Active Research Projects
PhD Projects
Land Use Land Cover Classification
The term Land Use and Land Cover (LULC) typically falls under the term land use classification. However, land use and Land Cover are two seperate themes which are addressed under land-use classification. Some classification can have singel level level (e.g, two-class grassland or developed areas) while other may have multiple levels of classification. One example of latter is National Land Cover Dataset NLCD. Often time, these LULC products are based on coarse spatial resolution data. While these datasets offer great deal of information in policy making, ofthen times at local level thsese data doesn’t align well with ground features. One way to deal with this problem is to use high-resolution remote sensing products.
The project’s primary goal is to investigate methodological approaches for mapping grassland-dominated landscapes using very high-resolution (VHR) remote sensing imagery over large areas. The goal is to evaluate methods to optimize the computational cost and improve the classification accuracy of land use and land cover products generated using VHR remote sensing. The research is mainly focused on analyzing orthoimagery acquired by the United States Department of Agriculture (USDA) through the National Agricultural Imagery Program (NAIP) at < 2 m pixel resolution and light detection and ranging (LiDAR) data. NAIP orthoimages are mainly collected during the agricultural growing seasons across the conterminous United States with a return interval of two to three years. The USGS 3D elevation program aims to complete the acquisition of nationwide LiDAR by 2023. In Texas, LiDAR data is available for almost all counties. In order to harness the power of high spatial resolution datasets, I mainly explore geographic object-based image analysis (GEOBIA) approaches using NAIP and LiDAR data. The analytical framework aims to answers the following questions:
a) Do supervised machine learning algorithms using the GEOBIA framework in image classification on NAIP imagery improve land use and land cover map accuracy compared to traditional classification algorithms?
b) Do LiDAR-derived metrics improve classification accuracy over multispectral NAIP imagery?
c) Do higher prediction accuracy derived from machine learning algorithms ensure high spatial accuracy?
d) Can GEOBIA be applied over time series of NAIP imagery to understand changes in land use and land cover? If so, how do these products compare to Landsat-based products?