133
14. REMOTE SENSING OF LAND COVER AND LAND COVER
CHANGE
10.
CALIBRATION, VALIDATION AND
INTEGRATION
It is possible to use correspondence or ‘confusion’ matrices to calibrate
remotely sensed estimates of cover against validated reference data to derive
estimates which are corrected for systematic bias in the remote sensing (e.g.,
Gonzalez-Alonso and Cuevas, 1993). Such techniques offer practical
solutions to the estimation of change statistics. Other techniques, such as
ratio-estimation and regression estimation, are well reported and understood
(e.g., Cochran, 1977). They offer enormous and largely unexplored potential
for refinement of monitoring methods.
11.
PRESENTATION OF RESULTS
Rather little attention has so far been given to the important problem of
how best to present and communicate information on land cover change.
Typically, this information is conveyed as simple tabulations or maps,
indicating ‘before and after’ extents of different land cover categories. If
spatially explicit information is available, then it becomes possible to
compute matrices of change, which, amongst other advantages, makes it
possible to record successional pathways. Shi and Ehlers (1996) give
examples of spatial representations of uncertainties in estimates of land cover
and land cover change, and, generally, there is enormous scope for greater
deployment of graphical devices, including 3-D projections and color, as a
means of conveying the location, extent and rate of change. The existence of
PC-based software, aimed at non-specialist users, (e.g., Haines Young et al.
1994), cries out for a more innovative approach to this task.
12.
SUMMARY AND CONCLUSIONS
The measurement of land cover change from remote sensing is presently
some way from operational status, despite a long history of the application of
Earth Observation for land cover mapping. The fundamental challenge is to
distinguish change from artifacts in the data with sufficient precision and
consistency. Key limiting factors have been identified above. Increased
spatial resolution from new sensors (e.g., IRS-1C) and those to be launched
shortly will lead to increased accuracy in the mapping of land boundaries.
Hyper-spectral systems offer the prospect of increased powers of
14. REMOTE SENSING OF LAND COVER AND LAND COVER
CHANGE
10.
CALIBRATION, VALIDATION AND
INTEGRATION
It is possible to use correspondence or ‘confusion’ matrices to calibrate
remotely sensed estimates of cover against validated reference data to derive
estimates which are corrected for systematic bias in the remote sensing (e.g.,
Gonzalez-Alonso and Cuevas, 1993). Such techniques offer practical
solutions to the estimation of change statistics. Other techniques, such as
ratio-estimation and regression estimation, are well reported and understood
(e.g., Cochran, 1977). They offer enormous and largely unexplored potential
for refinement of monitoring methods.
11.
PRESENTATION OF RESULTS
Rather little attention has so far been given to the important problem of
how best to present and communicate information on land cover change.
Typically, this information is conveyed as simple tabulations or maps,
indicating ‘before and after’ extents of different land cover categories. If
spatially explicit information is available, then it becomes possible to
compute matrices of change, which, amongst other advantages, makes it
possible to record successional pathways. Shi and Ehlers (1996) give
examples of spatial representations of uncertainties in estimates of land cover
and land cover change, and, generally, there is enormous scope for greater
deployment of graphical devices, including 3-D projections and color, as a
means of conveying the location, extent and rate of change. The existence of
PC-based software, aimed at non-specialist users, (e.g., Haines Young et al.
1994), cries out for a more innovative approach to this task.
12.
SUMMARY AND CONCLUSIONS
The measurement of land cover change from remote sensing is presently
some way from operational status, despite a long history of the application of
Earth Observation for land cover mapping. The fundamental challenge is to
distinguish change from artifacts in the data with sufficient precision and
consistency. Key limiting factors have been identified above. Increased
spatial resolution from new sensors (e.g., IRS-1C) and those to be launched
shortly will lead to increased accuracy in the mapping of land boundaries.
Hyper-spectral systems offer the prospect of increased powers of