Neural Network Ensembles: Combining Multiple Models for Downscaling of Soil Moisture
Soo See Chai1, Kok Luong Goh2

1Soo See Chai, Department of Software Engineering and Computing, Faculty of Computer Science and Information Technology, University of Sarawak Malaysia.
2Kok Luong Goh, International College of Advanced Technology Sarawak

Manuscript received on December 11, 2013. | Revised Manuscript received on December 15, 2013. | Manuscript published on December 25, 2013. | PP: 46-50 | Volume-2 Issue-2, December 2013. | Retrieval Number: B0618122213/2013©BEIESP

Open Access | Ethics and Policies | Cite
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (

Abstract: Soil moisture estimation is important for land surface modeling and climate modeling, with soil moisture being employed as a critical parameter. Although, the derivation of soil moisture from passive microwave remote sensing has been theoretically and practically proven to be possible, its spatial resolution however tends to be coarse-grained, at a range of about 20-40 km. As this does not satisfy the requirements of models using higher resolution grids, it is thus desirable to downscale soil moisture to finer resolutions of between 1 to 5 km. Neural network ensembles are known to be able to effectively improve the overgeneralization that arises from the combination of a set of neural network classifiers with a diverse range of error distributions. In this paper, a neural network ensemble method was explored to downscale soil moisture content from 20km to 2km resolution. The dataset used in this experiment was captured using low resolution L-band passive microwave observations from regional air-borne measurements in the study of Goulburn River catchment in Australia. The results have shown that by using a neural network ensemble, an average accuracy of 2.33% can be obtained for the downscaled soil moisture at a 2km resolution.
Keywords: Downscaling, ensemble neural network, radiometer, soil moisture.