Researches of spectral feature and growing monitoring of rice

Researches of spectral feature and growing monitoring of rice

Cha Zhongxing
Institute of Land survey and Planning of China

Wang Yanyi
Jiangsu Academy of Agrcultural Sciences

Qu Boolin
Institute of Land survey and Planning of China

Abstract
Using the method of power regression analysis, we find correlations between the Ratio Vegetation Index D/C (infrered to red, obtained from paddy spectral data) and LAI or TDM are very high. Before rice heading , the correlation coefficients (c.c.) are 0.96 and 0.93 respectively; after rice heading, the c.c. between D/C and TDM is -0.90 (5 varieties, 173 samples). Using the regression model obtained from one year to forecasting the LAI and TDM of another year, the precisions are: 96.4% and 94.6% respectively before rice handing, after rice heading it is 95.4% for forecasting TDM.

Introduction
Rice is one of the major crops in China. There are a great number of rice varieties and their growing periods and plant types are very different. So it is necessary to create a method to monitor the rice growth for the paddy management and yield estimation. In order to predict rice yield, agronomists usually measuse LAI and TDM during the rice growth period, but their methods are laborious and time-consuming.

With the development of remote sensing technology, it has been recognized that the crop spectral feature (such as light absorption, transmission and reflection) are closely determined by crop physiological characters which can express growing vigour and yield components of rice. From this point of view, the research was carried out the explore the relationship between rice spectral parameter and agronomic parameters during rice growing period and to monitor dynamically the rice growth.

Experiment Method
Experiments were carried out in Jiangsu Academy of Agricultural Sciences, Nanjing, 1987-1989,with 4 rice varieties (Yangan No.2-mediumjaponica, Xiu Shui No.4-late japonica, Nanjing No.11-medium indica, Shanyou No.63-hybrid medium indica), 3 fertilization levels and 2 replications. The 0.2 hectare experimental field was divided into 24 plots. Rice seedlings were transplanted on June 10th every year. Rice plant height, leaf area index(LAI), total dry biomass (TDM), tiller developments and phonological phases were measured every 15 days from the 7th day after transplantation.

The Exotech 100 Radiometer used in the experiment has 4 spectral bands: a 0.45-0.52um, b:052-0.60um, C: 0.63-0.69um, D:0.76-0.90um ( the same as the first 4 bands of TM) . All spectral data were collected by a polycordes on selected clear, windless days at 10:00-14:00 Binjing time. The Radiometer is set perpendicularly downward, 2m above the rice canopy. Five testing points were set in every plots. The average value of 5 tests was regarded as the spectral reflectance of the paddy plots. The average value of 5 tests was made once for every 1 or 2 weeks during the whole rice growing period.

Results and discussion

Fig. 1 shows that the paddy spectral features of blue, green and red bands are: low reflective, low transmissive and highly absorbed. The main reason is that the pigment especially the chlorophy11) strongly absorbs the photon. In red band the reflectance is very low because more than 90% of incident sun light energy is absorbed for rich photosynthesis. During the period from seedling transplantation to heading, long with the increase of rich leaf photosynthesis, the chlorophy11content in rice plant increases rapidly. With the arrival of rice heading time, rice reflectance in red bands increases. After milk stage, decreases but the rice panicle needs more nutrient, therefore the rice reflectance in red band increases more rapildly.

Fig. 1 Paddy reflectance character

There is a little reflectance peak in the green band and it increases with the growth of rice. Because the inner tissue of the rice leaf has high reflectance and high transmission in the near infrared (NIR) band, so there is a high reflectance and low absorption in the NIR band Fig. 2 shows that he NIR reflectance increases with the development during the growing period of the rice.

Fig. 2 Near infrared reflectance of rice

From seedling transplantation to heading stage, the reflectance of NIR increases with the increase of LAI, then a tendency of steadiness follows. During milk stage, along with large amount of nutrient transferred into the panicle and the change of inner tissue of the leaf, the reflectance of NIR decreases (Fig. 3)

Fig. 3 Relationship between LAI and NIR reflectance.

Agronomist had indicated that the value of rice biomass depends and LAI, and the former is closely related to yield. It is, therefore, reasonable to rely on R and NIR bands in studying the relationship between paddy spectrum and rice growth.

Relationship between paddy spectrum and LAI or TDM
Initial analysis of data shows that the relationship between rice reflectance of single band and LAI or TDM is not very ideal. But by using regression analysis form R and NIR band with 4 spectrum combinations with 6 mathmodels,, we have selected the best spectrum combination (D/C) and the best correlation model F( Y=AXB).

Because the reflectance before and after rice heading are different in data analysis, the rice growing period is divided into 2 parts.

Tab. 1 shows the poer regression coefficient between D/C and LAI or TDM is highest (0.961 and 0.928 respectively).

Using the D/C in 1988 and 1989 as one correlated variables and the LAI or TDM in 1988 and 1989 as the other correlated variables, we got the power regression coefficients of individual variety and total varieties of rice. Tab. 2 shows, before rice heading, correlation between D./C and LAI as well as TDM are significant for either individual variety and total varieties.

Using the D/C of 4 rice varieties in 1988 or 5 rice varieties in 1989 to make regression analysis with LAI or TDM (before rice heading, 82 samples), the results are:

In 1988	Y_L = 0.301 X^0.865	( r = 0.900) ......(1)
In 1988	Y_T = 8.684 X^1.100	( r = 0.892) ......(2)
In 1989	Y_L = 0.345 X^0.800	( r = 0.961) ......(3)
In 1989	Y_T= 12.44 X^0.992	( r = 0.928) ......(4)

( X=D/C, Y_L = LAI and Y_T = TDM )

Table. 1 The comparison of 6 regression analysis between LAI or TDM
and 4 composition of bands in 1989

LAI or TDM	Composition of bands	Mathmode1
		Y = AX+B	Y = (A / X)+B	Y = 1 / A+BX	Y = X / A+BX	Y=Ae^BX	Y = AX^B
		Correlation Coefficient
L A I	D/C	0.654	-0.669	-0.648	0.880	0.772	0.961
	D-C	0.585	-0.716	-0.592	0.821	0.774	0.832
	(D-C) / (D+C)	0.665	-0.880	-0.614	0.868	0.905	0.889
	C / (A+B+C)	-0.715	0.868	0.716	-0.695	-0.912	-0.880
T D M	D/C	0.684	-0.672	-0.620	0.873	0.764	0.928
	D-C	0.575	-0.708	-0.612	0.813	0.767	0.814
	(D-C) / (D+C)	0.704	-0.873	-0.625	0.863	0.900	0.871
	C / (A+B+C)	-0.711	0.853	0.702	-0.643	-0.903	-0.840

Table 2. Power correlation coefficient between D/C and TDM or LAI

Varceties	Before rice heading in 1988			Before rice heading in 1989			before rice heading in 1988 & 1989			After rice heading in 1988			After rice heading in 1989
	SAM.*	LAI	TDM	SAM.	LAI	TDM	SAM.	LAI	TDM	SAM	TDM		SAM.	TDM
	SAM.*	LAI	TDM	SAM.	LAI	TDM	SAM.	LAI	TDM	SAM	B.P.**	A.P.**	SAM.	B.P.	A.P.
Yangen No. 2	23	0.895	0.876	23	0.987	0.945	40	0.942	0.915	30	-0.678	-0.912	27	-0.689	-0.933
Nianjing No. 11	18	0.944	0.918	19	0.965	0.944	37	0.949	0.926	12	-0.304	-0.902	20	-0.487	-0.900
Shanyou No. 63	18	0.931	0.903	20	0.961	0.931	38	0.925	0.901	23	-0.854	-0.914	29	-0.691	-0.930
Xiu Shiu No. 4	23	0.932	0.927	18	0.949	0.937	41	0.940	0.911	30	-0.416	-0.912	14	-0.738	-0.916
ZaoDan 8	23	0.932	0.927	11	0.964	0.900	41	0.940	0.911	30	-0.416	-0.912	12	-0.374	-0.831
Total Varceties	82	0.900	0.892	91	0.961	0.928	173	0.933	0.903	95	-0.473	-0.904	103	-0.541	-0.901

* SAM. : Samples;
** B.P.: Before standardized processing
A.P.: After Standardized processing

When using the total of 1988 and 1989 (before rice heading 173 samples) to make the regression analysis with D/C and LAI or TDM, the reaults are:

Y = 0.354 X^0.794	(r = 0.933)......(5)
Y = 10.982 X^1.018	(r = 0.903)......(6)

From the above results of regression analysis, it is found that power correlation between D/C and LAI or TDM before heading is not limited by rice varieties.

After rice heading, the reflectance from paddy canopy is the mixed one from leaves and panicles of rice. Especially after the beginning of milk stage, because of differences in speed of grain filling and changes of the ratio of leaf and panicle, the spectral reflectances of different plots diverse greatly and the correlation between D/C and TDM (single variety or total varieties) is very low. But using following formula to make standardized process, the result is satisfactory .

Eq. 7

(X1' is the value after standardized processing, Xi is observed value and K is the number of observations after rice heading) Using 103 samples (after rice heading) to make the regression analysis between D/C and TDM, the correlation reached to extreme significance level (Tab. 2) and the formula as follows;

Y = 0.924 X^0.741

(r = -0.901)......(8)

(X is D/C and Y is TDM)

Test of mathematical model
Putting the graded D/C average value of 5 varieties in 1989 (before rice heading , 91 samples) into formula 1 and formula 2 (from 1988) and calculating the predicted value of LAI and TDM, the average rediction precision is 96.4% for TDM (Tab.3).

Grading the processed D/C value in 1988 (98 samples, after rice heading) into 5 classes, and putting it to formula 8, the prediction precisions for TDM are 91.2% to 98.5% (compare with the observed TDM value), the average precision is 95.4% (Tab.4)

Table.3 The comparison between predicted value and observed value of
LAI and TDM before rice heading.

D / C	range	< 10	10-20	20-30	30-40	< 50
	samples	20	22	14	25	4
	average value	4.56	19.81	24.29	34.75	41.57
LAI	predicted value	1.12	3.89	4.75	6.48	7.41
	observed value	1.16	3.56	4.94	6.40	7.32
	precision	96.5%	91.1%	96.1%	99.4%	98.8%
TDM	predicted value	48.09	231.90	290.20	430.30	510.21
	observed value	52.56	255.68	298.62	416.14	522.12
	precision	91.1%	90.3%	97.1%	96.7%	97.7%

Table 4. The comparison between predicted value and observed
value of TDM after rice heading

D / C	range	< 0.5	0.5-1.0	1.0-1.5	1.5-2.0	< 2.0
S.P.V.	samples	6	10	18	56	8
TDM	predicted	698.89	783.64	864.06	988.61	1116.22
	observed value	763.22	802.13	876.89	1019.38	1040.05
	precision	91.5%	97.7%	98.5%	96.9%	92.9%

The relationship between TDM and rice yield
Making linear regression analysis between the TDM value after rice heading and rice yield from 16 plots in 1989, the result is

Y = 0.707X + 315.069 (r = 0.803)......(9)

(X is average TDM after heading, Y is rice yield)

If biomass of each variety is measured according to its growing stages, perhaps the correlation between average TDM and yield will be much more ;significant.

Conclusion

Because there is some inner relationship between the reflectance of red band and the chlorophy11 content of rice leaf and between the reflectance of NIR band and the LAI of rice, D/C is very successful in reflecting the rice growth.
Before rice heading, the power correlation (. C.) between D/C and LAI or TDM is much closer than other kinds of combination. After rice heading, when standardized process is made, the P.C between D/C and TDM reaches to a very significance level.
Differences of rice varieties can be neglected in setting the correlation model between D/C and LAI or TDM.
It is quite possible to use "D/C - LAI" and "D/C - TDM" models in monitoring rice growth.
Rice yield can be predicted indirectly by suing the data of paddy spectrum.

References

N. K. Patll etc. The Relationship between the Rice spectral Reflection and Rice Yield. Int J. Remote Sensing, 6(5), 1985.
E. W. Lemaster etc. Seasonal Inspection of Model of Wheat Suits Spectral Albedo, Photogrammetic Engineering and Remote Sensing, 9, 1980.