Effect of Spatial Filtering on Characterizing Soil Properties from Imaging Spectrometer Data

Debsunder Dutta, Praveen Kumar, Jonathan A. Greenberg

Research output: Research - peer-reviewArticle

Abstract

Airborne imaging spectroscopy covering wavelength range of 0.35-2.5 μm can be used to quantify soil textural properties and chemical constituents. In this paper, we evaluate the effects of spatial resolution on the quantification of soil constituents using a lasso algorithm-based ensemble bootstrapping framework. Airborne visible infrared imaging spectrometer data collected at 7.6 m resolution over Bird's Point New Madrid (BPNM) floodway in Missouri, USA, is upscaled using a spatial filter to simulate a satellite-based sensor and generate multiple coarser resolution datasets, including the originally proposed 60.8 m hyperspectral infrared imager like data. The simulated data at multiple spatial resolutions are used in an ensemble lasso algorithm-based modeling framework for developing quantitative prediction models and spatial mapping of the soil constituents. We outline 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 model results demonstrate that the ensemble quantification method is scalable, and further the model structure indicates the persistence of important spectral features across spatial resolutions. The probability density functions of the constituents over the BPNM landscape show that it is similar for multiple spatial resolutions. Finally, a comparison of the model predictions with statistical central values together with the within pixel variance across fine to coarse resolutions indicate that the model accurately captures the median values of the fine subgrid that the coarse-resolution data is composed of. This study establishes the feasibility for quantifying soil constituents from space-borne hyperspectral sensors.

LanguageEnglish (US)
Article number7935357
Pages4149-4170
Number of pages22
JournalIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Volume10
Issue number9
DOIs
StatePublished - Sep 1 2017

Fingerprint

soil property
spectrometer
effect
Spectrometers
Imaging techniques
spatial resolution
prediction
soil
Sensors
sensor
bird
Infrared imaging
Model structures
Image sensors
Spectroscopy
Wavelength
AVIRIS
bootstrapping
probability density function
pixel

Keywords

  • Hyperspectral
  • HyspIRI
  • Lasso algorithm
  • multiresolution
  • remote sensing
  • soil properties

ASJC Scopus subject areas

  • Computers in Earth Sciences
  • Atmospheric Science

Cite this

Effect of Spatial Filtering on Characterizing Soil Properties from Imaging Spectrometer Data. / Dutta, Debsunder; Kumar, Praveen; Greenberg, Jonathan A.

In: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol. 10, No. 9, 7935357, 01.09.2017, p. 4149-4170.

Research output: Research - peer-reviewArticle

@article{c28f9c17002f49e88723e2a41fccff9b,
title = "Effect of Spatial Filtering on Characterizing Soil Properties from Imaging Spectrometer Data",
abstract = "Airborne imaging spectroscopy covering wavelength range of 0.35-2.5 μm can be used to quantify soil textural properties and chemical constituents. In this paper, we evaluate the effects of spatial resolution on the quantification of soil constituents using a lasso algorithm-based ensemble bootstrapping framework. Airborne visible infrared imaging spectrometer data collected at 7.6 m resolution over Bird's Point New Madrid (BPNM) floodway in Missouri, USA, is upscaled using a spatial filter to simulate a satellite-based sensor and generate multiple coarser resolution datasets, including the originally proposed 60.8 m hyperspectral infrared imager like data. The simulated data at multiple spatial resolutions are used in an ensemble lasso algorithm-based modeling framework for developing quantitative prediction models and spatial mapping of the soil constituents. We outline 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 model results demonstrate that the ensemble quantification method is scalable, and further the model structure indicates the persistence of important spectral features across spatial resolutions. The probability density functions of the constituents over the BPNM landscape show that it is similar for multiple spatial resolutions. Finally, a comparison of the model predictions with statistical central values together with the within pixel variance across fine to coarse resolutions indicate that the model accurately captures the median values of the fine subgrid that the coarse-resolution data is composed of. This study establishes the feasibility for quantifying soil constituents from space-borne hyperspectral sensors.",
keywords = "Hyperspectral, HyspIRI, Lasso algorithm, multiresolution, remote sensing, soil properties",
author = "Debsunder Dutta and Praveen Kumar and Greenberg, {Jonathan A.}",
year = "2017",
month = "9",
doi = "10.1109/JSTARS.2017.2701809",
volume = "10",
pages = "4149--4170",
journal = "IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing",
issn = "1939-1404",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "9",

}

TY - JOUR

T1 - Effect of Spatial Filtering on Characterizing Soil Properties from Imaging Spectrometer Data

AU - Dutta,Debsunder

AU - Kumar,Praveen

AU - Greenberg,Jonathan A.

PY - 2017/9/1

Y1 - 2017/9/1

N2 - Airborne imaging spectroscopy covering wavelength range of 0.35-2.5 μm can be used to quantify soil textural properties and chemical constituents. In this paper, we evaluate the effects of spatial resolution on the quantification of soil constituents using a lasso algorithm-based ensemble bootstrapping framework. Airborne visible infrared imaging spectrometer data collected at 7.6 m resolution over Bird's Point New Madrid (BPNM) floodway in Missouri, USA, is upscaled using a spatial filter to simulate a satellite-based sensor and generate multiple coarser resolution datasets, including the originally proposed 60.8 m hyperspectral infrared imager like data. The simulated data at multiple spatial resolutions are used in an ensemble lasso algorithm-based modeling framework for developing quantitative prediction models and spatial mapping of the soil constituents. We outline 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 model results demonstrate that the ensemble quantification method is scalable, and further the model structure indicates the persistence of important spectral features across spatial resolutions. The probability density functions of the constituents over the BPNM landscape show that it is similar for multiple spatial resolutions. Finally, a comparison of the model predictions with statistical central values together with the within pixel variance across fine to coarse resolutions indicate that the model accurately captures the median values of the fine subgrid that the coarse-resolution data is composed of. This study establishes the feasibility for quantifying soil constituents from space-borne hyperspectral sensors.

AB - Airborne imaging spectroscopy covering wavelength range of 0.35-2.5 μm can be used to quantify soil textural properties and chemical constituents. In this paper, we evaluate the effects of spatial resolution on the quantification of soil constituents using a lasso algorithm-based ensemble bootstrapping framework. Airborne visible infrared imaging spectrometer data collected at 7.6 m resolution over Bird's Point New Madrid (BPNM) floodway in Missouri, USA, is upscaled using a spatial filter to simulate a satellite-based sensor and generate multiple coarser resolution datasets, including the originally proposed 60.8 m hyperspectral infrared imager like data. The simulated data at multiple spatial resolutions are used in an ensemble lasso algorithm-based modeling framework for developing quantitative prediction models and spatial mapping of the soil constituents. We outline 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 model results demonstrate that the ensemble quantification method is scalable, and further the model structure indicates the persistence of important spectral features across spatial resolutions. The probability density functions of the constituents over the BPNM landscape show that it is similar for multiple spatial resolutions. Finally, a comparison of the model predictions with statistical central values together with the within pixel variance across fine to coarse resolutions indicate that the model accurately captures the median values of the fine subgrid that the coarse-resolution data is composed of. This study establishes the feasibility for quantifying soil constituents from space-borne hyperspectral sensors.

KW - Hyperspectral

KW - HyspIRI

KW - Lasso algorithm

KW - multiresolution

KW - remote sensing

KW - soil properties

UR - http://www.scopus.com/inward/record.url?scp=85020065821&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85020065821&partnerID=8YFLogxK

U2 - 10.1109/JSTARS.2017.2701809

DO - 10.1109/JSTARS.2017.2701809

M3 - Article

VL - 10

SP - 4149

EP - 4170

JO - IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

T2 - IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

JF - IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

SN - 1939-1404

IS - 9

M1 - 7935357

ER -