Jordan Graesser

Cropland/pastureland dynamics and the slowdown of deforestation in Latin America

Latin America has the planet’s largest land reserves for agriculture and had the most rapid agricultural expansion during the twenty-first century. A large portion of the expansion replaced forests, as shown by many local and regional studies. However, expansion varied regionally and also replaced other land covers. Further, it is important to distinguish between changes in cropland and pastureland as they produce food at different levels of efficiency and intensity. We used thirteen years (2001–2013) of MODerate Resolution Imaging Spectroradiometer satellite imagery to characterize cropland and pastureland expansion at multiple scales across Latin America. From 2001 to 2013, 17% of new cropland and 57% of new pastureland replaced forests throughout Latin America. Cropland expansion from 2001 to 2013 was less (44.27 Mha) than pastureland (96.9 Mha), but 44% of the 2013 cropland total was new cropland, versus 27% of the 2013 pastureland total, revealing higher regional expansion rates of row crop agriculture. The majority of cropland expansion was into pastureland within core agricultural regions of Argentina, Brazil, Bolivia, Paraguay, and Uruguay. On the contrary, pastureland largely expanded at frontiers, such as central Brazil, western Paraguay, and northern Guatemala. As others have suggested, regional agriculture is strongly influenced by globalization. Indeed, we find an overall decrease in agricultural expansion after 2007, coinciding with the global economic slowdown. The results illustrate agricultural cropland and pastureland expansion across Latin America is largely segregated, and emphasize the importance of distinguishing between the two agricultural systems, as they vary in land use intensity and efficiency.

Spatial refinement of census population distribution using remotely sensed estimates of impervious surfaces in Haiti

Previous research on the relationship between impervious surfaces and population has focused on limited areas. This paper considers the relationship for an entire country. Multiple spatial resolution optical imagery was integrated with census data to refine the spatial distribution of population in Haiti. A classification and regression tree (CART) methodology was used to create a percent impervious-area layer based on ten high-resolution training chips and a Landsat Enhanced Thematic Mapper (ETM+) mosaic. This estimate then became an input for a dasymetric mapping approach, where population was distributed proportionately to impervious areas with input from ancillary data sources. Haitis section boundaries (third administrative level) provided the mapping base because a strong relationship between imperviousness and population was observed at this administrative level. The accuracy of the technique was tested for a set of 110 municipalities in Haiti, making use of recent census data. The potential usefulness of imagery-based population estimates for areas where census data do not exist was also tested by dividing Haiti into northern and southern portions and using one to predict the other.

Mapping slums using spatial features in Accra, Ghana

In order to map the spatial extent and location of slum settlements multiple methodologies have been devised including remote sensing based methods and field based methods using surveys and census data. In this study we utilize spatial, structural, and contextual features (e.g., PanTex, Histogram of Oriented Gradients, Line Support Regions, Hough transforms and others) calculated at multiple spatial scales from high spatial resolution satellite data to map slum areas and compare these estimates to three field based slum maps: one from the UN Habitat/Accra Metropolitan Assembly (UNAMA) and two census data derived maps based on the UN Habitat definition of a slum, a simple slum/non-slum dichotomy map, and a slum index map. When comparing the remotely sensed derived slum areas to the UNAMA slum definition results indicate an overall accuracy of 94.3% and a Kappa of 0.91. When compared to the dichotomous, census derived slum maps the results are not as accurate. This reduced accuracy is due to the substantial over prediction of slums, especially if only one criterion was missing, using the census data. In relation to the slum index, the remote sensing estimates of slums were significantly correlated with an r2 of 0.45 and when population density was taken into account, the correlation increased to an r2 of 0.78. Overall, the remote sensing methodology provides a reasonable estimate of slum areas and variations within the city.

Complex settlement pattern extraction with multi-instance learning

Per-pixel (or single instance) based classification schemes which have proven to be very useful in thematic classification have shown to be inadequate when it comes to analyzing very high resolution remote sensing imagery. The main problem being that the pixel size (less than a meter) is too small as compared to the typical object size (100s of meters) and contains too little contextual information to accurately distinguish complex settlement types. One way to alleviate this problem is to consider a bigger window or patch/segment consisting a group of adjacent pixels which offers better spatial context than a single pixel. Unfortunately, this makes per-pixel based classification schemes ineffective. In this work, we look at a new class of machine learning approaches, called multi-instance learning, where instead of assigning class labels to individual instances (pixels), a label is assigned to the bag (all pixels in a window or segment). We applied this multi-instance learning approach for identifying two important urban patterns, namely formal and informal settlements. Experimental evaluation shows the better performance of multi-instance learning over several well-known single-instance classification schemes.

Assessing the relationship between spatial features derived from high resolution satellite imagery and census variables in Accra, Ghana

People living within cities tend to live near others who have similar characteristics such as socioeconomic status, religion or ethnicity. Generally the characteristics of these spatial groups manifest themselves at the neighborhood level and they live in similar dwellings. In this study we examine the ability to use spatial features and spectral information, derived from high spatial resolution satellite data, to map variations in dwellings to determine the amount of information we can derive about the people living within the city of Accra, Ghana. The spatial features examined are line support regions, PanTex, Histogram of Oriented Gradients, Local Binary Patterns, Hough and Fourier transforms, and the Normalized Difference Vegetation Index (NDVI). Results indicate that there are strong correlations between multiple spatial features and census variables at the neighborhood scale. This indicates that the spatial information derived from imagery can be used to map population characteristics at within city spatial scales.

Spatial patterns of DOC concentration and DOM optical properties in a Brazilian tropical river‐wetland system

The Cerrado (savanna) and Pantanal (wetland) biomes of central-western Brazil have experienced significant development activity in recent decades, including extensive land cover conversion from natural ecosystems to agriculture and urban expansion. The Cuiaba River transects the Cerrado biome prior to inundating large areas of the Pantanal, creating one of the largest biodiversity hotspots in the world. We measured dissolved organic carbon (DOC) and the optical absorbance and fluorescence properties of dissolved organic matter (DOM) from 40 sampling locations spanning Cerrado and Pantanal biomes during wet and dry seasons. In the upper, more agricultural region of the basin, DOC concentrations were highest in the rainy season with more aromatic and humified DOM. In contrast, DOC concentrations and DOM optical properties were more uniform for the more urbanized middle region of the basin between wet and dry seasons, as well as across sample locations. In the lower region of the basin, wet season connectivity between the river and the Pantanal floodplain led to high DOC concentrations, a four-fold increase in HIX (an indicator of DOM humification) and a 50% reduction in the spectral slope (SR). Basin-wide, wet season values for SR, HIX and FI (fluorescence index) indicated an increasing representation of terrestrially derived DOM that was more humified. Parallel factor analysis (PARAFAC) identified two terrestrially derived components (C1 and C2) representing 77% of total fluorescing DOM (fDOM). A third, protein-like fDOM component increased markedly during the wet season within the more urban-impacted region.

Contextural feature evaluation of multi-resolution imagery

Spatial information pertaining to the neighborhood mapping can only be available at the scales of centimeters to meters. In this study, we utilize spatial, structural and contextural features calculated at multiple spatial scales to assess feature-based urban mapping using two different high-spatial resolution sensors (SPOT versus Quickbird) and evaluate the difference in describing neighborhood at these scales. We compared the features that include PanTex, Histogram of Oriented Gradients, Line Support Regions, Hough transformations and others on two selected images that were orthorectified and coregistered. We ran a classification based on the stacks of features for each of the two panchromatic high-spatial resolution images. Our results showed that without using vegetation index as an input, SPOT and Quickbird accomplished comparable classification results, while adding NDVI in Quickbird classification, SPOT image alone does not accomplish comparable classification accuracy using Quickbird image mainly due to lack of vegetation information represented by Normalized Difference Vegetation Index (NDVI), which correlates substantial difference among neighborhoods. The future work can incorporate other multi-spectral information into classification while using SPOT imagery.