The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
2. DATA 
This study uses the National Airborne Field Experiment 2005 
(NAFE’05) data in the Goulbum catchment, south-eastern Aus 
tralia, in November 2005. This month-long field experiment 
provided extensive airborne passive microwave observations 
together with spatially distributed and in-situ ground monitoring 
of soil moisture [9]. The area monitored was a square of ap 
proximately 40x40km divided into two sub-areas namely the 
Merriwa area on the east side of this square and the Krui area to 
the west (Figure 1). 
During the field experiment, a small, two-seater plane from the 
Airborne Research Australia National facility called the Small 
Environmental Research Aircraft (SERA) was flown [10]. This 
aircraft was equipped with the Polarimetric L-band Multibeam 
Radiometer (PLMR) and thermal imager. The PLMR obtains 
data of both H- and V-polarized brightness temperatures (Tb) 
using a single receiver with a polarization switch at incidence 
angles +/-7°, +/-21.5° and +/-38.5° in either across track (push- 
broom) or along track configurations. The aircraft was flown at 
four different altitudes over either Krui or Merriwa area and re 
sulted in four different ground resolutions of: (i) 1km, (ii) 500m, 
(iii) 250m and (iv) 62.5m. 
Near-surface soil moisture data were measured across the 
NAFE’05 study area at a range of spatial scales. During the 
four weeks experiment, the near-surface soil moisture data was 
measured across the eight focus farms (Figure 1) concurrently 
with the aircraft overpasses. At the focus farms, soil moisture 
measurements were taken at many locations within the farm at 
various resolutions: 500m, 250m, 125m, 62.5m, 12.5m and 
6.25m. 
For the purpose of this study, the focus farm Roscommon in the 
Krui area with the characteristics in Table 1 is used. The air 
borne data with ground resolution of 250m is utilized. 
Area(ha) 
Topography 
Landuses 
Soils 
940 
Flat/Gently 
Grazing 
Red basaltic 
rolling 
clays and 
sandy soils 
Table 1. Main Characteristics of the Roscommon focus farm. 
3. BRIEF DESCRIPTION OF BACKPROPAGATION 
NEURAL NETWORK MODEL 
The Artificial Neural Network (ANN) was inspired by investi 
gations into the structure of the human brain that consists of in 
terconnected neurons. An ANN is made up of interconnecting 
artificial neurons within input, hidden and output layers. It has 
two modes of operation: training mode and operation/testing 
mode. In the training mode, neurons are trained using a par 
ticular input pattern to produce the desired output pattern. In the 
operation/testing mode, when a taught input pattern is detected 
at the input, the ANN will produce its associated output. A 
Backpropagation or feed-forward backpropagation ANN con 
sists of two processing parts within its neurons: forward and 
backward. When an input pattern is fed to the ANN during its 
training process, the ANN will try to learn and compare its pre 
dicted output value with the desired output value. The errors 
between the predicted and actual valued are then "backpropa- 
gated" through the network, and a gradient descent algorithm 
used to adjust the weights in the hidden and output layer nodes. 
The result is a network that produces the mapping between the 
input values and the output values via the neurons. 
Figure 1. The NAFE’05 study area in the 
Goulbum catchment. The focus farms within the two sub-areas 
of this catchment are also shown. 
62.5m footprint sizes. The area was entirely mapped at a par 
ticular altitude before descending to subsequently lower alti 
tudes. 
For the purpose of this study, the two-layer backpropagation 
ANN is being used. The architecture of this backpropagation 
NN is two inputs at the input layer, 4 neurons in the one hidden 
layer and one output at the output layer. The input of the back- 
propagation NN model is the H- and V-polarized brightness 
temperature value. The output of the NN model is the volumet 
ric soil moisture data. The ANN is trained to generate a map 
ping between the continuous input brightness temperature val 
ues and the continuous output value of soil moisture data. 
4. GROUND DATA 
The soil moisture within the top 5cm of the soil profile was 
monitored coincident with each aircraft flight either across the 
entire area or across the focus farm. Measurements of the top 
5cm soil moisture content were undertaken using an innovative 
Hydraprobe Data Acquisition System developed by The Uni 
versity of Melbourne that integrates a Global Positioning Sys 
tem and soil moisture sensor with a Geographic Information 
System [10]. During the sampling of the focus farms, very high 
resolution sampling was concentrated on a 150x150m area 
where soil moisture was measured at 12.5m (outer section) and 
6.25m (75m inner square) spacing. The area surrounding the 
very high resolution sampling areas was sampled at intermedi 
ate resolutions (125- to 250-m spacing). The remaining extent 
of the farm area was sampled at coarser resolution of 500m 
and/or 1km spacing.