Search

To purchase or authenticate to the full-text of this article, please visit this link: http://dx.doi.org/10.1111/j.1467-8306.2007.00577.x Byline: Robert Gilmore Pontius (*), Robert Walker ([dagger]), Robert Yao-Kumah (*), Eugenio Arima ([double dagger]), Stephen Aldrich ([dagger]), Marcellus Caldas (s.), Dante Vergara ([dagger]) Keywords: behavioral landscape model; change; land; pattern; scale Abstract: This article describes a quantitative assessment of the output from the Behavioral Landscape Model (BLM), which has been developed to simulate the spatial pattern of deforestation (i.e. forest fragmentation) in the Amazon basin in a manner consistent with human behavior. The assessment consists of eighteen runs for a section of the Transamazon Highway in the lower basin, where the BLM's simulated deforestation map for each run is compared to a reference map of 1999. The BLM simulates the transition from forest to non-forest in a spatially explicit manner in 20-m x 20-m pixels. The pixels are nested within a hierarchical stratification structure of household lots within larger development rectangles that emanate from the Transamazon Highway. Each of the eighteen runs derives from a unique combination of three model parameters. We have derived novel methods of assessment to consider (1) the nested stratification structure, (2) multiple resolutions, (3) a simpler model that predicts deforestation near the highway, (4) a null model that predicts forest persistence, and (5) a uniform model that has accuracy equal to the expected accuracy of a random spatial allocation. Results show that the model's specification of the overall quantity of non-forest is the most important factor that constrains and correlates with accuracy. A large source of location agreement is the BLM's assumption that deforestation within household lots occurs near roads. A large source of location disagreement is the BLM's less than perfect ability to simulate the proportion of deforestation by household lot. This article discusses implications of these results in the context of land change science and dynamic simulation modeling. Author Affiliation: (*)Department of International Development, Community, and Environment, Clark University ([dagger])Department of Geography, Michigan State University ([double dagger])Department of Environmental Studies, Hobart and William Smith Colleges, and Instituto do Homem e Meio Ambiente da Amazonia, Brazil (s.)Department of Geography, Kansas State University Article History: Initial submissions, February 2006; revised submissions, September and October 2006; final acceptance, October 2006 Article note: Correspondence: Department of International Development, Community, and Environment, School of Geography, Clark University, 950 Main Street, Worcester, MA 01610-1477, e-mail: rpontius@clarku.edu (Pontius); Department of Geography, Michigan State University, 118 Geography Bldg., East Lansing, MI, 48824, e-mail: rwalker@msu.edu (Walker), aldric30@msu.edu (Aldrich), vergarad@msu.edu (Vergara); Department of International Development, Community, and Environment, Clark University, e-mail: robby60@hotmail.com (Yao-Kumah); Department of Environmental Studies, Hobart and William Smith Colleges, 300 Pulteney Street, Geneva, NY 14456, e-mail: arima@hws.edu (Arima); Department of Geography, Kansas State University, 116 Seton Hall, Manhattan, KS 66506-2940, e-mail: caldasma@msu.edu (Caldas).

Remote sensing, or Earth Observation (EO), is increasingly used to understand Earth system dynamics and create continuous and categorical maps of biophysical properties and land cover, especially based on recent advances in machine learning (ML). ML models typically require large, spatially explicit training datasets to make accurate predictions. Training data (TD) are typically generated by digitizing polygons on high spatial-resolution imagery, by collecting in situ data, or by using pre-existing datasets. TD are often assumed to accurately represent the truth, but in practice almost always have error, stemming from (1) sample design, and (2) sample collection errors. The latter is particularly relevant for image-interpreted TD, an increasingly commonly used method due to its practicality and the increasing training sample size requirements of modern ML algorithms. TD errors can cause substantial errors in the maps created using ML algorithms, which may impact map use and interpretation. Despite these potential errors and their real-world consequences for map-based decisions, TD error is often not accounted for or reported in EO research. Here we review the current practices for collecting and handling TD. We identify the sources of TD error, and illustrate their impacts using several case studies representing different EO applications (infrastructure mapping, global surface flux estimates, and agricultural monitoring), and provide guidelines for minimizing and accounting for TD errors. To harmonize terminology, we distinguish TD from three other classes of data that should be used to create and assess ML models: training reference data, used to assess the quality of TD during data generation; validation data, used to iteratively improve models; and map reference data, used only for final accuracy assessment. We focus primarily on TD, but our advice is generally applicable to all four classes, and we ground our review in established best practices for map accuracy assessment literature. EO researchers should start by determining the tolerable levels of map error and appropriate error metrics. Next, TD error should be minimized during sample design by choosing a representative spatio-temporal collection strategy, by using spatially and temporally relevant imagery and ancillary data sources during TD creation, and by selecting a set of legend definitions supported by the data. Furthermore, TD error can be minimized during the collection of individual samples by using consensus-based collection strategies, by directly comparing interpreted training observations against expert-generated training reference data to derive TD error metrics, and by providing image interpreters with thorough application-specific training. We strongly advise that TD error is incorporated in model outputs, either directly in bias and variance estimates or, at a minimum, by documenting the sources and implications of error. TD should be fully documented and made available via an open TD repository, allowing others to replicate and assess its use. To guide researchers in this process, we propose three tiers of TD error accounting standards. Finally, we advise researchers to clearly communicate the magnitude and impacts of TD error on map outputs, with specific consideration given to the likely map audience. [ABSTRACT FROM AUTHOR]

The aim of this study is to improve the understanding of land changes in the Jiulong River watershed, a coastal watershed of Southeast China. We developed a stratified classification methodology for land mapping, which combines linear stretching, an Iterative Self-Organizing Data Analysis (ISODATA) clustering algorithm, and spatial reclassification. The stratified classification for 2002 generated less overall error than an unstratified classification. The stratified classifications were then used to examine temporal differences at 1986, 1996, 2002, 2007 and 2010. Intensity Analysis was applied to analyze land changes at three levels: time interval, category, and transition. Results showed that land use transformation has been accelerating. Woodland's gains and losses were dormant while the gains and losses of Agriculture, Orchard, Built-up and Bare land were active during all time intervals. Water's losses were active and stationary. The transitions from Agriculture, Orchard, and Water to Built-up were systematically targeting and stationary, while the transition from Woodland to Built-up was systematically avoiding and stationary.

Geospatial analysis and statistical analysis are coupled in this study to determine the dynamic linkage between landscape characteristics and water quality for the years 1996, 2002, and 2007 in a subtropical coastal watershed of Southeast China. The landscape characteristics include Percent of Built (%BL), Percent of Agriculture, Percent of Natural, Patch Density and Shannon's Diversity Index (SHDI), with water quality expressed in terms of COD(Mn) and NH(4)(+)-N. The %BL was consistently positively correlated with NH(4)(+)-N and COD(Mn) at time three points. SHDI is significantly positively correlated with COD(Mn) in 2002. The relationship between NH(4)(+)-N, COD(Mn) and landscape variables in the wet precipitation year 2007 is stronger, with R(2) = 0.892, than that in the dry precipitation years 1996 and 2002, which had R(2) values of 0.712 and 0.455, respectively. Two empirical regression models constructed in this study proved more suitable for predicting COD(Mn) than for predicting NH(4)(+)-N concentration in the unmonitored watersheds that do not have wastewater treatment plants. The calibrated regression equations have a better predictive ability over space within the wet precipitation year of 2007 than over time during the dry precipitation years from 1996 to 2002. Results show clearly that climatic variability influences the linkage of water quality-landscape characteristics and the fit of empirical regression models.