Non-Parametric Tools for Soil Moisture Mapping Using Active Microwave Data

Active microwave remote sensing has shown the capability to estimate soil moisture based on the large contrast between the dielectric properties of wet and dry soil. However, soil moisture response to microwave system from ground surface is mostly influenced by parameters such as: land cover, vegetation density, soil texture etc making the retrieval process more complex. In such conditions; non-parametric tools such as: neural networks and/or fuzzy logic systems could have more potential in compared to traditional techniques retrieving soil moisture from microwave sensors.

This study deals with the application of back-propagation neural network and fuzzy logic in estimating the surface soil moisture using Synthetic Aperture Radar (SAR) data. The potential of SAR images in spatial soil moisture estimation depends on the ability of these algorithms to define the complex relationship that exists between the backscattered energy and the moisture content of the soil.

The SAR backscattering values from Radarsat-1 images acquired in ScanSAR Mode were used along with Normalized Difference Vegetation Index (NDVI) and Vegetation Optical Depth (VOD) as inputs to neural network and fuzzy logic algorithms. The soil moisture data measured by ESTAR Instrument (Electronically Scanned Thinned Array Radiometer) during the SGP97 campaign (operated by NASA) were used as truth data in the training and the validation processes. In order to match the resolution of the measured soil moisture, the SAR spatial resolution was degraded to 800 meters using an averaging algorithm.

The performance of neural networks and fuzzy logic algorithms has been investigated by varying several parameters related to their structure and training process. The preliminary results showed that for neural networks, the variations of the number of hidden layers and neurons in each layer have no significant effect on classification accuracy. However, other decision making parameters such as threshold value have been shown to have more influence on the retrieval process.

Concerning, the fuzzy logic algorithm, the preliminary results showed that the cluster radius selection have a significant effect on classification accuracy. Further, the prediction made by neural network is higher than fuzzy logic in several runs of the model, but we found that the prediction made by fuzzy logic is more stable in nature. The additions of VOD and NDVI parameters have shown a significant effect on the final soil moisture classification accuracy for both models.