Modeling rice growth from characteristics of Reflectance spectra

 

Chwen-Ming Yang* and Muh-Rong Su
Agronomist and Assistant, Department of Agronomy
Taiwan Agricultural Research Institute, Council of Agriculture
189 Chung-Cheng Road, Wufeng, Taichung Hsien,
Taiwan 41301, ROC
Tel: (886)-4-330-2301 ext.135, Fax: (886)-4-330-2806
E-mail: cmyang@wufeng.tari.gov.tw

Keywords: Modeling, Rice Growth, Reflectance Spectra, Vegetation index, Satellite

Abstract
A three-year (1996-1998) experiment was conducted at Taiwan Agricultural Research Institute (TARI) Experimental Farm, Wufeng to modeling of rice growth from characteristics of reflectance spectra. The growth-waveband relationships were analyzed by different reflectance transformation techniques and were compared among several satellite sensors. The narrow-band ground remotely sensed spectral data were acquired by a portable spectroradiometer and the broad-band multispectral satellite inputs were simulated from the ground measurements. Physical growth characters leaf number (LN), plant height (PH), leaf area index (LAI), leaf dry weight (LDW), and aboveground dry weight (ADW) were measured periodically. Results indicate a diverse correlation coefficients between growth characters and different vegetation indices and the wavebands set in satellite sensors. For linear relationship, correlation with vegetation index (VI) was generally higher than with single-waveband of satellites sensors. In a curvilinear relationship, correlation between growth character (e.g. LAI) and VI (e.g. NDVI) was greatly improved, and growth character was also better estimated from reflectance of multi-waveband of satellite data. Results suggest that rice growth may be reasonably assessed and monitored, either from the ground or satellites, from characteristics of reflectance spectra when proper wavebands are selected.

Introduction
Remote sensing technology (RS) is currently an effective tool widely adopted in various aspects of exploitation and management of natural resources. It provides a timely detailed spatially distributed information allowing one to measure, monitor, and analyze the time sequence changes of target(s) possible. In agriculture, RS has been employed to the development of precision farming for the better of crop production and environment protection. Nevertheless, the surface reflectance spectra over a wide range of objects and conditions should be identified and interpreted into meaningful outputs prior to decision-making and applications. For such purposes, data bank including spectral properties of physical characteristics of crops and land cover and the corresponding mathematical models have to be established based on ground truth. The in situ remotely sensed datasets from aircraft or satellite can therefore be reasonably resolved and reconstructed to the original state.

Yang and Su (1997) simplified the relationship between LAI and the near-infrared reflectance (756 nm) of rice canopy to the reverse of the Mitscherlich function. Yang and Su (1998a, 1998b) found a curvilinear link between growth characters (LDW and LAI) and normalized difference vegetation index (NDVI) in rice crops from the ground spectral measurements. Su and Yang (1999) further supported the fact by suggesting exponential formulae for estimating the advancement of growth characters from NDVI. However, it would be invaluable to learn whether the narrow-band models be applicable to the broad-band measures of satellite sensors in the real world applications.

The present study is to modeling of rice growth from spectral characteristics of vegetative cover. The growth-waveband relationships were analyzed and compared by different reflectance transformation techniques among a number of satellite sensors. The broad-band multispectral satellite inputs (Landsat-TM, Landsat-MSS, and Spot-HRV) were simulated from the narrow-band ground measurements and rice growth was assessed from physical growth characters investigations.

Materials and Methods
Rice (Oryza sativa L. cv. Tainung 67) were grown in the experimental farm of TARI, Wufeng (24° 02’ N, 120° 40’ E, elevation of 85 m) on a loam soil during the 1st and the 2nd crops in 1996-1998. Seedlings were machine-transplanted to north-south rows with planting density of 1.92 × 105 hills ha-1 , in 1996 and 1997, and 2.22 × 105 hills ha-1 , in 1998, respectively. Each crop had 3 plots (replicates) with plot size of 18 m × 11 m. Cultivation and fertilization followed local cultural practices with furrow irrigation. Pest and weed control was applied as needed to avoid pest and weed interference. Growth characters (GC) of LN, PH, LAI, LDW, and ADW were measured periodically. PH was taken by a ruler. LN was counted at the time of area measurements, which was determined by a portable area meter (model LI-3000A, LI-COR Inc., USA). LAI was calculated by the area of green leaves over unit area of land. LDW and ADW were weighted after oven-dried at 80°C for 72 h.

Radiance from the incident solar radiation and vegetative cover were acquired under the same sun conditions to calculate reflectance spectra of rice periodically during the growing periods of 1996-1998. The reflectance (%) of individual wavelength was calculated by dividing the vegetation radiance measurements with the corresponding incident solar radiation measurements (Yang and Su, 1997, 1998a; Su and Yang, 1999). A LI-COR model-1800 portable spectroradiometer, with 2-nm bandpasses in the range of 350-1100 nm, was used for the ground measurements. It was connected to a quartz fiber-optic probe (LI-1800-10) and a remote cosine receptor pointed downward in a nadir-viewing about 1.0 m above rice vegetative surface. This distance was adjusted to plant height, by a tripod, to scan the upward reflected radiation of canopy. Measurements were made on clear or near cloudless days within 11:00-12:00 local standard time and the average values were used. Characteristics wavelengths of reflectance spectra were determined by using the first order differentiation in cope with valley and peak observations of spectral waves over six cropping seasons (Su and Yang, 1999). Wavelengths at 554, 674, and 754 nm were selected as the characteristics wavelengths (CW). They coincide with the absorption minimum and maximum of chlorophyll and the near-infrared boundary of the chlorophyll red-edge, respectively. The corresponding wavebands of LANDSAT-TM, LANDSAT-MSS, and SPOT-HRV sensors to CW are listed in Table 1.


Table 1. Comparisons of the spectral characteristics of different satellite sensors with the characteristics wavelengths (CW) determined from reflectance spectra of the ground measurements.

Landsat-TM

Landsat-MSS

Spot-HRV

CW

 

-------------------- nm --------------------

 

450-520 (TM1)

 

 

 

520-600 (TM2)

500-600(MSS1)

500-590(HRV1)

554(GREEN)

630-690 (TM3)

600-700(MSS2)

610-680(HRV2)

674(RED)

760-900 (TM4)

700-800(MSS3)

790-890(HRV3)

754(NIR)

 

800-1100(MSS4)

 

 


Six VIs were selected from the literature (Elvidge and Chen, 1995; Kanemasu, 1974; Rouse et al., 1974; Tucker, 1979) defined as the followings: (1) sum vegetation index (SVI): NIR+RED; (2) difference vegetation index (DVI): NIR-RED; (3) ratio vegetation index (RVI): NIR/ RED; (4) green-red ratio vegetation index (GRVI): GREEN/RED; (5) normalized difference vegetation index (NDVI): (NIR-RED)/(NIR+RED); and (6) soil-adjusted vegetation index (SAVI): (NIR-RED)× 1.5/ (NIR+RED+0.5). The broad-band inputs of satellite sensors were simulated from the ground narrow-band reflectance (Table 1). Correlation matrices were applied for GC and wavebands and vegetation indices comparisons. Regression analyses were performed to generate fitting-curves and equations in order to monitoring the comparative changes of GC and wavebands and vegetation indices as plants aged.


Table 2. Correlation coefficients for growth characters of rice and vegetation indices (VI) calculated from characteristics wavelengths (CW) and the simulated wavebands of different satellite sensors.

VI

Leaf Number

Plant Height

Leaf area index

Leaf dry weight

Aboveground dry weight

Landsat-TM

NDVI

0.791

0.372

0.833

0.731

0.188

SAVI

0.790

0.374

0.835

0.734

0.192

RVI

0.815

0.187

0.881

0.744

0.071

DVI

0.598

0.535

0.871

0.818

0.430

SVI

0.284

0.648

0.683

0.699

0.609

GRVI

0.830

0.024

0.800

0.631

-0.048

Landsat-MSS

NDVI

0.819

0.318

0.844

0.730

0.137

SAVI

0.817

0.322

0.846

0.733

0.143

RVI

0.829

0.181

0.873

0.734

0.054

DVI

0.596

0.510

0.853

0.795

0.415

SVI

0.053

0.623

0.470

0.522

0.645

GRVI

0.854

-0.063

0.756

0.575

-0.142

Spot-HRV

NDVI

0.791

0.380

0.835

0.735

0.193

SAVI

0.786

0.389

0.839

0.740

0.202

RVI

0.817

0.200

0.881

0.747

0.073

DVI

0.591

0.546

0.870

0.820

0.439

SVI

0.279

0.654

0.682

0.700

0.616

GRVI

0.854

-0.022

0.780

0.605

-0.107

CW

NDVI

0.800

0.329

0.825

0.717

0.155

SAVI

0.799

0.333

0.827

0.718

0.159

RVI

0.798

0.161

0.874

0.733

0.071

DVI

0.630

0.475

0.865

0.798

0.379

SVI

0.306

0.584

0.667

0.670

0.558

GRVI

0.783

0.069

0.810

0.653

0.016


Results and Discussion
Typical patterns of the measured GC during the growing seasons were shown in Figure 1. The curves were analogous to seasonal trends of NDVI in both crops (Figure 2), suggesting that NDVI was sensitive to progression of vegetative cover and hence be a promising parameter in estimating plant growth. The time-sequential changes of NDVI and reflectance spectra were caused by the ‘environmental’ and ‘growth’ effects inducing significant changes in plant structure and morphology during crop growth (Curran, 1983; Masoni et al., 1996; Sinclair et al., 1971). The turning period of NDVI curve corresponded to the phase transition, from vegetative growth to reproductive growth (Su and Yang, 1999). Because of the strong background effect and senescence, it would be better not to use NDVI in the early and the late plant development.


Table 3. Correlation coefficients for growth characters of rice and characteristics wavelengths (CW) and the simulated wavebands of different satellite sensors.

Waveband

Leaf
number

Plant
height

Leaf
area index

Leaf
dry weight

Aboveground
dry weight

Landsat-TM

TM1

-0.790

-0.339

-0.797

-0.690

-0.135

TM2

-0.754

-0.041

-0.608

-0.483

-0.187

TM3

0.838

-0.015

-0.718

-0567

-0.152

TM4

0.468

0.600

0.806

0.783

0.523

Landsat-MSS

MSS1

-0.767

-0.073

-0.638

-0.512

-0.155

MSS2

-0.839

-0.017

-0.697

-0.543

-0.189

MSS3

0.372

0.599

0.723

0.715

0.555

MSS4

0.391

0.665

0.762

0.761

0.582

Spot-HRV

HRV1

-0.757

-0.083

-0.632

-0.509

-0.148

HRV2

-0.839

-0.010

-0.711

-0.560

-0.165

HRV3

0.460

0.610

0.804

0.784

0.532

CW

GREEN

-0.678

-0.040

-0.530

-0.421

-0.199

RED

-0.828

-0.052

-0.736

-0.590

-0.107

NIR

0.503

0.538

0.801

0.764

0.470


Relations of GC and VIs were examined by correlation matrix analysis. Generally LAI showed a better correlation with these VIs among GC in all satellite inputs (Table 2). The relationships were further strengthened with curvilinear functions. For instance, correlation between LAI and NDVI was improved in an exponential relationship (Figure 3) with the determining factors (R2 ) greater than 0.76. The predicted values of LAI were linearly correlated to the observed values. Correlation coefficients for GC and the simulated wavebands of different satellite sensors may be positive or negative, whereas LAI had closer linkage in general (Table 3). By stepwise regression to the second order of the spectral parameters, estimation of GC can be formulated to exponential equations with R2 greater than 0.53 (data not shown). Results indicate that the normalized difference transformation technique is suitable for the compensation of sun conditions and the estimation of rice growth. It suggests a relative simple approach based on correlation between GC and VI for estimating and monitoring the ground cover and growth performance. This methodology is founded on the close linkage in vegetation spectral characteristics and GC rather than purely empirical curve-fitting and thus may be more universally applicable to a variety of applications. However, such relationships may be changed and their availability will be reduced under situations such as environmental stresses, pests infection, low sun angle, and strong soil background effects.

The simulated results also suggest that broad-band satellite data may be substituted for ground-based narrow-band measurements if spectral characteristics of wavebands are well-selected. Therefore, data bank established by narrow-band ground truth is applicable to spectral inputs collected from satellites in assessing plant growth of a crop. Due to many more interacting variables, however, the spectral data measured from satellite remote sensing are considerable more complicated than those from the ground platform. Many important parameters in relation to atmospheric and geographic calibrations need to be taken into consideration.

References

  • Curran, P. J., 1983. Multispectral remote sensing for the estimation of green leaf index. Philosophical Transactions of Royal Society, London, Series A, 309, pp.257-270.
  • Elvidge, C. D., and Z. Chen, 1995. Comparison of broad-band and narrow-band red and near-infrared vegetation indices. Remote Sensing of Environment, 54, pp.38-48.
  • Kanemasu, E. T., 1974. Seasonal canopy reflectance patterns of wheat, sorghum, and soybean. Remote Sensing of Environment, 3, pp.43-47.
  • Masoni, A., L. Ercoli, and M. Mariotti, 1996. Spectral properties of leaves deficient in iron, sulfur, magnesium, and manganese. Agronomy Journal, 88, pp.937-943.
  • Richardson, A. J., C. L. Wiegand, D. F. Wanjura, D. Dusek, and J. L. Steiner, 1992. Multisite analyses of spectral-biophysical data for sorghum. Remote Sensing of Environment, 41, pp.71-82.
  • Rouse, J. W., R. H. Haas, J. A. Schell, D. W. Deering, and J. C. Harlan, 1974. Monitoring the vernal advancement and retrogradiation (greenwave effect) of natural vegetation. NASA/GSFC Type III final report. Greenbelt, MD., USA, 371pp.
  • Sinclair, T. R., R. M. Hoffer, and M. M. Schreiber, 1971. Reflectance and internal structure of leaves from several crops during a growing season. Agronomy Journal, 63, pp.864-868.
  • Su, M.-R., and C.-M. Yang, 1999. Estimation of rice growth from reflectance spectra of vegetative cover. Journal of Photogrammetry and Remote Sensing, 4(3), (in press) Tucker, C. J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, 8, pp.127-150.
  • Yang, C.-M., and M.-R. Su, 1997. Analysis of reflectance spectrum of rice canopy. Chinese Journal of Agrometeorology, 4, pp.87-95.
  • Yang, C.-M., and C.-C. Ko, 1998. Seasonal changes in canopy spectra of sweet potato. Journal of Photogrammetry and Remote Sensing, 3(1), pp.13-28.
  • Yang, C.-M., and M.-R. Su, 1998a. Seasonal variations of reflectance spectrum and vegetation index in rice vegetation cover. In: Proceeding of the 3rd Asian Crop Science Conference. April 27-May 2, 1998. Chinese Society of Agronomy. Taichung, Taiwan, pp.574-593.
  • Yang, C.-M., and M.-R. Su, 1998b. Correlation of spectral reflectance to growth of rice vegetation. In: Proceedings of the 19th Asian Conference on Remote Sensing. November 16-20, 1998. National Mapping and Resource Information Authority and Asian Association on Remote Sensing. Manila, Philippines, pp.A-1-1-A-1-6.



Days after transplanting
Figure 1. Seasonal changes in physical growth characters of leaf area index (LAI), plant height (PH), leaf number (LN), and leaf dry weight (LDW) of rice (Oryza sativa L. cv. Tainung 67) vegetative cover during the 1st and the 2nd cropping seasons in 1996-1998.




Days after transplanting
Figure 2. Seasonal changes of normalized difference vegetation index (NDVI) calculated from characteristics wavelengths of rice reflectance spectra.


NDVI

Observed
LAI

Figure 3. The exponential relationships between leaf area index (LAI) and normalized difference vegetation index (NDVI) and the linear correlation between the predicted values and the observed values of LAI.