Index
| HSCaRS |
 Algorithms |
| Neural Networks |
| Marius Schamschula |
| Optics |
| Physics |
| AccessGrid Node |
| Arts & Sciences |
| A & M University |
Neural Networks:
The HSCaRS algorithms group is working in support of the hydrological modeling group. Hydrological models, not unlike meteorological models, need periodic updates from the environment. The problem to be solved is how to perform these updates. One candidate technology is the use of satellite based microwave sensors. These sensors offer poor resolution (when compared to optical and IR sensors). Pixels sizes of 32 x 32 km are to be expected, even in later generation passive sensors.
The goal of one aspect of algorithm development it to "disaggregate" remotely sensed data. Disaggregation is a process, which in our case takes low resolution microwave sensor data, and then adds other information, such as precipitation records and soil properties, to yield high resolution data, in this case soil moisture.
We have developed two simple neural networks: the Linear Disaggregation Network (LinDANet), and the Saturating Linear Disaggregation Network (SDANet). Work is ongoing with mmore advanced neural network algorithms.