Spatial Modelling and Scaling Issues - Environmental Modelling: Finding Simplicity in Complexity - page 72

June 4, 2001 August 25, 2001

Figure 5.4 Simulated surface reflectance (bottom) at TM scale (30m) from MODIS reflectance imagery (upper row, 500m) using a
STARFM algorithm at June 4 and August 25, 2001, separately (From Gao et al , 2006). The imagery is a red-green-blue composite
using near infrared, red, and green reflectance.

(Figure 5.4). In this way, high temporal and spatial reso-
lution data are produced for environmental monitoring
and modelling.

both themean function and the covariance of sample data,
whereas the universal kriging only requires determining a
variogram for the extrapolation of point data. Co-kriging
provides a consistent framework to incorporate auxiliary
information, which is easy to obtain, into the prediction.
This technique can predict the spatial data more precisely
(Papritz and Stein, 1999). Other robust measurements of
spatial interpolation include the fitting of splines (e.g. De
Boor, 1978) and the modelling process itself (as discussed
in Chapter 6).
Remote sensing has recently become the main source
of temporal and spatial environmental parameters at
large regions. However, spatial and temporal continuity
from multiple satellite sensors has become a concern.
First, roughly half the global land surface is obscured
due to persistent and transient cloud coverage as well as
ephemeral and seasonal snow effects (Moody et al ., 2005),
which results in extensive spatial data gaps. Thus the direct

5.4.3 Temporal andspatial continuityof input
parameters

Data continuity is needed to create consistent spatial
and temporal parameters in order to assure the reliable
model outputs at regional and global scales. To generate
temporally and spatially continuous input data, field and
remotely sensed measurements are interpolated at a large
extent using various approaches. The kriging method
denotes a set of techniques to predict (extrapolate) data
quantities and their spatial distribution from field point
observationdatasets. Commonly usedmethods are simple
kriging, universal kriging and co-kriging. Simple kriging is
a heterogeneous, best linear unbiasedpredictor in terms of