2.5D Morphogenesis: modeling landuse and landcover dynamics in the Ecuadorian Amazon
Joseph P. Messina
Department of GeographyMichigan State University USA
Stephen J. Walsh
Department of Geography University of North Carolina Chapel Hill USA
Plant Ecology September 2001, Volume 156, Issue 1, pp 75–88
Abstract
The Ecuadorian Amazon, lying in the headwaters of the Napo and Aguarico River valleys, is experiencing rapid change in Land Use and Land Cover (LULC) conditions and regional landscape diversity uniquely tied to the spontaneous agricultural colonization of the Oriente region of northeastern Ecuador beginning in the mid to late 1970s. Spontaneous colonization occurred on squattered lands located adjacent to oil company roads and in government development sectors composed of multiple 50 ha land parcels organized into `piano key' shaped family farms or fincas. Portions of these fincas were deforested for agricultural extensification depending upon the age of the finca and several site and situation factors. Because fincas are managed at the household level as spatially discrete, temporally independent units, land conversion at the finca-level is recognized as the chief proximate cause of deforestation within the region.
Focusing on the spatial and temporal dynamics of deforestation, agricultural extensification, and plant succession at the finca-level, and urbanization at the community-level, a cell-based morphogenetic model of Land Use and Land Cover Change (LULCC) was developed as the foundation for a predictive model of regional LULCC dynamics and landscape diversity. Here, LULC characteristics are determined using a time-series of remotely sensed data (i.e., Landsat Thematic Mapper (TM) and Multispectral Scanner (MSS)) using an experimental [semi-traditional] (hybrid unsupervised-supervised) classification scheme resulting in a time-series data set including LULC images for 1973, 1986, 1989, 1996, and 1999. Pixel histories of LULC type across the time-series were integrated into LULC trajectories and converted into seed or input data sets for LULC modeling to alternate time periods and for model validation. LULC simulations, achieved through cellular automata (CA) methodologies, were run on an annual basis to the year 2010 using 1973 as the initial conditions and the satellite time-series as the `check points' in the simulations. The model was developed using the Imagine Spatial Modeler of the ERDAS image processing software, and enhanced using the Spatial Modeler Language (SML). The model works by (a) simulating the present by extrapolating from the past using the image time-series, (b) validating the simulations via the remotely sensed time-series of past conditions and through field observations of current conditions, (c) allowing the model to iterate to the year 2010, and (d) comparing model outputs to an autoregressive time-series approach for annual conditions that are compared via paired t-tests of pattern metrics run at the landscape-level to define compositional and structural differences between successive model outputs.
Key words: Cellular automata, Deforestation and agricultural extensification, Pattern metrics , Satellite time-series, Simulation
Discussion and conclusions
Upon initial examination, the two images (Figure 4),one classified from satellite data and the other mod-eled through CA-based simulations appear somewhat different with smoothing occurring in the CA model run. However, by comparing the model output with landuse/landcover data of the finest resolution, both the successes and the flaws of the model become ap-parent. Recall that our attempt was to spatially modela subset of factors primarily associated with satellite change images, road and water patterns, and the posi-tion of the central city within the region. Demographicand socio-economic characteristics of households first assessed in 1990 through a probabilistic sample of ap-proximately 480 fincas, and repeated in 1999, have not yet been integrated into the analysis because of ongoing data preprocessing. Their subsequent inclu-sion should serve to reduce image variances between modeled and controlled views suggesting the impor-tance of household decision-making in deforestation and agricultural extensification and the resulting pat-terns, rates, and directions of land use and land cover change as central factors in altering regional biodi-versity at the landscape-level. In addition, any efforts for generalizability would be lost via overfitting the regional process based conditions to any model rules or parameters.
The model predicts (1996) the total landscape change within the ISA region reasonably well with slight overprediction of agricultural expansion and slight underprediction of urban expansion. The sum-mary correlations are lower than would be expected in a less complex environment or one operating at asynoptic scale than represented by the dynamic and fragmented landscape of the selected ISA. Specifi-cally, the low urban correlation is due in large partto the effects of roads (only a 1990 road coverage was used in this research, but in subsequent efforts an updated 2000 road coverage will be used so that the expansion of the road network and the improvements in the type of road surfaces can be included in the model simulations; however, it might be best to orga-nize the road coverage by road segments in which the type of surface by year could be maintained within the GIS database and integrated into the model on an an-nual basis to correspond to the simulation time-steps),random oil exploration activities, apparent change in social functions across the landscape, coffee and cat-tle prices, the development of a service economy in and adjacent to market towns, and the subdivision of fincas into relatively small parcels of land for resi-dential development and the small scale cultivation of commercial crops.
The relative regional homogeneity of the biophys-ical landscape (e.g., climatic and soil regimes), and the importance of local farmers in deforestation as a response to the geographic accessibility to towns and the expanding transportation infrastructure permits the wider application of the model to multiple social and physical scales. While, one cannot avoid the fact thatthe spatial fit of the model is less than ideal, it is impor-tant to realize that no methods exist to quantitativelyevaluate the fit. Consequently, the autoregressive time-series comparison was used to bridge that gap. Onelimitation to both simulation techniques is that bothare sensitive to initial conditions and small pertur-bations in the classification process. However, the autoregressive technique is much more sensitive toswings in polarity or positive/negative trends. Thismanifestation is easily seen with the class 3 predictedversus simulated classes (see Table 4). The simula-tions do suggest a more homogenous landscape with time, a scenario that fits our theoretical understand-ing of how in-migration of farmers into existing fincas through resale, sub-division of fincas to those engaged primarily in the burgeoning service sector, and the es-tablishment of new development sectors will alter the natural landscape through deforestation and agricul-tural extensification. But the selected ISA is already substantially converted to agriculture (and urban set-tings are increasing in number and areal extent), and the time since fincas were established has permit-ted sufficient time for secondary plant succession of crops to pasture and then to forest to occur, particu-larly given the soil infertility curves that link yieldsto LULC decisions and the motivation of farmers to consider opening new areas in their fincas for agri-cultural extensification instead of intensification of existing plots through fertilization and other land in-puts. The pattern metrics suggest that the landscapeis becoming more spatially homogenous with frag-mentation increasing slightly for some classes overthe simulated dates from 1998 to 2010 and decreas-ing and then increasing during the 1986 through 1996control period; the contiguity of the landscape is de-creasing with time for both the mapped and modeleddates. These non-linear swings in LULC are expected.We anticipate that for the selected ISA (and other ar-eas having developing urban centers and an expandingtransportation network) that the landscape will period-ically reflect an increase in fragmentation when newroads are constructed or when existing roads and trailsare improved. Greater geographic accessibility pro-vides an incentive to expand agriculture and to reapincome from the sale of wood products because ofthe greater connection to market centers and productmiddlemen. While lands transitioning through sec-ondary plant succession will restore deforested landto a forested condition with time, new areas will beconverted to keep up with the demand and opportunityfor the sale of subsistence and commercial products atformal and informal markets. If the use of fertilizersand technology in general becomes more widespread,the land parcels already in agricultural production asa consequence of previous deforestation may remainin production for a longer period of time since thedecay in crop yields will be altered. Then, we arelikely to see a possible reduction in lands being con-verted to alternate landuses as they transition back toforest classes. But greater geographic access, agri-cultural inputs, technology, and a greater regional(and possibly global) demand for agricultural prod-ucts may serve as sufficient inducements for farmers toexpand their agricultural base through extensificationeven with the sustainment of production in previouslydeforested land parcels through intensification. There-fore, the landscape (at least in this ISA with its specificsite and situation characteristics) may cycle from alandscape dominated by forest to one dominated by urban and agriculture, thereby reflecting a landscapeinterpretation via the metrics of homogeneity (for-est), heterogeneity (forest-agriculture), and a return tohomogeneity (agriculture).
The true measure of spatial complexity as applied in a CA context is one not yet fully realized in the literature. The tools exist to build complex (in the complicated not complexity sense) models to predict LULCC in a variety of environments and time peri-ods. Without further development of the theoretical underpinnings of complexity theory and spatial pat-tern, addressing for example, ‘what is a good fit,’the models run the very real risk of over-specification and unverifiable results. The comparison with the autoregressive technique illustrates the problem. While some landscape level characteristics can be predicted and meet traditional tests of significance, the spatialpattern of a place remains unknown. Using CA as aspatial surrogate to statistical tests allows the assign-ment of class to place. However, the actual time-seriesused is quite small and some basic assumptions of sta-tistical tests may be violated. Many of the results are predictable though given the constraints and built-inrules of CA.
The Oriente, like most places, is surprisingly com-plex. As snapshots in time, images provide the raw material to assist in the derivation of complex envi-ronmental and human processes – in effect, to see patterns instead of isolated points and relationships between different distributions. The challenge of mod-eling population-environment interactions closely par-allels the methodological concerns of much of socialscience. With improved data handling, the modeling scheme presented here is extendable to a variety of tropical environments and regional contexts,making cellular automata modeling not only a promoter of research into predictive spatial systems, but more importantly, an effective approach to modeling population-environment interactions across multiple spatial, temporal, and thematic domains.
This work combines the historically descriptive aspects of the nature-society discourse with remote sensing and landscape ecology pointing towards the development of a Northern Oriente analogue. How-ever, the challenges are many. None the least of which are:
– use of more suitable software for subsequent CA development,
– integration of endogenous and exogenous factorsin the model based upon the theories in human,political, and landscape ecology,
– extension of the satellite time-series data for model development and calibration,
– incorporation of time-lags and scale dependence of variables and their effects,
– development of non-linear statistical tools better suited for CA systems, and
– updated surfaces that represent proximate causes of deforestation and their hypothesized feedbacks across domains and thresholds of effects framed within a space-time context.
Full Text
ليست هناك تعليقات:
إرسال تعليق