A Basin-Scale Econometric Model for Projecting Future Amazonian Landscapes

Robert Walker (Principal Investigator) Michigan State University; Eustáquio Reis, Institute of Applied Economics Research (IPEA), Brazil

As an example, this image shows a situation in which deforestation precedes road-building. It depicts in red several settlement roads in 1988; deforested areas, as of 1988, are shown by the yellow polygons extending beyond the roads. Since the roads now pass through these old deforested areas, the figure suggests reverse causality, in which deforestation actually leads to road-building. This situation is probably common in areas of smallholder colonization. Download higher resolution image.

A Basin-Scale Econometric Model for Projecting Future Amazonian Landscapes was developed to predict forest loss associated with development scenarios in the Amazon basin. Given the scenarios, projections follow from results of econometric modeling based on economic theory and detailed local observation (led by Alexander Pfaff of Columbia University). The empirical analysis benefits from enhanced statistical “power” due to an expanded database using information for census tracts, not counties. This yields thousands of observations, enabling the use of fixed county effects to control for unobserved differences across space, as well as a partitioning of the sample to control for previous deforestation.

Findings applicable to most of the region show that new roads will raise deforestation rates. Further, contrary to a claim in the literature that road-building in highly cleared areas could lower deforestation rates, no results are significant and negative for any level of previous clearing. The project is extending this work based on understandings, new to such literature, about complex relationships between roads and deforestation. Project research has shown that road “endogeneity” can lead to mistaken statistical estimation of road effects, and is developing ways to minimize associated biases. Fieldwork documenting and addressing such interrelationships is led by Stephen Perz of the University of Florida. http://marajo.geo.msu.edu/lba/

Mapping and Modeling Land Use/Land Cover Dynamics in the Northern Ecuadorian Amazon

Stephen Walsh, Richard Bilsborrow (Principal Investigators), University of North Carolina-Chapel Hill

This Ecuador Project uses longitudinal household survey data collected in 1990 and 1999, a 2000 community survey, a multi-resolution satellite imagery time-series, GIS coverages of resource potentials and endowments, and field verification and geodetic control data to analyze the determinants of changes in land cover/land-use (LCLU) change at the plot, sector, and regional levels in the Northern Ecuadorian Amazon. The fundamental research questions revolve around (a) the rates, patterns, and mechanisms of forest conversion to agricultural and urban uses; (b) the relative importance of exogenous and endogenous variables to these land uses; (c) the associated scale-dependent drivers of LCLU dynamics and pattern-process relations operating across socio-economic and demographic, biophysical, and geographical domains; (d) rate and pattern of land conversion from forest to agricultural crops, pasture, secondary plant succession, and urbanization, as well as the rate and pattern of land abandonment at the farm level; and (e) plausible scenarios of future LCLU change and the policy implications as assessed through multi-level models, spatial lag models, neutral models, cellular automata models, and agent based models.

This graphic shows the initial landscape conditions and model outcomes of a generated cellular automata model: (a) Landsat TM LCLU classification for the year 1986 that sets the initial condition for the model, and (b) a predicted LCLU surface for 2010 relative to a defined scenario of LCLU dynamics.

Recent cellular automata and agent based models integrate space-time scales as well as global, regional, and local effects to derive rules and weights of variable behavior, neighborhood conditions, initial conditions, feedback mechanisms, and critical threshold. The spatial simulations are space-time sensitive and policy relevant and address important scenarios of LCLU change. The graphics below show (a) the LCLU classification of a portion of a 1986 Landsat TM image that was used to set the initial model conditions and (b) predicted LCLU for 2010 using growth or transition rules implemented within a cellular automata environment. This scenario examined deforestation and agricultural extensification at the farm-level as a consequence of terrain conditions, road features, population at the household, community characteristics, regional level demographics, and a generated surface of household income. http://www.cpc.unc.edu/projects/ecuador

Rural land-use change in the conterminous U.S.

Dan Brown (Principal Investigator), University of Michigan; Pierre Goovaerts, BioMedware, Inc.; Kathleen Bergen, University of Michigan

County-level land-use changes from 1950 to 2000, based on censuses of population, housing, and agriculture. A) change in population density; B) change in land area settled at “exurban densities” (i.e., 1 house per 1 to 40 acres); C) change in percent cropland (Brown et al. 2005).

This team of researchers at the University of Michigan, is investigating the land-cover consequences of land-use changes associated with “rural sprawl,” or the widespread development of exurban areas (e.g., outside cities, towns and suburbs) for relatively low-density settlement (Figure; Brown et al. 2005). The work is also supported by related grants from the NSF and USDA Forest Service. The team is using remote sensing to (a) map changes in landscape productivity (i.e., gross primary production) at a coarse resolution (i.e., 1km) over the 1990s and evaluate the relationships between these observed changes and demographic changes over the same period, (b) classify land-cover for four dates during the period 1985 to 2001 from moderate resolution Landsat images, (c) classify land-use and land-cover for individual parcels mapped using plat maps and historical aerial photographs from the period 1950 to 2000, and (c) develop geostatistical models to describe and model the spatial patterns of these changes over time. With support from an Earth System Science Fellowship (ESSFP) PhD grant (Amy Powers) the team is also investigating the spatial and temporal patterns of classification errors, so that future land-cover change investigations might benefit from an understanding of how patterned errors affect the accuracy of land-cover change information and models and analyses that rely on this information. The research provides a richer, multi-scale picture of how land-cover in exurban areas of the Eastern U.S. are changing in response to conversion of agricultural land-uses to low-density residential areas. Exurban areas are experiencing significant increases in productivity, resulting from the replacement of crops with perennial grasses and substantial amounts of increased tree cover. Data from U.S. censuses of housing and agriculture suggest that the changes are widespread and we can infer that the changes could play a significant role in forcing changes in regional scale ecosystem and atmospheric processes. This research is producing tools that take multiple approaches to modeling and that can be used in planning and policy applications associated with land-use and ecosystem impacts.

Spatial Predictive Modeling and Remote Sensing of Land Use Change in the Chesapeake Bay Watershed

Scott J. Goetz, Woods Hole Research Center; Nancy E. Bockstael, University of Maryland (Principal Investigators); Claire A. Jantz, Shippensburg University

Urban extent as of the year 2000 in the greater Washington DC metropolitan area. Download higher resolution image.

This work addresses mapping and monitoring urbanization in the Chesapeake Bay Watershed, where the spatial patterns of “sprawl” represent a set of conditions generally prevalent in much of the nation. Using innovative remote sensing techniques and spatially modeling the land-use change decision at the pixel (cell) and the individual property owner level, scenario analyses of future land-use dynamics provide critical quantitative insight into the impact of alternative land management and policy decisions. These are specifically aimed at addressing the effectiveness of implemented and proposed growth control policies. The project generated both remote sensing and spatially explicit socio-economic data to estimate and calibrate the parameters for the two different types of land-use change models. One, a cellular automata model (CA), was driven largely by observations of past patterns of land-use change from satellite observations. The other, an economically-based model (EC), was driven by mechanisms of the land-use change decision at the parcel level. The CA model does not incorporate the immense detail of the EC model, and the spatial pattern of development predicted by the two models is quite different. Modifications to the CA model, and manipulation of ‘parameter sets’ by incorporating rules from the economic model, were found useful for improving CA model performance – especially in capturing low-density residential development. This project may be the first serious attempt at developing both types of models for the same area, using as much common data as possible. The strengths and weaknesses of the two approaches were identified and this team of researchers plan to continue to revise each model in the light of new data and new lessons learned through continued collaboration.

Other LCLUC Prediction and Modeling Projects

• Small, Christopher - Columbia University. Development and Sensitivity Analysis of High Resolution Land Surface Parameters from Satellite Data and their Use in a Mesoscale Model – LDEO