Land cover monitoring by fractal analysis of digital images

article i nfo Quantitative information about the land cover is a fundamental issue for any soil management practice. The use of high-resolution remote sensing for this purpose is still questionable because of some poorly measurable or unknown parameters involved in the interaction between the electromagnetic waves and the soil surface. The recorded heterogeneous, highly variable and multi-temporal numerical databases require new analytical tools for signal's pattern recognition and space/time interpretation. In this research we use a novel and versatile low-cost technique (Fractal Image Informatics) for indirect measurement of the soil cover by in situ digital imagery and fractal interpretation. The roughness of gray level distribution across the images is measured in terms of Hurst exponent (H) computed at the global ("firmagram") and local ("reference lines") scales and correlated with some direct physical measurements of the residues' weight and the degree of surface covering. The comparative analysis of six contrasting tillage systems was carried out on Mollic Andosol of Michoacan State, Mexico. The results show that H extracted from the soil surface images reveals the differences in land cover in a statistically significant manner. The image roughness, and therefore its Hurst exponent, had a negative correlation with the total weight of plant residues (R 2 =0.80) as well as with the degree of soil covering (R 2 =0.92). Strong positive correlation (R 2 =0.86) was observed between the local H (extracted from the reference lines of the image) and global H (determined from the firmagram), confirming the scale invariance of the studied Andosol and the fractal nature of its digital images. All results are in qualitative and quantitative agreement with the in situ visualized patterns of residue distribution across the experimental plots. We conclude that Fractal Image Informatics is a precise and low-cost technique suitable to monitor the land cover in real time/space by digital imaging.