Files in this item



application/pdfDUTTA-DISSERTATION-2016.pdf (66MB)Restricted to U of Illinois
(no description provided)PDF


Title:Data driven estimation of soil and vegetation attributes using airborne remote sensing
Author(s):Dutta, Debsunder
Director of Research:Kumar, Praveen
Doctoral Committee Chair(s):Kumar, Praveen
Doctoral Committee Member(s):Greenberg, Jonathan A; Liang, Feng; Di Girolamo, Larry
Department / Program:Civil & Environmental Eng
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Remote Sensing
Soil Properties
Light Detection and Ranging (LiDAR)
Leaf Area Density
Abstract:Airborne remote sensing using imaging spectroscopy and LiDAR (Light Detection and Ranging) measurements enable us to quantify ecosystem and land surface attributes. In this study we use high resolution airborne remote sensing to characterize soil attributes and the structure of vegetation canopy. Soil texture, organic matter, and chemical constituents are critical to ecosystem functioning, plant growth, and food security. However, most of the soil data available globally are of coarse resolutions at scales of 1:5 million and lack quantitative information for modeling and land management decisions at field or catchment scales. Thus the need for a spatially contiguous quantitative soil information is of immense scientific merit which can be obtained using airborne and space-borne imaging spectroscopy. Towards this goal we systematically explore the feasibility of characterizing soil properties from imaging spectroscopy using data driven modeling approaches. We have developed a modeling framework for quantitative prediction of different soil attributes using airborne imaging spectroscopy and limited field soil grab sample datasets. The results of our analysis using fine resolution (7.6m) Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data collected over midwestern United States immediately after the large 2011 Mississippi River flood indicate the feasibility of using the developed models for quantitative spatial prediction of soil attributes over large areas (> 700 sq. km) of the landscape. The quantitative predictions reveal coherent spatial correlations of the difference in constituent concentrations with legacy landscape features, and immediate disturbances on the landscape due to extreme events. Further for model validation using independent test data, we demonstrate that the results are better represented as a probability density function compared to a single validation subset. We have simulated up-scaled datasets at multiple spatial resolutions ranging from 10m to 90m from the AVIRIS data, including future space based Hyperspectral Infrared Imager (HyspIRI) like observations. These datasets are used to investigate the applicability of the developed modeling framework over increasing spatial resolutions on the characterization of soil constituents. We have outlined an evaluation framework with a set of metrics that considers the point-scale model performance as well as the consistency of cross-scale spatial predictions. The results indicate that the ensemble quantification method is scalable over the entire range of airborne to space-borne spatial resolutions and establishes the feasibility of quantification of soil constituents from space- based observations. Further, we develop a retrieval framework from satellites, which combines the developed modeling framework and spectral similarity measures for global scale characterization of soils using a weighted constrained optimization framework. The retrieval algorithm takes advantage of the potential of repeat temporal satellite measurements to evolve a dynamic spectral library and improve soil characterization. Finally, we demonstrate that in addition to soil constituents, hyperspectral data can add value to characterizations of leaf area density (LAD) estimations for dense overlapping canopies. We develop a method for the estimation of the vertical distribution of foliage or LAD using a combination of airborne LiDAR and hyperspectral data using a feature based data fusion approach. Tree species classification from hyperspectral data is used to develop a novel ellipsoidal ‘tree shaped’ voxel approach for characterizing the LAD of individual trees in a riparian forest setting. We found that the tree shaped voxels represents a more realistic characterization of the upper and middle parts of the tree canopy in terms of higher LAD values, for trees of different heights in a forest stand.
Issue Date:2016-04-20
Rights Information:Copyright 2016 Debsunder Dutta
Date Available in IDEALS:2016-07-07
Date Deposited:2016-05

This item appears in the following Collection(s)

Item Statistics