You are on page 1of 17

Wavelet-based PROSPECT inversion for retrieving

leaf mass per area (LMA) and equivalent water


thickness (EWT) from leaf reflectance

Dong Li, Tao Cheng, Xia Yao, Zhaoying Zhang, Yongchao Tian, Yan Zhu, Weixing Cao

National Engineering and Technology Center for Information Agriculture (NETCIA) &
Jiangsu Collaborative Innovation Center for Modern Crop Production
Nanjing Agricultural University, Nanjing, China
Introduction
• Leaf mass per area (LMA) plays a
central role in various plant strategy
schemes.

• Equivalent water thickness (EWT) is an


important indicator of crop water status
for assessing drought risks in the
growing season.

• The PROSPECT was widely used to


estimate leaf biochemical parameters,
such as LMA and EWT.
ill-posed ?
Nanjing Agricultural University IGARSS 2016 Beijing, China
Introduction
• Existing solutions to alleviate the ill-posed problem
Stepwise reflectance-based method: one parameter is retrieved
through each iteration until all parameters are retrieved (Li and Wang,
2011).

(Li and Wang, 2011)

Problem: The improvement in accuracy is not consistent between


different data sets.
Nanjing Agricultural University IGARSS 2016 Beijing, China
Introduction
• Existing solutions to alleviate the ill-posed problem
VI-based method: used the vegetation index (VI) for constructing
the merit function for inversion (Liang et al., 2015).

(le Maire et al.,


2008)

Problem: It needs prior information to select the optimal VI in order


to achieve a higher estimation accuracy. The optimal VI maybe
depend on the special data set.
Nanjing Agricultural University IGARSS 2016 Beijing, China
Introduction
• Our solution to alleviate the ill-posed problem
Wavelet-based method: used the wavelet coefficient spectra from
continuous wavelet transform (CWT) for constructing the merit
function for inversion.

CWT

– Decompose a spectrum into a number of scale components.


– Low scales: narrow-range, high frequency information.
– High scales: broad-range, low frequency information.
Nanjing Agricultural University IGARSS 2016 Beijing, China
Introduction
• Advantages:
 Smoothing the reflectance spectra.
 Enhancing the absorption features of biochemical parameters.
 Multi-scale analysis for separating absorption information of
different parameters.
(Rivard et al., 2008; Blackburn and Ferwerda, 2008; Cheng et al.,
2012, 2014)

• Our objective is to evaluate the performance of wavelet-based


method in alleviating the ill-posed problem.

Nanjing Agricultural University IGARSS 2016 Beijing, China


Methods
• PROSPECT can simulate leaf reflectance spectra according to the
leaf biochemical parameters (forward modeling) or derive leaf
biochemical parameters from leaf reflectance spectra (backward
inversion).

Forward modeling Backward inversion

Nanjing Agricultural University IGARSS 2016 Beijing, China


Methods
• PROSPECT+CWT
• Integrating the continuous wavelet analysis into the PROSPECT
inversion process.

Forward modeling Backward inversion

Nanjing Agricultural University IGARSS 2016 Beijing, China


Data description
• A multi-species dataset (LOPEX) was collected
by the Joint Research Centre at Ispra in 1993.

• A single-species dataset (Wheat) was collected


in Rugao Exp. station of NETCIA in Jiangsu,
China in 2015.

 Two wheat cultivars, three N levels, two


density levels, three replicates.
 36 plots (5 m×6 m each)
 Leaf reflectance spectra, LMA and EWT
were collected over critical growth stages

Nanjing Agricultural University IGARSS 2016 Beijing, China


The effect of scale on inversion accuracies
• The R2 for EWT was insensitive to
scale for both LOPEX and Wheat.
However, the R2 for LMA varied with LMA/LOPEX
EWT/LOPEX
scale and the variations are different LMA/Wheat
for the two Inversion
data sets.strategy EWT/Wheat

Spectral range 1000 – 2200 nm


• TheChlorophyll
RMSE curves Fixed at 30 μg/cm
exhibited 2
similar
patterns R2 curves.
as did theFixed
Carotenoid at 10 μg/cm2
EWT 40 - 400 g/m2
• TheLMA 19 – 200 g/m when
LMA was overestimated
2

the Nscale was less1than


- 3 28 and then
became underestimated when the
scale was greater than for the Wheat
data set.

Nanjing Agricultural University


Results — LMA
Wavelet-based method proposed
in the study significantly improved
the inversion accuracy for LMA
compared to other methods.

Wavelet-based
Standard Stepwise Standard Stepwise
Wavelet- Wavelet-
Method reflectance reflectance reflectance reflectance
based based
-based -based -based -based
LOPEX Wheat
R2 -0.12 0.01 0.76 -4.36 -1.81 0.51
RMSE(g/m2) 26.16 24.51 12.16 14.67 10.62 4.43
Bias(g/m2)
Standard -15.71 -10.23
reflectance-based -3.49 13.32 reflectance-based
Stepwise 4.92 0.87
Nanjing Agricultural University
Results — EWT
Wavelet-based method improved
the inversion accuracy for EWT,
especially for Wheat.
The EWT was underestimated for
other methods.

Wavelet-based
Standard Stepwise Standard Stepwise
Wavelet- Wavelet-
Method reflectance reflectance reflectance reflectance
based based
-based -based -based -based
LOPEX Wheat
R2 0.83 0.83 0.87 -1.22 -1.47 0.66
RMSE(g/m2) 21.48 21.98 18.86 35.48 37.42 13.78
Bias(g/m2)
Standard -4.19 -5.29
reflectance-based -5.78 32.22 reflectance-based
Stepwise 34.21 -3.99
Nanjing Agricultural University
The advantage of CWT
• An example was used to describe the difference of inversion results
with standard reflectance-based and wavelet-based inversion method.

Standard relectance- Wavelet-based


measured
based inversion inversion
LMA (g/m2) 46 30 45
EWT (g/m2) 135 118 145

measured
standard reflectance-based wavelet-based
Nanjing Agricultural University
The advantage of CWT

standard reflectance-based wavelet-based

• Matches in the wavelet domain for the wavelet-based inversion


benefited from the enhancement of absorption features in reflectance
spectra and led to more accurate retrievals of LMA and EWT

Nanjing Agricultural University


Summary
• By integrating the CWA into the PROSPECT inversion process, we
achieved more accurate retrievals of LMA and EWT.

• The inversion of LMA for a single-species data set (Wheat) was


dependent on scale of wavelet decomposition due to its narrow range
and the best inversion was obtained at scale 28.

• This inversion method has a great potential for developing portable


instruments to predict foliar biomass and water content of crops
without the need of calibration models.

• The proposed method will be expanded to the canopy level in our


future work.

Nanjing Agricultural University IGARSS 2016 Beijing, China


Acknowledgement
• Thanks to field crew for the wheat experiment
• Thanks to S. Jacquemoud for making the LOPEX data set available
online
• Funding sources
– The NSFC (31470084)
– The Fundamental Research Funds for the Central Universities
(KYRC201401)
– The Academic Program Development of Jiangsu Higher Education Institutions
(PAPD)
– Jiangsu Distinguished Professor Program
– Jiangsu Entrepreneurship and Innovation Doctor Program

Nanjing Agricultural University IGARSS 2016 Beijing, China


Nanjing Agricultural University IGARSS 2016 Beijing, China

You might also like