David P. Lusch

Supervised classification of types of glaciated landscapes using digital elevation data

Abstract Automated approaches for identifying different types of glaciated landscapes using digitally processed elevation data were evaluated. We tested the ability of geomorphic measures (e.g. elevation, relative relief, roughness, and slope gradient) derived from digital elevation models (DEMs) to differentiate glaciated landscapes using maximum likelihood classification and artificial neural networks (ANN). The automated methods were trained and validated using an existing Quaternary geology map and a manual interpretation of the contour data portrayed on topographic quadrangles. The need for such methods arises from efforts to classify types of landscapes (e.g. ecoregions) in Michigan. One fundamental control of the landscape structure in Michigan, including soil type and vegetation, is the underlying sedimentary and landform assemblages produced by an array of glacial processes during the waning phase of the Pleistocene. Traditional methods for identifying different landscapes (e.g. ice-contact landscapes, stagnation landscapes) have relied on printed topographic maps and have been very effective, but time consuming. The maps resulting from the four supervised classification trials had between 51% and 61% agreement with the original Quaternary geology map. The output from the maximum likelihood classification had slightly higher agreements than the output from the neural net, which is attributed to the generalization inherent in the Quaternary geology map compared with the nature of the classifier for the neural net. The neural net, however, identifies significant detail and non-linear relationships between classification inputs and output classes. Future work should incorporate a map of soils into the classification.

A dynamic species distribution model of Glossina subgenus Morsitans: The identification of tsetse reservoirs and refugia

Tsetse flies are the primary vector for African trypanosomiasis, a neglected tropical disease that affects both humans and livestock across the continent of Africa. In 1973 tsetse were estimated to inhabit 22% of Kenya; by 1996 that number had risen to roughly 34%. Efforts to control the disease are hampered by a lack of information and costs associated with the identification of infested areas. To aid control efforts we have constructed the Tsetse Ecological Distribution Model (TED Model). The TED Model is a raster based dynamic species distribution model that predicts tsetse distributions at 250 m spatial resolution, based on habitat suitability and fly movement rates, at 16-day intervals. Although the TED Model can be parameterized to any tsetse subgenus/species requirements, for the purpose of this study the TED Model was parameterized to identify suitable habitat for Glossina subgenus Morsitans. Using the TED Model we have identified where and when Glossina subgenus Morsitans populations should be co...

Effects of winter temperatures on gypsy moth egg masses in the Great Lakes region of the United States

Accurate prediction of winter survival of gypsy moth (Lymantria dispar L.) eggs and phenology of egg hatch in spring are strongly dependent on temperature and are critical aspects of gypsy moth management programs. We monitored internal temperatures of egg masses at three heights aboveground level and at the four cardinal aspects on oak tree stems at two different locations in Michigan during the winter seasons of 1997/1998, 1998/1999 and 1999/2000. Effects of aspect were more strongly associated with observed egg mass temperatures than height above the ground surface. Instantaneous differences between egg mass temperatures on sunny days were as high as 30 ◦ C greater on the southern aspect vs. egg mass temperatures on other aspects, resulting in substantial differences in pre-hatch growing degree accumulations for egg masses on a single tree. Egg masses on southern and western aspects, where solar loading and temperatures were greatest, experienced substantial mortality. Mean survival of eggs averaged across the three seasons was less than 25% on southern and western aspects, compared with averages of 53 and 73% on eastern and northern aspects, respectively. Linear regression of mean monthly egg mass–air temperature differences (between north and south aspects) and mean daily solar flux density resulted in slope coefficient estimates of 0.13 and 0.21 ◦ CM J m −2 , and correlation coefficients of 0.81 and 0.82 at the two field locations, respectively. Using a simple egg mass hatch phenological model, such dissimilarities in temperature and growing degree day accumulation resulted in differences in estimated egg hatch dates of up to 25 days. Snow cover moderated egg mass temperature, with extreme seasonal minimum winter temperatures under snow cover as much as 7.1 ◦ C warmer than those

Developing land use/land cover parameterization for climate–land modelling in East Africa

Regional climate modeling studies now have numerous choices in selecting land use/land cover (LULC) products to provide land surface parameter information. The various LULC products were developed with different objectives, methods and data sources. Not all new LULC products have land classes that match the land class types defined in climate models. More importantly, when used in regional climate models, simulation results can vary significantly depending on the LULC products. Thus, developing appropriate LULC parameterization for climate models becomes critical depending on objectives and efforts. The objective of this paper is to develop the most accurate LULC scheme possible for East Africa for implementation in the Regional Atmospheric Modeling System (RAMS). A crosswalk procedure, based on assessments of various LULC products, was performed connecting land class types in RAMS and the newly created LULC scheme. No simulations are discussed here; rather, we present an outline of the procedures that were carried out to take advantage of the strengths of currently available LULC products, Africover and Global Land Cover 2000, for the purpose of conducting regional climate simulations.

An assessment of africover and GLC2000 using general agreement and videography techniques

Remote sensing science has grown to provide a range of land use land cover datasets. These datasets range in creation methodologies, objectives, and validation levels. This study evaluates Africover and Global Land Cover for the year 2000 (GLC) land use land covers for East Africa as land use land cover input for climate and land modeling research. Africover was produced from multi-temporal Landsat data under a relatively new classification scheme primarily using visual interpretation methods. GLC utilized the VEGETATION instrument onboard the Satellite Pour l’Observation de la Terre (SPOT) 4 using a similar classification scheme. A two-tiered assessment approach examining general class agreement and airborne videography techniques were performed. Differences between the products show moderate general agreement with a wide range of class level agreement. The video data was useful in developing both qualitative and quantitative assessment measures. Agricultural land use dynamics were particularly problematic causing much of the misclassification. The general objective was to assess and evaluate the land covers and their potential applications. Introduction Land use land cover change is a fundamental component of natural resource and global change research. These data represent a range of biophysical parameters and socioeconomic indicators thusly detailed and accurate land use land cover (LULC) information are vital to scientists, managers, and decision makers. The availability and quality of regional to continental scale remotely sensed LULC products have increased greatly in the past decade. At the same time remote sensing science and computational ability advances have made the incorporation of more advanced land cover products and statistical techniques possible in LULC mapping and monitoring. The capabilities and level of detail for regional to global modeling studies will benefit immensely from newer LULC data. When created by different agencies or groups, LULC datasets can often differ in temporal coverage, spatial resolutions, objectives, classification methodologies, and validation levels. This makes direct comparisons challenging and can limit applications. A range of assessment techniques provide approaches to deal with evaluation concerns. In this study assessments are needed to examine dataset validity and utility, illustrate strengths and weaknesses, and determine the overall quality and accuracy of the information. Numerous studies and methods have been implemented for assessing LULC products. These include both qualitative and quantitative evaluations. Two general methods of assessment, all with variations, are abundant throughout the literature. The first method is becoming more common and is identified as appropriate to evaluate larger scale LULC data. This method is a comparison of the levels of agreement/disagreement between general land categories of two generated LULC products. The second method is to examine a remotely sensed product against ground truthed or reference data, often finer resolution. Aerial video data capture, or videography, for land cover and vegetation condition assessment has proven to be one such useful method. Advances in digital technology and computer integration have increased the applicability of airborne video as a remote sensing tool. There are distinct logistical advantages to utilizing airborne video data in isolated and environmentally sensitive regions where there is limited preexisting aerial photography and poor infrastructure and where ground accessibility is difficult and expensive [1]. Study Site The study site is located in East Africa. For this assessment a two tiered study site was developed. The first level of assessment includes the countries of Burundi, Kenya, Rwanda, Uganda, and Tanzania. The second level of the assessment includes two aerial flight paths located in central and southern Kenya. Within the study sites, a wide range of LULC are present. Biophysical vegetation cover ranges from expansive savannas to dense forest to riparian wetlands. Land uses include a variety of human activities from intense agriculture to open cattle ranching to conservation. Data Africover is a project administered by the United Nations Food and Agricultural Organization (UN FAO) that was approved in 1994. Africover was initiated in response to requests to develop information on natural resources required at national and regional levels. The purposes range from disaster and early warning systems to agriculture and food security to biodiversity and global change research [2]. Africover methodology included combining digital automatic classification with traditional visual interpretation at 1:200:000 scale off Landsat satellite imagery. For smaller countries and specific areas 1:100,000 scale was used. A 21class regional aggregate for East Africa was evaluated for the study region. 5005 0-7803-9050-4/05/

Glacial history of Michigan, U.S.A.: A regional perspective

20.00 ©2005 IEEE. 5005 Africover needs assisted in generating the Land Cover Classification System (LCCS) developed by the UN FAO. The LCCS is a hierarchical, priori system that in theory can be applied to any region of the globe for comparison purposes [3]. The categorical assignment includes broad sweeping LULC classes followed by a series of descriptive characteristics. Determining parameters include vegetation or non-vegetated surfaces, terrestrial or aquatic, cultivated and managed, natural and semi-natural, life-form, cover layer, feature height, spatial distribution, leaf type and phenology. Global Land Cover 2000 is available from the Joint Research Centre’s Global Vegetation Monitoring Unit. The overall objective in creating GLC was to provide a harmonized land cover database for the globe as part of the Millennium Ecosystem Assessment. The database was designed to serve users from science programs, policy makers, environmental convention secretariats, international and non-governmental organizations and development-aid projects [4]. More than 30 partner institutions participated in the project utilizing 14 months of pre-processed VEGETATION sensor data via the System Pour l’observation de la Terre (SPOT) 4 satellite. The data used was collected between 11/1/1999 – 12/31/2000. The classification methodology was based off the UN FAO LCCS for compatibility purposes. A continental Africa version was evaluated here. Methods For the general agreement analysis each land cover class was aggregated into one of nine general land cover classes. Using the LCCS scheme, aggregating to broad classes for comparison purposes is systematic. However, the classifications were performed by different multi-teamed groups at different scales possibly resulting in different interpretations. The videography was recorded onboard an aircraft flown at a mean altitude of 1000 meters above sea level in order to capture land cover at fine scale. A Global Positioning System recorded location information during the flight directly to the footage. Video tape data was transferred digitally at a rate of 30-frames per second via an IEEE 1394 FireWire cable. The capture resolution was at 720x480 in accordance with National Television System Committee standards. Approximately five hours was recorded over the 900 kilometers traveled. The field of view (FOV) was calculated in order to geolink the video data to the LULC data. The GPS points were re-sampled to a representative level for accurately depicting the airplane’s flight path over the landscape. The GPS points were treated as network nodes and snapped together to form a polyline flight path. Two flight line FOV shapefiles were created and merged together as an end result. The land covers were clipped using the flight line FOV shapefile. Area estimates were generated to identify classes captured and assist in sampling design. The FOV area was calculated to cover approximately 12.6 hectares per frame. The number of possible FOVs, or sampling units in this case, was calculated by dividing the area of each class by the FOV area. For Africover, this resulted in a range of possible samples from 11 for irrigated agriculture to 772 for tree and shrub savanna. The number of possible FOVs for GLC ranged from 8 for sandy desert and dunes to 844 for open grassland with sparse shrubs. The georegistered digital video was then scanned through finding a stratified random sampling scheme of each GPS point to determine whether that point was correctly or incorrectly classified. The interpretation process included three individuals judging a sample point together. A sampling point, or area in a frame in which that point was located, was determined whether the classification was correct or incorrect. Results Using the LCCS framework both GLC and Africover classes were aggregated into broad land categories to examine general agreement/disagreement. All classes were grouped into the broad level classes of forest, woodland/shrubland, grassland, agriculture, barren, water, and urban. The overall agreement between GLC and Africover at general levels using the LCCS framework is approximately 41% calculated from the confusion table (Table 1). Class agreement was calculated for each land category. Measuring Africover against GLC, class agreement ranged between 0.3% 90%. Class agreements levels for forest, woodland/shrubland, grassland, agriculture, barren, water, and urban were 22%, 22%, 58%, 37%, 0.32%, 90%, and 11% respectively. Measuring GLC against Africover for forest, woodland/shrubland, grassland, agriculture, barren, water, urban class agreement produced 21%, 16%, 46%, 54%, 13%, 92%, and 63% respectively. Using the video footage to build an error matrix for Africover, for the flightlines sampled, an overall accuracy of 55% (213/391) was calculated. No correct sample points were identified for open to closed grassland. Aquatic agriculture which had a small regional set of samples, was identified as mostly correct. Closed trees also had relatively high accuracy.

Mapping the physiography of Michigan with GIS

Publisher Summary This chapter discusses the glacial history of Michigan, U.S.A. At its maximum, the Laurentide Ice Sheet in the Great Lakes region, comprised the Green Bay, Lake Michigan, Saginaw and Erie lobes, extended south of Michigan into central Ohio, Northwestern Indiana, and North-Eastern Illinois. The Saginaw lobe, the thinnest of the four lobes, melted the fastest uncovering south-central Lower Michigan first. During the next 5500 years, continued recession was interrupted by minor ice margin readvances in the ice sheet that formed a series of end moraines across central Lower Michigan. While retreating, a series of proglacial lakes formed at the margin of the Laurentide Ice Sheet at sites where the land sloped towards the ice front. The earliest proglacial lakes to form in Lower Michigan were glacial Lake Chicago and Lake Maumee within the Lake Michigan and Erie basins. The lake level stabilised long enough to form three successively lower beach ridges. The chapter discusses that after the Lake Michigan and Saginaw lobes had melted Northward into Northern Lower Michigan, the Laurentide Ice Sheet surged back into Lake Michigan, Huron and Erie basins. This surge, called the Port Huron advance, resulted in the formation of glacial Lake Saginaw and Lake Whittlesey in the Lake Huron and Erie lake basins, respectively. The Port Huron end moraine, which marks this significant readvance, is a very prominent topographic feature throughout much of Michigan. Continued melting after the Port Huron advance, once again exposed the thumb of Michigan, leading to the formation of glacial Lake Warren around 12,800 B.P.

Conflating LiDAR data and multispectral imagery for efficient building detection

Abstract We present a new physiographic map of Michigan, that is also available interactively, online. Only four, small-scale physiographic maps of Michigan had been previously published. Our mapping project made use of a wide variety of spatial data, in a GIS environment, to visualize and delineate the physical landscape in more detail than has been done previously. We also examined many of the unit boundaries in the field, using a GIS running on a GPS-enabled laptop. Unlike previous physiographic maps, the online version of the map enables users to query the criteria used to define each of the 224 boundaries of its 10 major and 91 minor physiographic units. The interactive nature of the online version of the map is a unique enhancement to physiographic maps and mapping. Our study also provides data on the number and types of criteria used to define each of the 224 unit boundaries within the map. Most of our unit boundaries are based on data derived from 10-m raster elevation data and NRCS soils data, e.g., relief, soil wetness, escarpments, landscape fabric, and parent material characteristics. Data gleaned from NRCS SSURGO county-scale soil maps were a strength of the project. [Key words: Michigan, physiography, landforms, soils, GIS, mapping]

Feature extraction from 3D lidar point clouds using image processing methods

Abstract. Light detection and ranging (LiDAR) point cloud data can contain millions of point returns from a diverse range of surface features, and directly reconstructing buildings from these data is challenging. Trees and other vegetation pose a particular problem in many built environments. This paper investigates several efficient procedures for detecting buildings and excluding vegetation using LiDAR and imagery data. Two general approaches for identifying and filtering out returns from vegetation are investigated: the first uses a normalized difference vegetation index (NDVI) image, while the second uses height differences. The utility of an entropy filter for improving NDVI filter performance as well as two distinct approaches for height-difference modeling are also evaluated. All methods use efficient raster-based algorithms for filtering while retaining the high spatial precision of the vector LiDAR point returns. Following removal of nonbuilding points, remaining points are segmented into distinct building features. In addition, we place particular emphasis on the analysis of processing challenges and special cases as well as the accuracy of these different methods on a large-volume LiDAR dataset covering a challenging build environment.

Accounting for positional uncertainty in historical shoreline change analysis without ground reference information

Airborne LiDAR data have become cost-effective to produce at local and regional scales across the United States and internationally. These data are typically collected and processed into surface data products by contractors for state and local communities. Current algorithms for advanced processing of LiDAR point cloud data are normally implemented in specialized, expensive software that is not available for many users, and these users are therefore unable to experiment with the LiDAR point cloud data directly for extracting desired feature classes. The objective of this research is to identify and assess automated, readily implementable GIS procedures to extract features like buildings, vegetated areas, parking lots and roads from LiDAR data using standard image processing tools, as such tools are relatively mature with many effective classification methods. The final procedure adopted employs four distinct stages. First, interpolation is used to transfer the 3D points to a high-resolution raster. Raster grids of both height and intensity are generated. Second, multiple raster maps - a normalized surface model (nDSM), difference of returns, slope, and the LiDAR intensity map - are conflated to generate a multi-channel image. Third, a feature space of this image is created. Finally, supervised classification on the feature space is implemented. The approach is demonstrated in both a conceptual model and on a complex real-world case study, and its strengths and limitations are addressed.