MLSC-Week 5

EL
Machine Learning for Soil and Crop Management

Prof. Somsubhra Chakraborty
Agricultural and Food Engineering Department
Indian Institute of Technology Kharagpur
PT
Week 5: DIFFUSE REFLECTANCE
SPECTROSCOPY: BASICS AND
APPLICATIONS FOR CROP AND SOIL
N
LECTURE 21
• Multispectral Remote Sensing
• Hyperspectral Remote Sensing
EL
• Basics of Soil Spectroscopy
• Spectral Library
PT
N
• Soil Spectroscopy
• MIR
EL
• Vis-NIR
• Spectral library
PT
• Chemometrics
N
PANCHROMATIC BAND
• A panchromatic band (black and white band) is one band that usually contains a
couple of hundred nanometers bandwidth
EL
• The bandwidth enables it to hold a high signal-noise, making the panchromatic data
available at a high spatial resolution
• This image can be gathered with a higher resolution since the spectral range gives the
smaller detectors allowance to be utilized while sustaining the high signal-noise ratio
PT
N
HTTPS://CRISP.NUS.EDU.SG/~RESEARCH/TUTORIAL/PAN.GIF
SAMPLING THE SPECTRUM
EL
PT
N
MULTISPECTRAL AND HYPERSPECTRAL
• Multispectral imaging measures light in a small number (typically 3 to 15) of spectral
bands
EL
• Hyperspectral imaging is a special case of spectral imaging where often hundreds of
contiguous spectral bands are available
PT
N
MULTISPECTRAL AND HYPERSPECTRAL
EL
PT
N
MULTISPECTRAL VS HYPERSPECTRAL RS
EL
PT
N
MULTISPECTRAL VS HYPERSPECTRAL RS
EL
PT
N
HYPERSPECTRAL SENSORS
EL
PT
N
SOIL SPECTROSCOPY
• Can characterize soil properties efficiently
• Defined as the study of the spectral signature of a soil material
EL
• The spectral signature relates to soil characteristics such as organic and mineral
components
• Measurements are fast, cost-effective and nondestructive and can be made both in the
laboratory and in situ in the field
PT
• Soil composition and characteristics are encoded in the spectrum at specific
wavelengths of the electromagnetic spectrum
• For example, mid-infrared spectra have encoded information on soil mineralogy or soil
organic matter composition, which can be assessed quantitatively or qualitatively
N
using the absorption or reflectance at specific wavelengths
ELECTROMAGNETIC SPECTRUM
• Visible : soil colour
• The γ rays, X rays and infrared spectra : especially elemental composition and soil mineralogy
EL
PT
N
Lillesand et al. (2015)
SOIL SPECTROSCOPY
• Infrared spectra: sensitive to both organic and inorganic soil materials
• The mid-infrared (MIR) spectra: direct information on soil organic and mineral
EL
components of the soil than the visible and near-infrared (VisNIR) range
• Various components of the soil organic matter have very distinct spectral signature in
the mid-infrared range. The reason is that the fundamental molecular vibrations occur
in the mid-infrared range, while the overtones and combinations occur in the VisNIR
PT
• The absorption features detected in the vis-NIR: fewer, broader and more complex
than those recorded in the MIR
• Development of chemometrics: advanced soil spectroscopy
N
SOIL SPECTROSCOPY: POINTS TO BE REMEMBER!
• Infrared spectral bands are largely non-specific: they are not linearly related to a single
soil property and overlap between properties
EL
• This is particularly true for vis-NIR range of the spectra
• To extract these complex patterns and obtain quantitative estimates of a soil property,
mathematical transfer functions are used to correlate spectral wavelengths to soil
properties
PT
• The transfer function is calibrated using the spectral wavelengths as independent
variables and the laboratory measured values of the soil properties as the dependent
variable
• Once calibrated on the spectra, the soil property can be predicted using the spectral
information only
N
• PCR, PLSR, RF, SVR etc.
SOIL SPECTROSCOPY: POINTS TO BE REMEMBER!
• Fast
• Cost-effective (almost zero recurring cost)
• Needs minimum or no sample pretreatments
EL
• High Throughput and Non-invasive
• Portable
• Parameters:
• OC, Available N, P, K
• pH and clay
PT
• Moisture and heavy metals
N
Morgan et al. (2009)
SPECTRORADIOMETER
EL
PT
ASD FieldSpec 4 Hi-Res: High Resolution
Spectroradiometer PSR+ 3500 Spectroradiometer
N
HOW TO BUILD A SPECTRAL LIBRARY?
EL
PT
N
Viscarra-Rossel and McBratney (2008)
SOIL SPECTRA
EL
PT
N
Geeves et al. (1994)
SOIL SPECTRA
EL
PT
N
Reflectance soil spectra from Geeves et al. (1994) where the colour represents the amount
of clay (low to high clay content) derived from conventional laboratory soil analysis Wadoux et al. (2021)
• Geeves GW, Cresswell HP, Murphy BW, Gessler PI, Chartres CJ, Little IP, Bowman GM (1994) Physical, chemical
and morphological properties of soils in the wheat-belt of southern NSW and northern Victoria. NSW
Department of Conservation; Land Management/CSIRO Division of Soils Occasional Report, CSIRO
EL
• Lillesand T, Kiefer RW, Chipman J (2015) Remote sensing and image interpretation. Wiley, New York
• Viscarra-Rossel RA, McBratney AB (2008) Diffuse reflectance spectroscopy as a tool for digital soil mapping. In:
Digital soil mapping with limited data. Springer, Berlin, pp 165–172
PT
• Wadoux, A.M.J.C., Malone, B., Minasny, B., Fajardo, M., McBratney, A.B., (2021). Soil Spectral Inference with R
Analysing Digital Soil Spectra using the R Programming Environment. Springer.
N
N
PT
EL
EL
PT
N
LECTURE 22
• Soil Spectra
EL
• Spectroradiometer Specs
• Spectral Resolution
PT
N
• Soil Spectra
EL
• Spectral Resolution
• FWHM
PT
• Sampling Interval
N
HOW TO BUILD A SPECTRAL LIBRARY?
EL
PT
N
Viscarra-Rossel and McBratney (2008)
SOIL SPECTRA
EL
PT
N
Geeves et al. (1994)
SOIL SPECTRA
• Absorption features are typically very
narrow (< 20 nm), so narrow band
widths are necessary.
EL
• Many important band widths are also
fairly shallow, so high radiometric
resolution and SNR is also necessary.
PT
N Reflectance soil spectra from Geeves et al. (1994) where the colour represents the amount
of clay (low to high clay content) derived from conventional laboratory soil analysis
Wadoux et al. (2021)

WHY WE USE SPECTRORADIOMETER?
• Ground based Remote Sensing
EL
• - Helps understanding basics of target and EM interaction
• - Point/single pixel based study
• - Spectral ground truth of Satellite Remote Sensing
PT
• - Act as a reference for upscale to satellite /airborne
• Imaging sensors with required bands

N
VisNIR-DRS BASICS
• Visible and near infrared
diffuse reflectance
spectroscopy is a technology
EL
that uses a mug light or
contact probe to produce
reflections at different
wavelengths (350-2,500 nm)
• Field portable
PT
• Simple
• Reflectance patterns can
be correlated to a number
of soil properties
N
HOW VisNIR DRS WORKS?
EL
PT
N
HOW VisNIR DRS WORKS?
One can measure the light energy at
various wavelengths = a spectrum
The maxima and minima of spectral
EL
reflectance curves – minima are caused
by absorption, and we call these
absorption features or absorption bands.
Differences in absorption and scattering

for different wavelengths can be used to
PT
identify the compounds.
N
SPECTRORADIOMETER SPECS
EL
PT
N
1. A Light source
2. A dispersive unit (monochromator)
EL
3. A detector
4. (Fibers)
5. Absorbance / reflectance standard
PT
N
EL
PT
N
SPECTRORADIOMETER : WORKING MECHANISM
EL
PT
N
SPECTRORADIOMETER : WHITE REFERENCE
• 99% reflectance
• Optimize spectrometer
EL
• White reference: simplifies reflectance
• Removes ambient temperature effect
• Removes humidity effect
• Optimize : probes change
PT
N
SPECTRAL RESOLUTION
• Spectral sampling interval is the spacing between
sample points in the spectrum.
• Sampling is independent of resolution and in ASD
spectroradiometers is between 2 and 5 times per
EL
FWHM.
• The sampling interval here is 1.4 nm for the region
350 - 1000 nm and 2 nm for the region 1000 - 2500
nm.
• Spectral resolution is defined as the full-width-half-
maximum (FWHM) of the instrument response to a
PT
monochromatic source.
• These spectral resolution values have been measured
by calculating the FWHM of a near monochromatic
peak in a spectrum acquired when viewing the output
of a monochromator with the FieldSpec3®
spectroradiometer.
N
SPECTRAL RESOLUTION
• Spectral sampling refers to the interval at which DNs
are collected or sampled along the response curve
EL
• FWHM refers to the detector response derived from
exposure to a calibrated monochromatic source
• To be detected , feature must be larger than the
PT
FWHM of the spectrometer
N
• https://fsf.nerc.ac.uk/instruments/asd_fieldspec.shtml
• Viscarra-Rossel RA, McBratney AB (2008) Diffuse reflectance spectroscopy as a tool for digital soil mapping. In:
EL
Digital soil mapping with limited data. Springer, Berlin, pp 165–172
PT
N
N
PT
EL
EL
PT
N
LECTURE 23
• FOV
• Spectral Preprocessing
EL
• Spectral Trimming
• Savitzky-Golay filter
PT
• Standard Normal Variate
N
• FOV
• Spectral Trimming
EL
• Moving Window Average
• Savitzky-Golay filter
PT
• SNV
N
PRE AND POST-DISPERSIVE SPECTROMETER
• Pre-dispersive spectrometer:
• The ambient stray light signal can represent a large fraction of the total light signal
EL
measured by the detector
• Therefore, it can be a major source of error
• Post-Dispersive spectrometer
• The ambient stray light scattered from the sample is also collected, but with the post-
PT
dispersive instrument only ambient stray light of the same wavelength as that being
measured by the detector is added to the signal.
• Thus, the stray light signal represents a much smaller fraction of the total light signal
measured by the detector resulting in an increase in instrument baseline stability
N
FOV AND COVERAGE AREA
The small size of the foreoptics allows
positioning the foreoptics at a greater
distance from the surface under
observation.
EL
Large FOV: few measurement for larger
spatial coverage
Small FOV: No self-shadowing
PT
N
FOV AND COVERAGE AREA
EL
PT
N
FIELD AND LAB SCANNING
EL
PT
https://fsf.nerc.ac.uk/instruments/asd_fieldspec.shtml
N
FOREOPTICS AND ACCESSORIES
EL
PT
Contact Probe Hi-Brite Contact Probe Hi-Brite Muglight
N
Plant Probe Leaf Clip Pistol Grips
https://www.malvernpanalytical.com/en/products/product-range/asd-range/fieldspec-range/fieldspec-4-hi-res-ng-spectroradiometer#accessories
CONVERSION OF UNIT OF MEASURMENTS
EL
PT
N
y cm–1 = 10,000,000 / y nm
WHY SPECTRAL PREPROCESSING?
• Errors in the recorded spectral data
EL
• Scanning soil with non-uniform particle sizes such as soils, the
reflectance spectrum is often accompanied by scattering noise
PT
• First step in spectral data analysis
• No single best method
• The best preprocessing method depends on the sensitivity of the

N
following analysis to random variation in spectra
SPECTRAL PREPROCESSING: REFLECTANCE TO ABSORBANCE
EL
PT
N
SPECTRAL PREPROCESSING: SPECTRAL TRIMMING
• Trim wavelength from 350 to 499 nm (ultraviolet

Low SNR
EL
to green) and 2451 to 2500 nm
• Retained wavelengths in the spectral range of
500–2450 nm
PT
N Trimmed spectrum

SPECTRAL PREPROCESSING: MOVING WINDOW
• In a MW average operation, each wavelength
EL
value is taken as the average of the neighbouring
wavelengths
• The original spectra are smoothed, which
PT
reduces the information content but also the
noise that it contains too
• The user has to specify the size of the window,

N
i.e. over how many wavelengths the values are
averaged Wadoux et al. (2021)

SPECTRAL PREPROCESSING: SAVITZKY-GOLAY FILTERING
• The Savitzky-Golay filter fits a local polynomial
EL
regression (of order k) on a series of spectral
values to determine the smoothed value for
each wavelength, for a given filter length (also
PT
called the window size)
• Proposed by Savitzky and Golay (1964)
N
Reyna et al. (2017)
SPECTRAL PREPROCESSING: LIGHT SCATTERING EFFECTS
• Light scattered by soil samples results in deviations dependent on the wavelength, path
EL
length and sensitivity of the detector
• The deviations are difficult to quantify during soil scanning
• A pre-processing step is used to correct for deviations due to light scattering
PT
N
SPECTRAL PREPROCESSING: SNV
• Proposed by Barnes et al. (1989)
EL
• Corrects for single light scattering
• Also known as z-transformation or as centring
and scaling Barnes et al. (1989)
PT
• Operates per spectrum or row-wise
• Normalizes each spectrum

N
SPECTRAL PREPROCESSING: MSC
• Multiplicative scatter correction (MSC) is used to compensate for multiplicative
EL
deviations dependent from the wavelength
• The correction aligns each spectrum to a reference spectrum so that baseline and
amplification effects are at the same average level in every spectrum
PT
N
• As this reference spectrum is unknown, the mean spectrum of a
EL
given spectral library, denoted xref.
• Xref represents the mean scattering and offset
• Each spectrum xi is then fitted to the reference spectrum using
PT
the least squares method:
• Particle size and path length effects should vary randomly from
N
sample to sample, and therefore the average should reasonably
reduce these effects. This is the main assumption behind MSC.

EL
PT
N
https://towardsdatascience.com/scatter-correction-and-outlier-detection-in-nir-spectroscopy-7ec924af668
SPECTRAL PREPROCESSING: DETRENDING
• A practical alternative to SNV and MSC is to remove the mean
EL
value or a linear trend from the spectra
• This method is called detrending
• Can be used in combination with SNV or MSC
PT
N
SPECTRAL PREPROCESSING: CR
• Convex hull or continuum removal (CR): one
EL
type of baseline method that works by fitting
a convex hull to each spectrum and
computing the deviations from the hull
PT
• CR accentuates the absorption bands in the
spectra whileN minimizing brightness
differences
• Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard Normal Variate Transformation and De-Trending of
Near-Infrared Diffuse Reflectance Spectra. Applied Spectroscopy, 43(5), 772–777.
EL
• Morgan, C.L.S., Waiser, T.H., Brown, D.J., Hallmark, C.T., 2009. Simulated in situ characterization of soil organic
and inorganic carbon with visible near-infrared diffuse reflectance spectroscopy. Geoderma 151, 249–256.
• Reyna, L.; Dube, F.; Barrera, J.A.; Zagal, E. Potential Model Overfitting in Predicting Soil Carbon Content by
Visible and Near-Infrared Spectroscopy. Appl. Sci. 2017, 7, 708. https://doi.org/10.3390/app7070708
PT
N
N
PT
EL
EL
PT
N
LECTURE 24
• Derivative Spectra
• VisNIR DRS Soil Applications
EL
PT
N
• Boosted Regression Tree
• Clay prediction
EL
• First Derivative Spectra
• Inorganic Carbon
PT
• Variable Indicators
N
SPECTRAL PREPROCESSING: DERIVATIVE
• Converting the spectra to first- or second-order derivatives
EL
• Increasing the SNR
• Removes both additive and multiplicative effects on the spectra
• First-derivative: detrends the spectrum
PT
• Computing the derivative of a spectrum is usually performed after
a trimming or initial smoothing

N
• When computing the first- and second- derivatives with the SG
filter, an initial smoothing is not necessary

EL
PT
N
EL
PT
N
WHY DERIVATIVE SPECTRA?
Derivative: change in reflectance over a bandwidth Δ𝜆𝜆 = 𝜆𝜆𝑗𝑗 − 𝜆𝜆𝑖𝑖 , where 𝜆𝜆𝑗𝑗 >𝜆𝜆𝑖𝑖
𝑑𝑑 𝑑𝑑 (𝑛𝑛−1)
nth derivative= 𝑠𝑠 (Tsai and Philipot, 1998).
𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑(𝑛𝑛−1)
 Increasing resolution
EL
 Background elimination
 Noise filtration
PT
N
http://www.youngin.com/application/AN-0608-0115EN.pdf
WHY DERIVATIVE SPECTRA?
EL
PT
N http://www.youngin.com/application/AN-0608-0115EN.pdf
 High resolution: separate 2 analytes  Suppression of broad bands

with similar 𝜆𝜆𝑚𝑚𝑚𝑚𝑚𝑚 values
SPECTRAL PREPROCESSING + CHEMOMETRIC ALGORITHMS
 1st derivative of reflectance spectra

 Discrete wavelet transform
 Classification : linear discriminant analysis, support vector machines,
EL
and random forest
 Multivariate regression/ML:
 stepwise multiple linear regression
 partial least squares regression
 principal component regression
PT
 penalized spline
 GAM
 RF
 SVM
 ANN
 CNN
N
 ACO
 …..and many more
CLASSICAL VISNIR DRS SOIL APPLICATION
EL
PT
N Viscarra Rossel and Beherens (2010)
• 3768 samples from all 50 U.S. states and two tropical

territories and an additional 416 samples from 36
EL
different countries in Africa (125), Asia (104), the
Americas (75) and Europe (112)
• Boosted Regression Tree outperformed PLSR
PT
• Predicted ordinal clay mineralogy levels for
montmorillonite and kaolinite, with 88% and 96%
accuracy, respectively
N
Brown et al. (2006)
Viscarra Rossel et al. (2006)
EL
PT
N
EL
PT
N
CLASSICAL DRS SOIL APPLICATION
EL
PT
N
CLASSICAL DRS SOIL APPLICATION
EL
PT
N
VISNIR DRS BASED CLAY PREDICTION
EL
PT
N
Waiser et al. (2007)
VISNIR DRS BASED OC AND IC PREDICTION
EL
PT
N
VISNIR DRS BASED SURFACE SOIL MOISTURE PREDICTION
EL
PT
N
Zhu et al. (2010)
VISNIR DRS BASED SURFACE SOIL MOISTURE PREDICTION
EL
PT Zhu et al. (2010)
N
VISNIR DRS + VARIABLE INDICATORS
• A total of 300 samples were collected from three catenas of Transylvanian Plain, Romania.
• First derivative spectra were used to calculate Pearson's correlation coefficient (r),
EL
biweight midcorrelation (bicor), mutual information based adjacency (AMI), variable
importance in the projection (VIP), and their combinations.
• This variable indicator suite was combined with an ordered predictor selection (OPS)
method to choose the optimum number of spectral variables (NSV).
PT
• This method was tested with partial least squares regression (PLSR) and support vector
regression (SVR) with independent validation.
• Results indicated that the variable indicator-based SVR model yielded superior
predictability relative to full-spectrum PLSR model for all soil parameters.
• Moreover, both PLSR and SVR optimal models used the identical best variable indicators.
N
While AMI appeared as the best indicator for four soil attributes (clay, TN, TC and LOI),
bicor was selected as the best indicator for sand and silt.
Raj et al. (2018)

EL
PT
N
Raj et al. (2018)
EL
PT
N
Raj et al. (2018)
EL
PT
N
Raj et al. (2018)
• Brown, D.J., Shepherd, K.D., Walsh, M.G., Dewayne Mays, M., Reinsch, T.G., 2006. Global soil
characterization with VNIR diffuse reflectance spectroscopy. Geoderma 132 (3–4), 273–290.
• Morgan, C.L.S., Waiser, T.H., Brown, D.J., Hallmark, C.T., 2009. Simulated in situ characterization of soil
EL
organic and inorganic carbon with visible near-infrared diffuse reflectance spectroscopy. Geoderma 151,
249–256.
• Raj, A., Chakraborty, S., Duda, B.M., Weindorf, D.C., Li, B., Roy, S., et al., 2018. Soil mapping via diffuse
reflectance spectroscopy based on variable indicators: an ordered predictor selection approach.
Geoderma 314, 146–159.
PT
• Viscarra Rossel, R.A., Walvoort, D.J.J., McBratney, A.B., Janik, L.J., Skjemstad, J.O., 2006. Visible, near
infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of
various soil properties. Geoderma 131, 59–75.
• Waiser, T.H., Morgan, C.L.S., Brown, D.J., Hallmark, C.T., 2007. In situ characterization of soil clay content
with visible near-infrared diffuse reflectance spectroscopy. Soil Science Society of America Journal 71,
389–396.
N
N
PT
EL
EL
PT
N
LECTURE 25
• VisNIR DRS Soil Applications
• VisNIR DRS Crop Applications
EL
PT
N
• Soil Arsenic Pool
• OSC
EL
• EPO
• RSI
PT
• NDSI
N
VISNIR DRS FOR SOIL ARSENIC PREDICTION
EL
PT
N
Chakraborty et al. (2017a)
EL
PT
N
Chakraborty et al. (2017a)
EL
PT
N
VISNIR DRS FOR SOIL ARSENIC POOL PREDICTION
• Analysis of soil solid As phases has commonly used sequential extraction: Time consuming,
EL
destructive, and costly
• VisNIR DRS spectral data for rapidly predicting total As and five different solid As phases (Mg,
PO4, Ox, HCl and org pools)
PT
N
Chakraborty et al. (2017b)
EL
PT
N
Chakraborty et al. (2017)
EL
PT
N
EL
PT
N
DRS FOR PETROLEUM CONTAMINATED SOIL
EL
PT
N
VISNIR DRS : EPO
• The mathematical principle behind EPO :Roger et al. (2003).
• The EPO algorithm is based on the proposition that the field moist spectral matrix (X) can
be expressed as an amalgamation of a useful section (XP) that may describe a targeted soil
EL
property (e.g., clay), an extraneous section including the influence of moisture (XQ), and
spectral noise (R)
• The useful part (XP) can be computed via the difference matrix D, which is the difference
between the spectra of soil with moisture and spectra of dried soil.
• Note that, Q can be calculated from the singular value decomposition (SVD) of D and
PT
subsequently used for deriving the projection matrix P from P=I−Q, where I denotes the
identity matrix.
• Specifically, SVD helps in removing the space of the moisture, ultimately generating an
orthogonal space that remains unaffected by the moisture effect.
• Notably, the number of EPO components (c) can be derived from the SVD of D.
Alternatively; Q can be calculated from the principal component analysis (PCA) of DTD.
N
VISNIR DRS : EPO
EL
Results indicated that using
PT
EPO1 and EPO2 it was possible
to remove 5% and 29% of the
error introduced by the wet/in
situ measurements, respectively
N
PENETROMETER MOUNTED VISNIR DRS : EPO-PLS
• A penetrometer-mounted VisNIR probe was
used for collection of VisNIR spectra from soils
in situ.
• For each sampling location, VisNIR spectra
EL
were collected in situ using the penetrometer-
mounted probe and from airdried and ground
soil cores.
• An EPO was applied to in situ VisNIR spectra to
remove the effects of soil moisture and
intactness from the spectra.
• After application of the EPO, PLS models
PT
calibrated using a spectral library containing
only spectra from air-dried and ground soils
were used to predict clay content from in situ
spectra.
• Prior to application of the EPO, PLS was unable
to accurately predict clay content from in situ
spectra. After application of the EPO, PLS
performance on in situ spectra was quite good
N with average RMSE and R2 across all sampling
areas of 88 g kg−1 and 0.76, respectively.
Ackerson et al. (2017)

USE OF DRS AND CHEMOMETRICS: WATER DEFICIT IN RICE
• The changes in sucrose, reducing and total
sugar content due to water-deficit stress in rice
leaves were modeled .
EL
• Novel spectral indices in near infrared (NIR)
range: ratio spectral index (RSI) and normalised
difference spectral indices (NDSI) sensitive to
sucrose, reducing sugar and total sugar content
were identified which were subsequently
PT
calibrated and validated.
• The best performing multivariate models for
sucrose, reducing sugars and total sugars were
found to be, MARS, ANN and MARS,
respectively with respect to RPD values of 2.08,
2.44, and 1.9.
N
Das et al. (2018)
USE OF DRS AND CHEMOMETRICS: WATER DEFICIT IN RICE
EL
PT
N
Das et al. (2018)
USE OF DRS AND CHEMOMETRICS: SALINITY STRESS IN RICE
• VisNIR DRS was used for
leaf nutrient estimation in
rice under salinity stress.
EL
• Novel spectral indices in
VisNIR range were
developed and validated
• PLSR- and PCA-based
machine learning models
PT
were evaluated
• PLSR-based models
provided superior
performances N
Das et al. (2020)
USE OF DRS FOR CROP BASED AIR QUALITY MONITORING
• In urban areas: poor air quality
• Assesses the potential of hyperspectral tree
leaf reflectance for monitoring traffic related
EL
air pollution
• 29 Carpinus betulus saplings were exposed to
an environment with either high or low traffic
intensity
• The local air quality was estimated by leaf
saturation isothermal remanent magnetization
(SIRM)
PT
• Leaf chlorophyll content index (CCI), specific
leaf area (SLA) and water content (WC) were
determined
• To gain insight in the link between leaf
reflectance and air quality, the correlation
between SIRM and several spectral features
was determined
•
N
Spectral reflectance for wavelength bands in
the red and SWIR around the red edge, were
correlated to SIRM with Pearson correlations of
up to R = −0.85 (R2 = 0.72)
Brackx et al. (2017)

DRS APPLICATION FOR ALGAE
Rapid assessment
of algal biomass
and pigment
EL
contents
PT
N
Duppeti et al. (2017)
• Ackerson, J.P., Morgan, C.L.S., Ge, Y., 2017. Penetrometer-mounted VisNIR spectroscopy: application of EPO-PLS to in situ VisNIR
spectra. Geoderma 286, 131–138.
• Brackx, M., Van, W.S., Verhelst, J., Scheunders, P., Samson, R., 2016. Hyperspectral leaf reflectance of Carpinus betulus L. saplings
for urban air quality estimation. Environ. Pollut. 220, 159e167.
• Chakraborty, S., B. Li, D.C. Weindorf, and C.L.S. Morgan . 2019. External parameter orthogonalisation of Eastern European VisNIR-
DRS soil spectra. Geoderma 337:65-75.
EL
• Chakraborty, S., B. Li, S. Deb, S. Paul, D.C. Weindorf, and B.S. Das. 2017b. Predicting soil arsenic pools by visible near infrared
diffuse reflectance spectroscopy. Geoderma 296: 30-37.
• Chakraborty, S.,Weindorf, D.C., Deb, S., Li, B., Paul, S., Choudhury, A., Ray, D.P., 2017a. Rapid assessment of regional soil arsenic
pollution risk via diffuse reflectance spectroscopy. Geoderma 289, 72–81.
• Chakraborty, S., D.C. Weindorf, C.L.S. Morgan, Y. Ge, J. Galbraith, B. Li, and C.S. Kahlon. 2010. Rapid identification of oil
contaminated soils using visible near-infrared diffuse reflectance spectroscopy. Journal of Environment Quality 39(4): 1378-1387.
• Das, B., Manohara, K.K., Mahajan, G.R., Sahoo, R.N., 2020. Spectroscopy based novel spectral indices, PCA- and PLSR-coupled
machine learning models for salinity stress phenotyping of rice. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 229, 117983
PT
• Das, B., Sahoo, R. N., Pargal, S., Krishna, G., Verma, R., Chinnusamy, V., et al. (2018). Quantitative monitoring of sucrose, reducing
sugar and total sugar dynamics for phenotyping of water-deficit stress tolerance in rice through spectroscopy and chemometrics.
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 192(5), 41–51.
• Duppeti, H., Chakraborty, S., Das, B. S., Mallick, N., Kotamreddy, J. N. R. (2017). Rapid assessment of algal biomass and pigment
contents using diffuse reflectance spectroscopy and chemometrics. Algal research, 27, 274-285.
• Viscarra Rossel, R.A., Beherens, T., 2010. Using data mining tomodel and interpret soil diffuse reflectance spectra. Geoderma 158,
46–54.
N
N
PT
EL

MLSC-Week 5

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MLSC-Week 5

Uploaded by

Copyright:

Available Formats

EL

Machine Learning for Soil and Crop Management

Wadoux et al. (2021)

• Ground based Remote Sensing

• - Point/single pixel based study

• - Spectral ground truth of Satellite Remote Sensing

• Imaging sensors with required bands

The maxima and minima of spectral

Differences in absorption and scattering

are collected or sampled along the response curve

exposure to a calibrated monochromatic source

• To be detected , feature must be larger than the

Small FOV: No self-shadowing

• Errors in the recorded spectral data

reflectance spectrum is often accompanied by scattering noise

• No single best method

• The best preprocessing method depends on the sensitivity of the

• Trim wavelength from 350 to 499 nm (ultraviolet

• Retained wavelengths in the spectral range of

Wadoux et al. (2021)

• In a MW average operation, each wavelength

• The original spectra are smoothed, which

noise that it contains too

• The user has to specify the size of the window,

averaged Wadoux et al. (2021)

• The Savitzky-Golay filter fits a local polynomial

values to determine the smoothed value for

each wavelength, for a given filter length (also

• Proposed by Savitzky and Golay (1964)

• The deviations are difficult to quantify during soil scanning

• A pre-processing step is used to correct for deviations due to light scattering

• Proposed by Barnes et al. (1989)

• Also known as z-transformation or as centring

and scaling Barnes et al. (1989)

• Normalizes each spectrum

• Multiplicative scatter correction (MSC) is used to compensate for multiplicative

amplification effects are at the same average level in every spectrum

• As this reference spectrum is unknown, the mean spectrum of a

• Xref represents the mean scattering and offset

• Each spectrum xi is then fitted to the reference spectrum using

reduce these effects. This is the main assumption behind MSC.

• A practical alternative to SNV and MSC is to remove the mean

• This method is called detrending

• Can be used in combination with SNV or MSC

• Convex hull or continuum removal (CR): one

a convex hull to each spectrum and

computing the deviations from the hull

spectra whileN minimizing brightness

• Converting the spectra to first- or second-order derivatives

• Removes both additive and multiplicative effects on the spectra

• First-derivative: detrends the spectrum

a trimming or initial smoothing

filter, an initial smoothing is not necessary

 High resolution: separate 2 analytes  Suppression of broad bands

 1st derivative of reflectance spectra

• 3768 samples from all 50 U.S. states and two tropical

• Boosted Regression Tree outperformed PLSR

Viscarra Rossel et al. (2006)

Raj et al. (2018)

PO4, Ox, HCl and org pools)

Ackerson et al. (2017)

Brackx et al. (2017)

You might also like