PROGRESSIVE SAMPLING - FIDELITY AND ACCURACY
K. Tempfli
International Institute for Aerospace Survey and Earth Sciences (ITC)
Enschede, The Netherlands
ABSTRACT
Progressive sampling of terrain relief results in an incomplete grid
DEM. The grid density is locally adapted to the terrain variability.
Progressive densification of the sampling grid is controlled by an on-
line analysis of the already measured elevation values using as
criterion a threshold for the second differences. Grid spacing,
threshold, and number of densification runs determine the achievable
fidelity of the terrain representation. The concept of transfer function
is utilized to quantify fidelity. Accuracy of the DEM is estimated from
transfer function and power spectrum of the terrain relief.
1. EXORDIUM
The comparative test /13/ of photogrammetrically sampled digital eleva-
tion models (DEMs), which was conducted by ISPRS Working Group III/3
(1980-84), has shown that accuracy prediction and accuracy assessment of
a DEM still constitute a problem for many DEM producers. Resolution 3 of
Commission III, which was approved by the congress in Rio de Janeiro
recommends pursuing the endeavour in developing methods for accuracy
assessment. In /11/, an approach was elaborated for regular grid DEMs,
thus equispaced sampling (ES), which is based on the transfer function
of the system of sampling and subsequent interpolation and spectral
analysis of its input, i.e., terrain and additive measuring error. The
empirical evaluation of the method--using artificial, computer-generated
surface and profile data--was encouraging.
Progressive sampling (PS) /5/ aims photogrammetrically collecting
precise and comprehensive terrain relief data. In order to make sampling
time-efficient and to obtain data in a favourable structure, elevation
is measured in a grid of variable density, locally adapted to the fluc-
tuations of terrain relief. Collecting data by progressively increasing
the sampling density requires on-line data analysis and synthesis, which
can be realized on a computer-supported analogue instrument /3/ or an
analytical plotter /1/, /8/ and /12/. The irregular sampling pattern and
its dependency on terrain geometry make analytical accuracy estimation
more complex than for ES. After reviewing the approach to accuracy as-
sessment for ES, a tentative extension of this concept to PS will be
presented.
2. EQUISPACED SAMPLING
A digital model of a (topographic) surface is attained by sampling the
surface and reconstructing it from the sample by interpolation. The
£76539-