A model-based neural network methodology to estimate water and ice contents in planetary soils using neutron fluxes detected by in-situ and/or airborne deployment of neutron detectors is proposed and shown to be effective. Focusing of epithermal and thermal energy regimes, the neutron fluxes are computed [1] as function of the medium physical properties and used to train neural networks in the inverse mode. For an homogeneous soil the model-based neural network shows satisfactory performances in retrieving the percentage of water. For a soil modelled as layered, neural networks designed to retrieve both depth and thickness of an ice layer beneath the soil surface provides good results only in a limited range of configurations. However, it has been found that training two networks to independently retrieve the two parameters results in more accurate results. It has been also found that multiple measurements help improve the accuracy of the inversion for this configuration.
A. Luciani, P. Panfili, R. Furfaro, B.D. Ganapol, D. Mostacci (2009). Estimating water and ice content on planetary soils using neutron measurements: a neural network approach. RADIATION EFFECTS AND DEFECTS IN SOLIDS, 164, 345-349 [10.1080/10420150902811656].
Estimating water and ice content on planetary soils using neutron measurements: a neural network approach
MOSTACCI, DOMIZIANO
2009
Abstract
A model-based neural network methodology to estimate water and ice contents in planetary soils using neutron fluxes detected by in-situ and/or airborne deployment of neutron detectors is proposed and shown to be effective. Focusing of epithermal and thermal energy regimes, the neutron fluxes are computed [1] as function of the medium physical properties and used to train neural networks in the inverse mode. For an homogeneous soil the model-based neural network shows satisfactory performances in retrieving the percentage of water. For a soil modelled as layered, neural networks designed to retrieve both depth and thickness of an ice layer beneath the soil surface provides good results only in a limited range of configurations. However, it has been found that training two networks to independently retrieve the two parameters results in more accurate results. It has been also found that multiple measurements help improve the accuracy of the inversion for this configuration.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.