Machine Vision and Applications (2012) 23:937–952 DOI 10.1007/s00138-011-0375-3 ORIGINAL PAPER Soil-moisture estimation from TerraSAR-X data using neural networks Matej Kseneman · Dušan Gleich · Božidar Potoˇ cnik Received: 4 October 2010 / Revised: 5 July 2011 / Accepted: 22 September 2011 / Published online: 11 October 2011 © Springer-Verlag 2011 Abstract Early prediction of natural disasters like floods and landslides is essential for reasons of public safety. This can be attained by processing Synthetic-Aperture Radar (SAR) images and retrieving soil-moisture parameters. In this article, TerraSAR-X product images are investigated in combination with a water-cloud model based on the Shi semi- empirical model to determine the accuracy of soil-moisture parameter retrieval. SAR images were captured between Jan- uary 2008 and September 2010 in the vicinity of the city Maribor, Slovenia, at different incidence angles. The water- cloud model provides acceptable estimated soil-moisture parameters at bare or scarcely vegetated soil areas. How- ever, this model is too sensitive to speckle noise; therefore, a pre-processing step for speckle-noise reduction is carried out. Afterwards, self-organizing neural networks (SOM) are used to segment the areas at which the performance of this model is poor, and at the same time neural networks are also used for a more accurate approximation of model parame- ters’ values. Ground-truth is measured using the Pico64 sen- sor located on the field, simultaneously with capturing SAR images, in order to enable the comparison and validation of the obtained results. Experimental results show that the pro- posed method outperforms the water-cloud model accuracy over all incidence angles. M. Kseneman (B ) · D. Gleich · B. Potoˇ cnik Faculty of Electrical Engineering and Computer Science, University of Maribor, Maribor, Slovenia e-mail: matej.kseneman@gmail.com D. Gleich e-mail: dusan.gleich@uni-mb.si B. Potoˇ cnik e-mail: bozo.potocnik@uni-mb.si Keywords Model-based de-speckling · Soil-moisture estimation · Self-organizing maps · Feed-forward back-propagation neural network · TerraSAR-X 1 Introduction The soil water content, also known as soil-moisture, plays an important role in estimation, classification and model- ling of various large-scale ecological and climate processes, such as agriculture, flood forecasting, climate change, surface run-off, ground-water replenishment etc. Several different approaches for soil-moisture parameter retrieval, based on the active and passive microwave radiation, had been devel- oped over the past few decades. Research in this field is still undergoing [23, 1]; new technologies (e.g. better SAR satel- lite image resolution, and satellite constellations bringing more polarized images gathered at the exact time) are being developed, and new moisture-estimation models are arising. However, one of the greatest challenges is still to find a model that would work accurately on all underlying surfaces. Many studies have shown that microwave radiation data, although they depend on various radar parameters, such as frequency, incidence angle and polarization, are also corre- lated with the terrain slope, dielectric properties related to soil-moisture [33, 9, 17], and surface roughness. Soil-mois- ture estimation techniques have been very popular in the past few years [25], mainly because of resolution up to 1 m and vast surface coverage [2]. Several different empirical models have been used for esti- mation of soil-moisture based on SAR data [32, 27, 11, 29], meanwhile the TerraSAR-X relationship is investigated in [3, 5]. However, the Shi empirical model [29] proved to be the most efficient. This model is based on theoreti- cal foundations, mainly on the integral equation model (IEM) [16], and is extended and modified to accommodate 123