30 October 2019 in International Journal of Geographical Information Science
(2019). GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge discovery and beyond. International Journal of Geographical Information Science. Ahead of Print.
29 September 2007 in GeoJournal
22 June 2018 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Establishing up-to-date large scale building maps is essential to understand the urban dynamics, such as estimating population, urban planning, and many other applications. Although many computer vision tasks have been successfully carried out with deep convolutional neural networks, there is a growing need to understand their large scale impact on building mapping with remote sensing imagery. Taking advantage of the scalability of convolutional neural networks (CNNs) and using only few areas with the abundance of building footprints, for the first time we conduct a comparative analysis of four state-of-the-art CNNs for extracting building footprints across the entire continental United States. The four CNN architectures namely: Branch-out CNN, fully convolutional network (FCN), conditional random field as recurrent neural network (CRFasRNN), and SegNet, support semantic pixelwise labeling and focus on capturing textural information at multiscale. We use 1-meter resolution aerial images from National Agriculture Imagery Program as the test-bed, and compare the extraction results across the four methods. In addition, we propose to combine signed-distance labels with SegNet, the preferred CNN architecture identified by our extensive evaluations, to advance building extraction results to instance level. We further demonstrate the usefulness of fusing additional near IR information into the building extraction framework. Large scale experimental evaluations are conducted and reported using metrics that include: Precision, recall rate, intersection over union, and the number of buildings extracted. With the improved CNN model and no requirement of further postprocessing, we have generated building maps for the United States with an average processing time less than one minute for an area of size $\sim \!\!{\text{56}}$ $\text{km}^2$. The quality of extracted buildings and processing time demonstrated that the proposed CNN based framework fits the need of building extraction at scale.
1 February 2001 in Journal of Water Resources Planning and Management
At a watershed scale, land-use change can increase runoff, flooding, and nonpoint source pollution and degrade downstream water bodies. Thus it is important to assess the potential hydrologic impacts of land-use change prior to watershed development. The L-THIA (Long-Term Hydrologic Impact Assessment) model is a tool to initially assess how land-use change affects annual average runoff and is based only on readily available data. Because L-THIA is relatively new, it is important to test it against other, well-accepted methods. The U.S. Environmental Protection Agency's SWMM (Storm Water Management Model), a well-known and widely used model, was used to perform runoff calculations for comparison with L-THIA. Applications of L-THIA and SWMM to two small watersheds in Chicago show that L-THIA predicts annual average runoff between 1.1 and 23.7% higher than SWMM. The agreement between the results is higher for larger watersheds. Both models predict a linear relationship between average annual runoff and incre...
1 January 2018 in Remote Sensing of Environment
Although remote sensing has long been used to aid in the estimation of population, it has usually been in the context of spatial disaggregation of national census data, with the census counts serving both as observational data for specifying models and as constraints on model outputs. Here we present a framework for estimating populations from the bottom up, entirely independently of national census data, a critical need in areas without recent and reliable census data. To make observations of population density, we replace national census data with a microcensus, in which we enumerate population for a sample of small areas within the states of Kano and Kaduna in northern Nigeria. Using supervised texture-based classifiers with very high resolution satellite imagery, we produce a binary map of human settlement at 8-meter resolution across the two states and then a more refined classification consisting of 7 residential types and 1 non-residential type. Using the residential types and a model linking them to the population density observations, we produce population estimates across the two states in a gridded raster format, at approximately 90-meter resolution. We also demonstrate a simulation framework for capturing uncertainty and presenting estimates as prediction intervals for any region of interest of any size and composition within the study region. Used in concert with previously published demographic estimates, our population estimates allowed for predictions of the population under 5 in ten administrative wards that fit strongly with reference data collected during polio vaccination campaigns.