplicación de métodos cuantitativos a problemas ambientales y territoriales.

This speaker also appears in: