378 
10. DISCUSSION 
REFERENCES 
The true status of spatially fixed sample units 
or cells are expected to have heterogeneous 
variance and lack independence, caused by 
landscape level processes such as regional land 
use practices, climatic patterns, and 
physiographic gradients. Therefore, no new 
complications are introduced by heterogeneous and 
dependent errors propagated from regional 
calibration and deterministic prediction models 
applied to 40 km 2 sample units, or small area 
estimation techniques for ancillary data applied 
to 640 km 2 cells. 
It is frequently assumed that sampling errors 
associated with a systematic sample of plots in 
space are independent and identically 
distributed. These unrealistic assumptions will 
not bias estimates of stratum status, but there 
would be loss of efficiency, and bias in the 
estimated covariance matrix for stratum stratus 
estimates. Biased estimates of the covariance 
matrix might adversely affect important tests of 
hypothesis, and stepwise regression analytical 
models. Therefore, heterogeneity and lack of 
independence among should be expressed in the 
statistical models. 
Additional statistical details need development 
before the hypothetical example in this paper 
could be implemented. This example is 
univariate, where status is defined as proportion 
of forest. More detailed categories would be 
required in a true landscape monitoring system, 
and the estimators in this paper would have to be 
developed for the multivariate case. Estimating 
model prediction error with remeasurements of 
permanent plots would require multivariate roots 
of polynomial matrix equations. Combining 
ancillary data from other monitoring sources 
would require multivariate, small area estimation 
techniques to estimate status of individual 
cells. It is assumed that the stratum is 
homogeneous, but multivariate spatial trends in 
status might be expected. Multivariate 
geostatistical methods used to estimate spatial 
trends and heterogeneous variance among sample 
units must deal with propagated heterogeneity and 
dependence from multivariate calibration and 
deterministic prediction models. Multivariate 
logit transformations, or the multivariate 
Dirichlet distribution might be needed to better 
deal with skewed error distributions for 
proportion estimates that approach zero. 
The procedures outlined in this paper might have 
conceptual appeal to some, but they have never 
been put into operation within a broad scale, 
landscape level, environmental monitoring system. 
More work is needed to verify their applicability 
and feasibility. Alternatives, such as an 
interpenetrating design without the model based 
Kalman filter might be less risky, but could be 
less efficient, and might be incapable of testing 
deterministic models to improve understanding of 
system dynamics. Contingency plans should be 
made in case- a design based or model based 
approach is found unacceptable. 
Czaplewski, R.L., 1990. Kalman filter to update 
forest cover estimates. In State of the Art 
Methodology of Forest Inventory. Syracuse, 
NY-USA, in press. 
Czaplewski, R.L., Catts, G.P., and Snook, P.W., 
1987. National land cover monitoring using large, 
permanent photoplots. In Land and Resource 
Evaluation for National Planning in the Tropics. 
Chetumal, Mexico, pp. 197-202. 
Czaplewski, R.L., Alig, R.J., and Cost, N.D., 
1988. Monitoring land/forest cover using the 
Kalman filter, a proposal. In Forest Growth 
Modelling and Prediction. Minneapolis, MN-USA, 
pp. 1089-1096. 
Dixon, B.L., and Howitt, R.E., 1979. Continuous 
forest inventory using a linear filter. Forest 
Science. 25:675-689. 
Gregoire, T.G. and Walters, D.K., 1988. Composite 
vector estimators derived by weighting inversely 
proportional to variance. Canadian Journal of 
Forest Research. 18:282-284. 
Hay, A.M., 1988. The derivation of global 
estimates from a confusion matrix. International 
Journal of Remote Sensing. 9:1395-1398. 
Maybeck, P.S., 1979. Stochastic Models. 
Estimation and Control. Vol. 2. Academic Press, 
New York. 423pp. 
Sorenson, H.W., 1985. Kalman filtering: Theqry 
and Applications. IEEE Press Inc., New York. 
457pp. 
Tenenbein, A., 1972. A double sampling scheme for 
estimating from misclassified multinomial data 
with applications to sampling inspection. 
Technometrics 14:187-202.