Development of the Global Multispectral Imager Thermal Emissive FCDRs - Satellite-based Applications on Climate Change - page 74

5.3.2 Band Transfer for FCDR Generation

Once the global testing database, including simulated proxy data for various version
of AVHRR and MODIS sensors, as well as scene characteristics is established,
statistical analysis is conducted to derive the relationships between similar bands
of different sensors.
Figure 5.6a-c illustrates the linear relationships between Aqua MODIS band 31
and band 4 of NOAA-16 AVHRR, NOAA-17 AVHRR, and NOAA-18 AVHRR,
respectively. Band transfer equations among these sensors at 11-

m band and
statistics are listed in Table 5.3 . For all the selected sensor pairs, linear relationship
fits very well, with R 2 higher than 0.999 and very small root-mean-square error
(RMSE), which are around MODIS sensor accuracy requirements.
Similarly, Fig. 5.7a-c illustrates the linear relationships between Aqua MODIS
band 32 and band 5 of NOAA-16 AVHRR, NOAA-17 AVHRR, and NOAA-18
AVHRR, respectively. Table 5.4 lists the band transfer equations and statistics of
the 12-

μ

m band. For all the selected sensor pairs, linear relationship fits even better
than the 11- μ m band, with R 2 higher than 0.9999 and root-mean-square errors
(RMSE) at 0.0412223, 0.0262475, and 0.0373566 K, respectively. These errors
meet MODIS sensor accuracy requirements.
For VIIRS, Terra MODIS, and other version of AVHRR sensors, band transfer
equations can be determined similarly using the global testing database. More
accurate band transfer equations can be obtained if scene characteristics are taken
into account (Hao et al. 2005b ). So, the band mapping approach is feasible for
construction of long-term thermal emissive FCDRs.

μ

5.4 Conclusion and Discussions

In this chapter, technical approaches for thermal emissive FCDR generation with
AVHRR, MODIS, and VIIRS measurements are presented, and preliminary results
are demonstrated and analyzed based on a global testing database constructed with
global AIRS measurements of selected 8 days during year 2002 and 2008. The
performance of spectral mapping is investigated by comparing AIRS-simulated
MODIS and aggregated MODIS measurements. As the differences between AIRS-
simulated MODIS and aggregated MODIS measurements are quite small, it is
feasible to use AIRS-simulated proxy datasets for FCDR generation. Band transfer
equations are derived, and performances of band transfer models are analyzed
statistically. For all the selected sensor pairs, determinant coefficients of linear
band transfer equations are almost 1, and the root-mean-square errors are very
small, accurate enough for most applications.
Certainly, the presented approach relies on consistent calibration of selected
sensors. Calibration consistency and stability are critical for Climate Data Record
generation. Inter-sensor calibration based on comparison of measurements from
different sensors at collocated sites can help to improve the calibration consistency