Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 31

PocketLAI

A smartphone app for estimating leaf area


index


Roberto Confalonieri, Marco Foi
University of Milan, Cassandra lab
PocketLAI IRRI, 8-11 October 2013
Leaf area index

Total one-sided area of leaf tissue per unit ground surface
Key variable for analyzing the interaction between plants
and atmosphere
o amount of radiation intercepted
o plant water requirements
o CO
2
sequestration
o assimilation of exogenous information in simulation
models
o forecasting purposes
PocketLAI IRRI, 8-11 October 2013
How to derive LAI?

It can be measured
1. collection of leaves
2. measurement of their area
dedicated instruments
acquiring and processing leaf images
PocketLAI IRRI, 8-11 October 2013
How to derive LAI?

It can be estimated
o allometric relationships
o inversion of light transmittance models
LAI-2000, LAI-2200
ceptometers (AccuPAR, SUNSCAN)
CI-100
hemispherical camera

Suitable for extensive campaigns
PocketLAI IRRI, 8-11 October 2013
Context: commercial tools

LAI-2000 (now LAI-2200) and ceptometers
quite expensive (4500 - 10000 $)
characterized by low portability (1224109 - 651443 cm;
4.15 - 6.5 kg with cases)
long and expensive maintenance services in case of
damages
field campaigns can be interrupted!
PocketLAI IRRI, 8-11 October 2013
Context: smartphones

Production volumes a bit larger than those of LAI-2000 and
ceptometers
Competing pressure is leading producers to provide
devices with wonderful hardware
High quality sensors and computational capability for a few
hundred dollars (games!)
the possibility of making/receiving telephone calls is
becoming somehow secondary
PocketLAI IRRI, 8-11 October 2013
Indirect methods

Light transmittance models relating gap fraction (P
0
) and LAI
Poisson model for random spatial distribution of infinitely
small leaves



o is the probe angle
o k is the extinction coefficient, function of and
(Campbell s ellipsoidal leaf angle distribution)
PocketLAI IRRI, 8-11 October 2013
( )
( ) LAI k
e P

=
u
u
0
How LAI is estimated?

Solutions: modified versions of the model
Ceptometers
o information on canopy structure & direct/diffuse
radiation
LAI-2000
o five probe angles: no other information needed

57.5 zenith angle (Warren-Wilson, 1963; Baret et al., 2010)
o neither information on canopy structure, nor different
angles are needed
o we started from here
PocketLAI IRRI, 8-11 October 2013
PocketLAI - flowchart


PocketLAI IRRI, 8-11 October 2013
STOP
START
Specify a code for the
measure and start the
acquisition procedure
Rotate the device along its main axis starting
with the display oriented downward
and concluding with
it in vertical position
Wait 5 sec. to allow the
user to put the device
below the canopy. Then
give a signal to user
Is the angle between
the vertical and the normal to the
screen = 57.5?
NO
YES
Acquire information
Process information
Store
data
Export data
in tabular and
GIS format
Put the device above
the canopy
Visualize
results
on the
display
(only for App-L)
segmentation
luminance
Two methods:
PocketLAI


PocketLAI IRRI, 8-11 October 2013
An integrated quick manual
is available
PocketLAI


PocketLAI IRRI, 8-11 October 2013
PocketLAI IRRI, 8-11 October 2013
PocketLAI


segmentation based on
pixels intensity (leaves
darker than cloudy sky)
segmentation based on
pixels chromatic values in
an HSB color space (blue
pixels detected)
PocketLAI


PocketLAI IRRI, 8-11 October 2013
weve seen that angles
are important for indirect
methods
we used the device
accelerometer to derive
an inclinometer
|
|
|
.
|

\
|
= =
|
|
|
.
|

\
|
=
1
0
0
R Rg
G
G
G
G
pz
py
px
p
( )
|
|
|
.
|

\
|

=
| |
| | |
cos sin 0
sin cos 0
0 0 1
x
R ( )
|
|
|
.
|

\
|
=
u u
u u
u
cos 0 sin
0 1 0
sin 0 cos
y
R
( )
|
|
|
.
|

\
|
=
1 0 0
0 cos sin
0 sin cos

z
R
2 2
cos sin
tan
pz py
px
pz py
px
xyz
G G
G
G G
G
+

=
|
|
.
|

\
|
+

=
| |
u
pz
py
xyz
G
G
= | tan
images
automatically acquired at 57.5
while the user is rotating the device
PocketLAI


PocketLAI IRRI, 8-11 October 2013
Images are acquired in
live-preview mode
Baret et al. (2010) Agr. For. Met. 150, 1393-1401
Processing


PocketLAI IRRI, 8-11 October 2013
( )
|
|
.
|

\
|

=
) 5 . 57 cos(
5 . 0
0
5 . 57
LAI
e P

1) segmentation
2) Luminance (L, candela m
-2
) using device exposimeter
S t
k N
L


=
2
| =
above
below
L
L
P
0
N is the f-number or focal ratio
t (sec) is the exposure time
S is the ISO sensitivity
k is the reflected-light constant
correction factor for scattered radiation below
the canopy
PocketLAI Tests

1. Evaluation of PocketLAI and other indirect methods
(AccuPAR ceptometer, LAI-2000) for rice by adapting the
ISO 5725 protocol for the validation of analytical methods


2. Comparison with AccuPAR ceptometer for canopy
structures deviating from the assumption behind the light
transmittance model implemented in PocketLAI
PocketLAI IRRI, 8-11 October 2013
Confalonieri et al. (2013) Comput. Electron. Agric. 96, 67-74
Francone et al. (2013) Field Crop. Res. in press
PocketLAI Tests (1), methods


ISO 5725-2: Accuracy
PocketLAI IRRI, 8-11 October 2013
0
2
4
6
8
media = 4
media = 4
mean = 4
mean = 4
0
2
4
6
8
media = 4
media = 6
mean = 4
mean = 6
Trueness
Precision
Repeatability
(variability just due to
the replicate effect)
Reproducibility
(variability due to
different users, etc.)
Inter-laboratory tests
or ring tests
or ring trials
PocketLAI Tests (1), methods

Inter-laboratory tests (ISO 5725):
different laboratories
different operators for each laboratory
different levels for the analyte
different replicates for level, lab, operator
presence of reference materials
PocketLAI IRRI, 8-11 October 2013
trueness
repeatability
linearity
reproducibility
PocketLAI Tests (1), methods

Inter-laboratory tests (ISO 5725):
different laboratories
different operators for each laboratory
different levels for the analyte
different replicates for level, lab, operator
presence of reference materials
PocketLAI IRRI, 8-11 October 2013
4 groups, each composed of 4 students
different sowing densities
different moments during crop cycle
homogeneous (small) plots
split in quadrants
each quadrant corresponds
to a single replicate (to
avoid trampling effects)
LAI from destructive method (20 plants plot
-1
)
PocketLAI Tests (1), methods

Inter-laboratory tests:
1. outliers detection (among laboratories for each level)
Cochrans test (variances)
Grubbs test (means)
2. trueness (RMSE, ME, CRM), linearity
3. Precision
standard deviation, limit ( ) of repeatability
standard deviation, limit ( ) of reproducibility
PocketLAI IRRI, 8-11 October 2013
t s r
r
2 =
t s R
R
2 =
t = critical value of the Student t distribution (2 tails)
at 95% confidence level for n-1 freedom degrees
PocketLAI Tests (1), results


PocketLAI IRRI, 8-11 October 2013
App-L is the most accurate
The other methods present similar performances
PocketLAI Tests (1), results


PocketLAI IRRI, 8-11 October 2013
a
: repeatability limit. 1
b
: relative standard deviation of repeatability. 2
c
: reproducibility limit. 3
d
: relative standard deviation of reproducibility. 4
e
: laboratory 3 is an outlier according to the Cochran test. 5
f
: corrected value (s
r
set equal to s
R
in case s
r
> s
R
; Orwitz, 1995; Scaglia et al., 2011). 6
App-G:
one of the
most precise
App-L:
decidedly the
less precise
PocketLAI Tests (1), results


PocketLAI IRRI, 8-11 October 2013
No relationships between repeatability and LAI limit
Only App-L (luminance) presents a significant relationship
between reproducibility and LAI
Repeatability Reproducibility
PocketLAI Tests (2)

Assumption behind models for light transmittance into the
canopy:
Small leaves randomly distributed within the canopy
PocketLAI IRRI, 8-11 October 2013
Rice canopies do not deviate
so much
but others do!
PocketLAI Tests (2), methods

Three canopies markedly deviating from the assumption
behind light transmittance model
o maize: big leaves, row seeded
o giant reed: big leaves, row-planted, plants far from
each other, marked tillering aptitude
o grassland: often not homogeneous (no random
distribution)
We just compared PocketLAI estimates with those provided
by the AccuPAR ceptometer
PocketLAI IRRI, 8-11 October 2013
10 cm 10 cm
PocketLAI Tests (2), methods

Measurement protocols (1):
ceptometer
o maize: Facchi et al. (2010)
o giant reed & grassland:
PocketLAI
o maize:
o giant reed & grassland:
PocketLAI IRRI, 8-11 October 2013
Facchi et al. (2010) J. Ag. Eng. 1, 33-40
2 replicates
5 readings below 2 replicates
5 readings 2 replicates
PocketLAI Tests (2), results

PocketLAI IRRI, 8-11 October 2013
maize giant reed grassland
Giant reed (& maize):
overestimation for low values (big leaves + clumping effect)
underestimation for values higher than 5 m
2
m
-2
Grassland:
no systematic under/overestimations
more replicates needed (heterogeneity, AccuPAR 80 cm-probe)
PocketLAI Conclusions

PocketLAI resulted comparable for rice to commercial
tools for both trueness and precision
It represent a nice solution
o in contexts characterized by low availability of
resources
o when portability is an important requisite
o in cases when a large number of instruments is needed
PocketLAI IRRI, 8-11 October 2013
PocketLAI Remarks

It presented a device-effect (especially for low-cost
products), thus requiring to apply calibration facilities
(and the need for developing automatic calibration
facilities)
It resulted less robust in cases of canopies markedly
deviating from the assumptions behind light transmittance
models are (maize, giant reed, grassland)
An automatic selection of the segmentation algorithm
would
o improve user experience and
o reduce subjectivity in evaluating how cloudy the sky is
PocketLAI IRRI, 8-11 October 2013
Pocket N


PocketLAI IRRI, 8-11 October 2013
coming soon



PocketLAI IRRI, 8-11 October 2013


Simplicity is complexity resolved
(C. Brncui)
Thanks so much for the invitation
and for your kind attention

You might also like