High-Pass Filtering for Thermal Data Sharpening: 2D Matrix-Based Approach for Enhanced Spatial Detail in Raster Data

Abstract

Thermal remote sensing data generally have coarser spatial resolution than panchromatic imagery, limiting the representation of fine-scale spatial variations. This study develops a parameterized High-Pass Filtering (HPF) approach to enhance the spatial detail of Landsat 9 Band 10 thermal data from a 30 m grid to a target 15 m grid using panchromatic information from Band 8. The method applies Gaussian filtering to extract high-frequency spatial detail, followed by normalization and controlled detail injection using three α values (0.25, 0.50, and 1.00) and three σ values (1, 2, and 3), resulting in nine experimental combinations. The sharpened thermal rasters were evaluated using descriptive statistics, RMSE, MAE, Bias, R², Pearson correlation, and B10–PAN spatial correlation. The results show that increasing α produces more spatially focused thermal patterns and strengthens the relationship with PAN, with B10–PAN correlation increasing from approximately 0.336–0.338 before sharpening to 0.6680 at the highest tested combination. However, stronger detail injection also increases thermal variability and deviation from the original thermal values, while the mean remains relatively stable at approximately 23.05°C. Increasing σ primarily affects the filtering characteristics and valid-data coverage, with the number of valid pixels decreasing across the tested σ levels. Overall, the parameterized HPF framework provides a simple approach for investigating thermal spatial-detail enhancement while demonstrating a clear trade-off between spatial association with PAN and consistency with the original thermal data.