Geoscience Reference
In-Depth Information
3. Results and Discussions
Comparison of ANN and IAPP Retrieval Accuracies
Retrieval results are evaluated by the RMS differences between the retrievals
from IAPP and ANN with the collocated NCEP analysis. The root mean square
error of the retrieval is defined as
K
1
Ç
2
(
X
-
X
)
RMS =
(2)
NCEP
RETER
K
i
1
where
X
NCEP
and
X
RETR
are the NCEP analysis and retrieved temperature
respectively and
K
is the total number of comparisons.
Figure 2(a) shows the RMS retrieval errors for the training dataset from
the trained neural network and IAPP. The neural network temperature retrievals
are found to be superior to the IAPP retrievals at all levels, especially at the
low levels. This is because in the IAPP a fast and accurate transmittance model
is generated for the RTE calculation which is significantly dependent on the
microwave forward transfer model. Therefore it is very difficult to calculate
accurate brightness temperatures of the low-level and window channels due to
the difficulty of obtaining enough accurate surface parameters, especially the
surface emissivity. The microwave emissivity varies with the surface conditions,
local zenith angle and other factors. Due to these factors brightness temperatures
for surface-viewing channels can differ by several tens of degrees with a same
atmospheric profile. Li et al. (2000) also pointed out that the retrieval could be
subject to a large error especially in the lower atmospheric levels due to some
uncertainties, such as the failure of low-level cloud check, surface type
uncertainty and emissivity error.
Figure 2(b) shows the RMS retrieval errors on the neural network and
IAPP over the testing dataset. The results are not satisfactory as compared to
the training data set. This is because the maximum frequency of match-up data
found over the ocean in the testing dataset was quite less in numbers. From the
figures (Figs 2a and 2b), it can be stated that the retrieve results from the
neural network in training dataset are better than testing dataset but they are
still comparable. Especially the results from the neural network are significantly
better than those from IAPP at higher level near the 300-100 hPa. It also appears
that the neural network method is capable of quite accurately and rapidly
retrieving vertical temperature profiles in the testing dataset with the least
number of collocated points. If enough collocated data of other region are
available and used to train the neural network, the retrieval results may improve.
In the statistical comparison of ANN versus IAPP, the overall 2567 and 993
collocated data pairs have been used for computing the training and testing
data sets respectively.
Figure 2(c) illustrates the overall RMS errors of the neural network and
IAPP schemes. The figure shows that the surface has the largest RMS value of
Search WWH ::
Custom Search