The aim of data conflation is to synergise geospatial information from different sources into a common framework,
which can be realised using multivariate geostatistics. Recently, multiple-point geostatistics (MPG) has been proposed
for data conflation. Instead of the variogram, MPG borrows structures from the training image, so the spatial correlation
is characterised by multiple-point statistics. In pattern-based MPG, two sets of data can be integrated by utilising the
secondary data as a locally varying mean (LVM). The training image provides a spatial correlation model and is
incorporated to facilitate reproduction of similar local patterns in the predicted image. However, the current patternbased
MPG gathers similar patterns based on a prototype class, which extracts spatial structures in an arbitrary way. In
this paper, we proposed an improved pattern-based MPG for conflation of digital elevation models (DEMs). In this
approach, a new strategy for forming prototype class is applied, which is based on the residual surface, vector
ruggedness measure (VRM) and ridge valley class (RVC) of terrain data. The method was tested on the SRTM and
GMTED2010 data. SRTM data at the spatial resolution of 3 arc-second was simulated by conflating sparse elevation
point data and GMTED2010 data at a coarser spatial resolution of 7.5 arc-second. The proposed MPG method was
compared with the traditional pattern-based MPG simulation. Several kriging predictors were applied to provide LVMs
for MPG simulation. The result shows that the new method can achieve more precise prediction and retain more spatial
details than the benchmarks.
Uncertainty is an integral component in thematic mapping, and descriptors such as percent correctly classified pixels
(PCC) and Kappa coefficients of agreement have been devised as thematic accuracy metrics. However, such spatially
averaged measures neither offer hints about spatial variation in accuracy, nor are they useful for error propagation in
derivatives due to the deficiency that spatial dependency is not properly accommodated. Geostatistics provides a good
framework for spatial uncertainty characterization, as conditional simulation is designed for generating equal-probable
realizations of often sparsely sampled fields of concern, which can be summarized for error statistics or subjected to
particular geo-processing to facilitate error propagation. Often, for modeling errors in area-class maps depicting
distributions of spatial classes, stochastic indicator simulation is employed. Unfortunately, indicator approaches suffer
from non-invariant behaviors in simulated classes as class labels are drawn from intervals of class probabilities that are
arbitrarily ordered. Discriminant space-based models have been proposed to enhance consistency in mapping spatial
classes and replicability in modeling spatial categorical uncertainty. This paper explores bivariate (rather than univariate)
discriminant models and extends uncertainty modeling from single-time to bi-temporal area-class maps. Experiment
using simulated data sets was carried out to quantify errors in area classes and their propagation in change analysis. It
was found that there are significant differences between the results obtained by discriminant models and those by
indicator geostatistics. Further investigations are anticipated incorporating real data for mapping and propagating errors
in area classes.
Earlier research has discussed the concept of discriminant space and its applications in area-class mapping and
uncertainty characterization. Both simple univariate cases with b=1 (b being the dimension of the discriminant space)
and multivariate cases with b>1 were analyzed with simulated and real data sets, respectively. This paper describes
combined use of generalized linear models and kriging for scalable area-class mapping, with the former deterministically
predicting mean class responses and the latter making use of spatially correlated residuals in the predictive class models.
Scalability in area-class mapping is facilitated by scale-dependent prediction of mean class responses and kriging of the
residuals over specific gridding cells. The methodology was implemented with topographic data and Landsat TM
imagery concerning land cover mapping in central western Montana, which confirmed the effectiveness of the proposed
strategy combining regression and kriging for scalable mapping of area classes.
Earlier research has introduced the concept of discriminant space, which is spanned by the covariates underlying
area-class occurrences, for increased consistency, interpretability, and replicability in area-class mapping and uncertainty
characterization. While simple univariate cases with b=1 (b being the dimension of the discriminant space) were
investigated previously using simulated data, real world applications are usually multivariate with b>1, thus giving rises
to the need for developing discriminant models in spaces of higher dimensionality for increased applicability. This paper
describes combined use of generalized linear modeling and kriging for area-class mapping, with the former
deterministically predicting mean class responses while the latter making use of spatially correlated residuals in the
predictive class models. Scalability in area-class mapping is facilitated by flexible implementation of scale-dependent
prediction of mean class responses and point- vs. area-support kriging of the residuals. This is followed by an empirical
study concerning land cover mapping in central western Montana, which confirmed the effectiveness of the proposed
strategy combining regression and kriging for scale-dependent mapping of area classes.
The rational function model (RFM), also known as rational polynomial coefficients (RPCs) or rational polynomial
camera (RPC) model, is a generalized sensor model. Different from rigorous sensor model, RFM does not need to obtain
the interior and exterior orientation geometry and other physical properties associated with the physical sensor. RFMs
were first adopted by Space Imaging company as a replacement for rigorous sensor models, and it drew much attention
from the commercial satellite data vendors who rapidly followed the suit in order to protect the confidential information
of the sensors. This paper focuses on the solution for rational polynomial coefficients, RFM-based stereo-model
reconstitution, and positional accuracy analysis. As RPCs do not have obvious physical meanings and their solution is
iterative, analytical approaches to accuracy analysis may not be feasible; computer simulation is thus adopted to quantify
accuracy in RPC-determined positional data. The simulation-based strategy is efficient in mapping local features in
positional errors, which contain both the systematic and random components.
Despite developments in error analysis for discrete objects and interval/ratio fields, there exist conceptual problems with the case of nominal fields. This paper seeks to consolidate a conceptual framework based on the discriminant space for categorical mapping and error modeling. The discriminant space is defined upon the essential properties and processes underlying occurrences of spatial classes, and lends itself to geostatistical analysis and modeling. The discriminant space furnishes consistency in categorical mapping by imposing class-conditional mean structures that are associated with discriminant or "environmental" variables in various statistical models, and facilitates physically interpretable and scale-dependent error modeling. Further research will focus on models and methods based on multi-dimensional discriminant space and at multiple scales.
Nowadays, large quantity of data at faster repeatability is generated from various remote sensors and prompts for spatio-temporally integrated strategies for data handling and information extraction. Change detection is one of the essential techniques for near real-time analysis in remote sensing of the environment. Assuming overall phonological conditions being comparable, change detection is performed either on two-point timescale (bi-temporal) or on a continuous timescale (temporal trajectory analysis), with the latter having the advantage of minimizing the influence of phenology. Univariate image differencing is the most widely applied change detection algorithm, which involves subtracting one date of imagery from a second date that has been co-registered to the first. With "perfect" data, positive and negative values would represent areas of change in the resultant difference imagery, and zero values representing no change.
To quantify the uncertainty in remotely sensed change detection, a geostatistical framework is proposed so that the mean and standard error in pixel or parcel-based difference between the means of the bi-temporal image/map subsets are computed with spatial and temporal dependence accounted for properly, paving the way for probabilistic mapping of changes. To make the proposed approach adaptable to both regular and irregular sampling schemes, block co-kriging is formulated to evaluate means and standard errors in the differences between spatially aggregated means. The geostatistical framework for uncertainty mapping in bi-temporal image/map-based change detection is tested using simulated data sets, whose spatial and temporal correlation can be prescribed. It is anticipated that the geostatistical approach advocated in this paper will make valuable addition to the literature on spatial uncertainty in remote sensing and change detection.
Bayesian networks are used for reasoning under uncertainty. This paper examines the use of Bayesian networks for integrating multi-temporal remotely sensed data with landform data derived from digital elevation models (DEM) and groundwater data to produce maps showing areas affected by salinity in the Yellow River Delta of China. Incorporating prior knowledge about the relationships between input attributes and their relationship with salinity, a conditional probabilistic network is used to impose a known relationship between input attributes and salinity status. The results are compared with maximum likelihood classification techniques using single-date Landsat TM imagery. They show a large improvement on the maximum likelihood classifier. The network is used to produce a time-series of landcover and salinity maps for the Yellow River Delta.