John Rogan

Measuring the Physical Composition of Urban Morphology Using Multiple Endmember Spectral Mixture Models

The application of multiple endmember spectral mixture analysis (MESMA) to map the physical composition of urban morphology using Landsat Thematic Mapper (TM) data is evaluated and tested. MESMA models mixed pixels as linear combinations of pure spectra, called endmembers, while allowing the types and number of endmembers to vary on a per-pixel basis. A total of 63 two-, three-, and four-endmember models were applied to a Landsat TM image for Los Angeles County, and a smaller subset of these models was chosen based on fraction and root-mean-squared error (RMSE) criteria. From this subset, an optimal model was selected for each pixel based on optimization for maximum area coverage. The resultant endmember fractions were then mapped into four main components of urban land cover: Vegetation, Impervious surfaces, Soil, and Water/Shade. The mapped fractions were validated using aerial photos. The results showed that a majority of the image could be modeled successfully with two- or three-endmember models. The validation results indicated the robustness of MESMA for deriving spatially continuous variables quantified at the sub-pixel level. These parameters can be readily integrated into a wide range of applications and models concerned with physical, economic, and/or sociodemographic phenomena that influence the morphological patterns of the city.

Contextual land-cover classification: incorporating spatial dependence in land-cover classification models using random forests and the Getis statistic

Land-cover characterization of large heterogeneous landscapes is challenging because of the confusion caused by high intra-class variability and heterogeneous landscape artefacts. Neighbourhood context can be used to supplement spectral information, and a novel way of incorporating spatial dependence in a heterogeneous region is tested here using an ensemble learning technique called random forests and a measure of local spatial dependence called the Getis statistic. The overall Kappa accuracy of the random forest classifier that used a combination of spectral and local spatial (Getis) variables at three different neighbourhood sizes (3 × 3, 7 × 7, and 11 × 11) ranged from 0.85 to 0.92. This accuracy was higher than that of a non-spatial random forest classifier having an overall Kappa accuracy of 0.78, which was run using the spectral variables only. This study demonstrated that the use of the Getis statistic with different neighbourhood sizes leads to substantial increase in per class classification accuracy of heterogeneous land-cover categories.

Global Trends in Seasonality of Normalized Difference Vegetation Index (NDVI), 1982–2011

A 30-year series of global monthly Normalized Difference Vegetation Index (NDVI) imagery derived from the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI3g archive was analyzed for the presence of trends in changing seasonality. Using the Seasonal Trend Analysis (STA) procedure, over half (56.30%) of land surfaces were found to exhibit significant trends. Almost half (46.10%) of the significant trends belonged to three classes of seasonal trends (or changes). Class 1 consisted of areas that experienced a uniform increase in NDVI throughout the year, and was primarily associated with forested areas, particularly broadleaf forests. Class 2 consisted of areas experiencing an increase in the amplitude of the annual seasonal signal whereby increases in NDVI in the green season were balanced by decreases in the brown season. These areas were found primarily in grassland and shrubland regions. Class 3 was found primarily in the Taiga and Tundra biomes and exhibited increases in the annual summer peak in NDVI. While no single attribution of cause could be determined for each of these classes, it was evident that they are primarily found in natural areas (as opposed to anthropogenic land cover conversions) and that they are consistent with climate-related ameliorations of growing conditions during the study period.

An evaluation of bagging, boosting, and random forests for land-cover classification in Cape Cod, Massachusetts, USA

The iterative and convergent nature of ensemble learning algorithms provides potential for improving classification of complex landscapes. This study performs land-cover classification in a heterogeneous Massachusetts landscape by comparing three ensemble learning techniques (bagging, boosting, and random forests) and a non-ensemble learning algorithm (classification trees) using multiple criteria related to algorithm and training data characteristics. The ensemble learning algorithms had comparably high accuracy (Kappa range: 0.76-0.78), which was 11% higher than that of classification trees. Ensemble learning techniques were not influenced by calibration data variability, were robust to one-fifth calibration data noise, and insensitive to a 50% reduction in calibration data size.

Mapping Selective Logging in Mixed Deciduous Forest: A Comparison of Machine Learning Algorithms

This study assesses the performance of five Machine Learning Algorithms (MLAs) in a chronically modified mixed deciduous forest in Massachusetts (USA) in terms of their ability to detect selective timber logging and to cope with deficient reference datasets. Multitemporal Landsat Enhanced Thematic Mapperplus (ETM+) imagery is used to assess the performance of three Artificial Neural Networks ‐ Multi-Layer Perceptron, ARTMAP, Self-Organizing Map, and two Classification Tree splitting algorithms: gini and entropy rules. MLA performance evaluations are based on susceptibility to reduced training set size, noise, and variations in the training set, as well as the operability/transparency of the classification process. Classification trees produced the most accurate selective logging maps (gini and entropy rule decision tree mean overall map accuracy � 94 percent and mean per-class kappa of 0.59 and 0.60, respectively). Classification trees are shown to be more robust and accurate when faced with deficient training data, regardless of splitting rule. Of the neural network algorithms, self-organizing maps were least sensitive to the introduction of noise and variations in training data. Given their robust classification capabilities and transparency of the classselection process, classification trees are preferable algorithms for mapping selective logging and have potential in other forest monitoring applications.

Mapping seasonal trends in vegetation using AVHRR-NDVI time series in the Yucatán Peninsula, Mexico

This research examines the spatio-temporal trends in Advanced Very High Resolution Radiometer (AVHRR) Global Inventory Modelling and Mapping Studies (GIMMS) normalized difference vegetation index (NDVI) time series to ascribe land use change and precipitation to observed changes in land cover from 1982 to 2007 in the Mexican Yucatán Peninsula, using seasonal trend analysis (STA). In addition to discrete land cover transitions across the study region, patterns of agricultural intensification, urban expansion and afforestation in protected areas have enacted changes to the seasonal patterns of apparent greenness observed through STA greenness parameters. The results indicate that the seasonal variation in NDVI can be used to distinguish among different land cover transitions, and the primary differences among these transitions were in changes in overall greenness, peak annual greenness and the timing of the growing season. Associations between greenness trends and precipitation were weak, indicating a human-dominated system for the 26 years examined. Changes in the states of Campeche, Quintana Roo and Yucatán appear to be associated with pasture cultivation, urban expansion-extensive cultivation and urban expansion-intensive cultivation, respectively.

Mapping Burn Severity of Mediterranean-Type Vegetation Using Satellite Multispectral Data

Knowledge of the spatial distribution of burn severity immediately following a fire is needed to locate areas requiring management for environmental impacts and timber salvage, and for validation of fire risk and fire behavior models. We evaluated methods for mapping post-fire burn severity in southern California Mediterranean-type ecosystems using satellite images calibrated and validated by field-collected data. The effects of spectral transforms, temporal dimensionality, classifiers, and sensor type on the accuracy of burn severity classification were analyzed. We mapped and analyzed the distributions of five categories of burn severity or land cover for two southern California wildfires based primarily on classification of Landsat TM/ETM+ data, with IKONOS MS data also being evaluated. Map accuracy was assessed relative to field-based classification of burn severity of randomly located plots, using the Composite Burn Index approach. Maps based on the multi-temporal Kauth Thomas transform of Landsat TM/ETM+ data and maximum likelihood classifier had the highest overall accuracy (64 and 55%) and kappa values (0.51 and 0.37) for the two study areas. Forested lands were classified at a much higher level of accuracy (overall accuracy near 80%), while accurate classification of burn severity in shrublands was more challenging (overall accuracy less than 50%). The lower stature vegetation of shrublands typically experiences crown-burning fires, such that range of burn severity for shrublands is more limited.

InSAR detects increase in surface subsidence caused by an Arctic tundra fire

Wildfire is a major disturbance in the Arctic tundra and boreal forests, having a significant impact on soil hydrology, carbon cycling, and permafrost dynamics. This study explores the use of the microwave Interferometric Synthetic Aperture Radar (InSAR) technique to map and quantify ground surface subsidence caused by the Anaktuvuk River fire on the North Slope of Alaska. We detected an increase of up to 8 cm of thaw-season ground subsidence after the fire, which is due to a combination of thickened active layer and permafrost thaw subsidence. Our results illustrate the effectiveness and potential of using InSAR to quantify fire impacts on the Arctic tundra, especially in regions underlain by ice-rich permafrost. Our study also suggests that surface subsidence is a more comprehensive indicator of fire impacts on ice-rich permafrost terrain than changes in active layer thickness alone.

A step-wise land-cover classification of the tropical forests of the Southern Yucatan, Mexico

Analysis of land-cover change in the seasonal tropical forests of the Southern Yucatán, Mexico presents a number of significant challenges for the fine-scale land-cover information required of land-change science. Subtle variation in mature forest types across the regional ecocline is compounded by vegetation transitions following agricultural land uses. Such complex mapping environments require innovation in multispectral classification methodologies. This research presents an application of a step-wise maximum likelihood/In-Process Classification Assessment (IPCA) procedure. This hybrid supervised and unsupervised classification methodology allows for exploration of underlying characteristics of Landsat Thematic Mapper (TM) imagery in tropical environments. Once spectrally separable classes have been identified, field data then determine the ecological definition of vegetation types with special attention paid to areas of unknown or mixed classes. A post-field assessment re-classification using the Dempster–Shafer method reduced the original 25 spectral classes to 14 ecologically distinctive classes, providing the fine-tuned land-cover distinctions that are required for both environmental and socioeconomic research questions. The overall map accuracy was 87% with an average per-class accuracy of 86%. Per-class accuracy ranged from as low as 45% for pasture grass to a high of 100% for tall-stature evergreen upland forest, low and medium-stature semi-deciduous upland forest and deciduous forest.

Hurricane disturbance mapping using MODIS EVI data in the southeastern Yucatán, Mexico

This letter evaluated the use of Moderate Resolution Imaging Spectroradiometer (MODIS) 250 m Enhanced Vegetation Index (EVI) standard product data (MOD/MYD13Q1 C5) to map the damage caused by Hurricane Dean (August 2007) to the forests in the Yucatán Peninsula of Mexico using a two-step vetting procedure. Sequences of MODIS EVI 16-day composite products captured before and after the hurricane were compared against 93 field damage plots to select an appropriate set of pre- and post-damage data. Aqua pairwise combinations produced the highest damage detection overall accuracy compared with Terra (82.4% vs. 73.8%, respectively) because of advantageous timing of the Aqua EVI compositing, relative to the hurricane event. The most accurate EVI damage map (91.4% overall) revealed highest damage detection in Saffir–Simpson hurricane wind scale zone 5 (i.e. wind speed >248 km h−1, i.e. 95%), followed by 93% in zone 4 (210–249 km h−1) and 87% in zone 3 (178–209 km h−1). Results indicate that MODIS EVI products provide timely and accurate maps of hurricane damage in subtropical forests.