Abstract
Landscape monitoring is an important issue in landscape ecology and forest ecosystem management. This study used geographic information systems to integrate landscape models at different levels to analsyze landscape changes of the Liukuei ecosystem management area in Taiwan. Landscape maps of 1988 and 1996 wee derived from aerial photographs using digital photogrammetry. GIS programs wee used to calculate the landscape structure indices under different conditions. Markov models were then used to predict trends of landscape changes. Preliminary results indicate that natural processes and human interference both significantly affected the structures and land cover distributions of the study area. Without continued reforestation, natural forests would reclaim man-made stands gradually. Timber harvesting would not cause long-lasting effects on the landscape. Forestation, on the other hand, may induce landslides or create bare land. To address spatial variability of landscape changes, probabilistic models will be developed to examine factors associated with landscape changes and to predict probabilities of landscape changes spatially.

1. Introduction
Landscapes are assemblages of habitats, communities and land use types, and the spatial configuration of these landscape elements can be attributed to a combination of environmental correlates and human forces (Forman and Godron 1986). Landscapes are dynamic in structure, function, and spatial pattern. Therefore, monitoring landscape dynamics is a basic but important aspect of landscape ecology and ecosystem management. In human-dominated landscapes, changes in a landscape are often due to management practices. Therefore, protecting and preserving ecosystems requires an ability to predict the direct and indirect, temporal, and spatial effects of human activities (Costanza 1987).

Scientists often use landscape models to describe or explain landscape dynamics. Baker (1987) classified landscape models into three categories: (1) whole landscape models, (2) distributional landscape models, and (3) spatial landscape models. Whole landscape models focus on the value of a variable or several variables in a particular land area. Distributional landscape models emphasize changes in the distribution of land cover types. Spatial models, on the other hand, use the location and configuration of landscape elements in projecting change, and can explicitly produce maps of these changing spatial configurations.

There are many application examples of landscape models. For example, Turner (1989), Li (1990) and Dunning et al. (1992) used various landscape indices to measure landscape structures under different conditions. Markov models are probably the most commonly used distributional models. Burnham (1973) used a Markov simulation model to depict the intertemporal landscape changes in a southern Mississippi alluvial valley. Lindsay and Dunn (1979) applied a Markov model to project future distributional patterns of land uses in New Hampshire. With some modifications, distributional landscape models can be used to derive spatial landscape models. For example, Turner (1987) included neighborhood influences in deriving Markov transition probabilities to account for spatial variability. In recent years, probabilistic models such as logic models have become popular in modeling spatial landscape changes. For instance, Dale and others (1993) used a multi-nominal logic model to evaluate the deforestation pattern in Brazil. Hsu (1996) applied a binary logic model to examine factors affecting land use changes and to predict possible land use patterns in the future.

Landscape models at the three different levels are interrelated. Whole landscape models describe the characteristics of a certain landscape; distributional models examine or predict changes of landscape distribution over time; and spatial models determine where the changes might occur. However, those models were rarely integrated. Since landscape models are directly associated with spatial landscape patterns, the use of geographic information systems (GIS) can facilitate the integration of models at different levels. Therefore, the objective of this study is to proposed a framework for integrated landscape analysis. Various landscape indices, Markov models, and multinomial logic models are used to analyze the landscape changes in managed forest landscape in Taiwan. The results can provide useful information to decision-makers.