Spectral Indices and Agronomic Variables Relationship of Cotton (Gossypium sps.) under Sowing Dates and Nitrogen Levels

 
Mohammad Shamim Ansari
Faculty of Land and Food Resources,
University of Melbourne, Australia
Mail id: [email protected]

Abstract:
Remote sensing plays an important role to estimate agronomic parameters and yield. To study the relationship between spectral indices and agronomic variables, yield attribute and yield of cotton (Gossypium sps.) species, field experiment was conducted at Punjab Agricultural University, Ludhiana, India. The spectral reflectances were taken in the two spectral bands, red (R) band (625-689 nm) and Near Infrared (NIR) band (760-897 nm) from the crop canopy to calculate spectral indices, Radiance Ratio (RR=NIR/R) and Normalized Difference Vegetation Index [NDVI=(NIR-R)/(NIR+R)]. The correlation studies are discussed between the agronomic variables i.e., plant height, leaf area index, chlorophyll content, total and dry matter partitioning with the spectral indices viz., RR and NDVI. To see the best regression equation, data were run in three models i.e. linear model 'A', quadratic model 'B' and multiple linear model 'C', in which independent variable 'days after sowing' was added in linear model 'A'. Similarly the correlation between yields attributes and spectral indices were calculated from the linear model under different date of sowing and nitrogen levels. The results are also presented of correlation between seed cotton yield and the spectral indices of different date of observations from the linear and quadratic models. The results show that plant growth parameters, yield attributes and seed cotton yield can be estimated from RR and/or NDVI.

Introduction
Cotton being a commercial crop plays an important role in agrarian and industrial economy of a country. Recent developments in the field of remote sensing have opened new avenues in the field of agriculture for getting information about crop condition and crop productivity. The measurement of leaf area index (LAI) and other agronomic parameters like chlorophyll, biomass of the crop by conventional methods is time consuming, especially when the measurements are required throughout the growth cycle of the crop. Spectral measurements from crops have shown promise for use in estimating these parameters during the growth cycle of the crop. To know the relationship between spectral data and agronomic variables, there was need to create variable growth conditions. In order to create variable growth conditions of American and Desi cotton, different sowing dates and nitrogen (N) levels were tested. N deficiency in plant leads to reduction of LAI, chlorophyll content, biomass and ultimately reduction in yield. The reflections of visible radiation and near infrared radiation from the most heavily fertilized plots were proportionally much less and higher than that of unfertilized plots, respectively (Stanhill et al 1972). It has widely been reported (Chang and Collins 1983 and Banninger 1990, among many others) that vegetation under stress show changes in reflectance in the near-infrared bands (750-1300 nm) and red band (680 nm).

Management of essential nutrients particularly nitrogen in relation to time of sowing is the most critical factor which greatly influences growth and yield of American (Gossypium hirsutum) and Desi cotton (G. arboreum) cotton (Gill, 1963). To utilize the full potential of remote sensing for the condition and monitor crop growth, it is essential to quantify the relationship between agronomic parameters and spectral properties of the crop. These above agronomic parameters were well correlated with the spectral indices in different crops (Qi et al 1995, Thenkabail et al 2000 and Zhao et al 2007). There is a lack of wok on monitoring of cotton crop condition under nitrogen and dates of sowing situation. This paper presents the results of the experiment, which was conducted on two species i.e., American (Gossypium hirsutum) & Desi cotton (G. arboreum) to correlate the spectral indices and agronomic variables of cotton crop under sowing dates and nitrogen levels.

METHODOLOGY

Site and layout
An experiment was conducted on cotton crop during 1997-98 Kharif season on a sandy loam soil at the experimental farm of the Punjab Agricultural University, Ludhiana, India. in the cotton – wheat rotation. The soil was low in available nitrogen, medium in available phosphorus and high in available potassium content throughout the profile. The experiment was laid out in split-plot design with two cotton species {American cotton (Gossypium hirsutum) cv. F 846 and Desi cotton (G. arboreum) cv. LD 327} and two dates of sowing (D1, 1 May 1997 and D2, 29 May 1997) in the main plot and five nitrogen (N) levels (viz., 0, 40, 80, 120 and 160 kg N/ha) in the sub-plot replicated three times. Nitrogen was applied as per treatment in the form of urea. Half of the N (in the form of urea) was applied at sowing and rest half at the appearance of first flower. The 30 kg P2O5/ha in the form of single super phosphate was drilled at the time of sowing. The all other agronomic practices were carried out as per the "PAU Agronomic Package and Practices for Kharif Crops'.

Spectral Observation and Spectral Indices Calculation
Spectral reflectance measurements were recorded between 0930 and 1130 hours in direct sunlight using an indigenously developed hand held operated Multiband Ground Truth Radiometer (GTR 12- 0.4) in two spectral bands, red (R) (625-689 nm) and Near Infrared (NIR) (760-897 nm) at fortnightly interval starting at 35 DAS up to harvest. Average of five measurements was made on each experimental plot with radiometer elevated 1 m vertically above the top of the crop canopy surface. All spectral measurements were normalized with the irradiance obtained by BaSO4 standard. The radiance ratio (RR) was calculated by dividing percent NIR reflectance value by percent red (R) reflectance value (Rouse et al 1973). This is very useful for monitoring crop growth condition (Tucker et al 1981, Kamat et al 1983, Hatfield 1983, Wiegand & Richardson 1984 and Richardson et al 1992). The normalized difference vegetation index (NDVI) were calculated as below: NDVI [(NIR -R)/(NIR+R)]. Fasheun and Balogun (1992) suggested that RR and NDVI were related/ sensitive with biochemical properties of leaf and phenological stage of crop respectively.

Agronomic Measurements and Statistical Procedure
Growth parameters i.e., leaf area index (LAI), plant height, total dry matter and dry matter partitioning, leaf area index (LAI), etc. were measured concurrently with the spectral data. The yield attributing parameters and phonologies (Ansari and Mahey 2002) were also taken to correlate with spectral vegetation indices. The chlorophyll content was determined by the method of Arnon 1949 in two dates of observation. Seed cotton yield of three picking was expressed in tonnes per hectare (t/ha). The growth and yield attribute parameters were correlated with vegetation indices, using the three models viz., simple linear (model A), quadratic (model B) and multiple linear (model C) in which 'Days After Sowing' (DAS) was used as additional variables. The simple linear, quadratic, and multiple linear regressions were done as per the procedure outlined by Gomez and Gomez (1983).

RESULTS AND DISCUSSION

Relationship between Growth Variables and Spectral Indices
The two dates of sowing and five nitrogen levels are combined and are expressed in American (Gossypium hirsutum) and Desi (G. arboreum) cotton as separately for combined dates of sowing (D1 +D2) and nitrogen levels (N) (Table 1). The numbers of data taken in correlation studies are expressed by 'n' in each case. The level of significance in percentage is expressed by 0.01 and 0.05.

The result shows that the plant height, stem, reproductive, root and total dry matter are highly significantly correlated at a, 0.01 with NDVI and high correlation (r) are found of Desi cotton as compared to American cotton species in model A, improved in model B and further improved in model C as affected by both sowing dates and nitrogen levels (Table 1). The plant height, stem and total dry matter are highly correlated with NDVI in quadratic model B irrespective of sowing dates and nitrogen treatment effect. However these correlations are significantly better in Desi cotton than American cotton. LAI and chlorophyll content are highly correlated and significant with RR in model B and with NDVI in model C of American cotton. The LAI is significantly correlated at 1 percent level of significance with RR than NDVI in linear model A. The correlation coefficient 'r' values further improved in model B and again further improved in model C in both cotton species. The LAI is best correlated in American cotton than Desi cotton. The leaf dry matter is highly correlated with spectral indices of American cotton affected by nitrogen in all models. The leaf dry matter is also highly correlated with NDVI of Desi cotton in model B and did not improved in model C. The chlorophyll content is significantly correlated with RR than NDVI in linear model A and slight increase in quadratic model B under American cotton than Desi cotton, except under nitrogen treatment. This 'r' values further improved in multiple model C with NDVI in Desi cotton than American cotton species under sowing dates and nitrogen treatment. The similar results were shown by Dubey et al 1994 and Rao et al 1997 that the spectral indices significantly correlated with LAI and dry biomass production of wheat crop. Spectral discrimination of nitrogen treatments on crop growth has been reported by research workers in Israel (Stanhill et al 1972), the USA (Walburg et al 1982 and Hinzman et al 1986), the Netherlands (Clevers 1986), Sweden (Kleman and Fagerlund 1987) and India (Patel et al 1985 and Mahey et al 1989). The effect of nitrogen treatment with the agronomic parameters and spectral indices were discussed and found that agronomic parameters were highly correlated with spectral indices (Ansari et al 2006).

Estimation of Agronomic Parameters
The LAI, plant height, stem, leaf, reproductive and total dry matter are regressed as a dependent variable for the selected highly correlated spectral indices of American and Desi cotton species, separately and validated (Table 2). The first sowing date (1 May 1997) data are used for regression model fit and second sowing date (29 May 1997) data are used for test of model. The coefficient of determination 'R2' are presented for model fit and validated model. The standard error (SE) of estimate also has been shown in table.

LAI is regressed in the linear model A for both American and Desi cotton and validated by test data for prediction. American cotton LAI regression and prediction are presented in Fig 1. The LAI of American cotton is explained by RR with R2 = 0.75. This predicted model explained 0.79 coefficient of determination. The correlation in Table 1 shows that the regression and prediction model will improve in model C especially with NDVI of both cotton species, which is not worked out here.

Table 1. Correlation coefficient (r) between agronomic variables (x) and vegetation indices (y =RR or NDVI) of American & Desi cotton treatment combinations affected by sowing dates and nitrogen separately.

Table 2. Regression and validated prediction model for the agronomic parameters with spectral indices.

Fig. 1 (a) LAI regression model against Radiance Ratio (RR) of American cotton from the 1st Sowing date (1 May 1997) data set and (b) the predicted LAI from the test data of 2nd sowing date (29 May 1997) data.

The linear regression equation for the Desi cotton gives the model fit and validated R2 values (0.59 and 0.61, respectively) and less than American cotton. The other agronomic parameters are also predicted from the regression model with spectral indices (Table 2). The total dry matter is predicted with model fit R2 = 0.90 by NDVI for Desi cotton in linear model rather than quadratic model but the test data did not give higher validation R2 value. Also in quadratic model the R2 did not improve both in fit model and predicted model. The plant height, leaf and stem dry matter are predicted by NDVI for Desi cotton in quadratic model with 0.58, 0.68 and 0.36 R2 values, respectively. The regression equation is worked out for the reproductive dry matter in quadratic model with maximum model fit R2 0.43 of American cotton with NDVI but fail to validate by test data (R2 = 0.11). Other agronomic parameters were also worked out and found less model fit R2. These agronomic parameters were regressed and validated in different crops (LAI by Qi et al 1995, leaf area index and plant height by Thenkabail et al 2000 and leaf area index (LAI), aboveground biomass by Zhao et al 2007) and found highly correlated related and estimated the these parameters from derived spectral indices.

Relationship between Spectral Parameters and Seed Cotton Yield and Yield Attributes
The linear correlation between spectral indices and yield attributes and seed cotton yield are calculated for the Sowing Dates and Nitrogen treatments to see the effect of sowing dates and nitrogen on the spectral characteristics with the change in yield attributes properties (Table 3). Seed cotton yield is significantly and linearly correlated with NDVI (r = 0.74) than RR (r = 0.59) in timely (D1) sown crop. Similarly number of flower, total boll and boll weight per plant are significantly and linearly correlated with NDVI in timely sown crop than RR. The number of flower, open boll, unopened boll and total boll per plant are significantly correlated with RR with

Table 3 Comparison of linear correlation 'r' between spectral indices (y = RR and NDVI) and seed cotton yield (x) and yield attributes (x) under sowing dates and Nitrogen levels of American and Desi cotton

Table 4 Correlation coefficient (r) between date wise spectral parameters (x=RR or NDVI) and seed cotton yield (y)

high 'r' values than NDVI in N0, N40, N80 and N120 nitrogen treatment (Table 3). Among these yield attributes, the number of open boll per plant is maximum correlated (0.82 to 0.86). Seed cotton yield is significantly and linearly correlated with N0 and N120 among other nitrogen treatments with RR (r = 0.62 and 0.61, respectively). Cotton weight per boll is negatively and significantly correlated with RR and NDVI while boll weight per plant is insignificantly correlated among nitrogen levels. The flower number also is negatively correlated with RR in N160 treatment.

The correlation is studied in two models i.e. linear model A and quadratic model B for each RR and NDVI at timely and late sown American and Desi cotton species yield (Table 4). In timely sown American cotton, the seed cotton yield is quadratically and significantly (0.01 level of significance) correlated with RR at 95 DAS (r = 0.74). In late sown crop, seed cotton yield is quadratically correlated at 35, 50, 110 and 125 DAS with RR (r = 0.74, 0.72, 0.73 and 0.84, respectively) and at 35 and 125 DAS with NDVI (r = 0.76 and 0.81 respectively). Correlation in Table 4 shows that timely sown American cotton species yield is significantly correlated with RR than the NDVI in quadratic model B from maximum vegetative growth period to maturity (95 DAS to 185 DAS). While late sown crop yield is highly correlated in quadratic model B with RR during initial vegetative growth period to maximum vegetative growth period (35 to 125 DAS).

Timely sown Desi cotton yield is highly linearly correlated with RR at 125 DAS (r = 0.86) and quadratically correlated with NDVI at 35 (r = 0.70) and 155 (r = 0.65) DAS. Late sown Desi cotton yield is quadratically correlated with RR at 65 to 125 DAS. The seed cotton yield is also linearly correlated with NDVI at 80, 95 and 140 DAS and 'r' values further improved with quadratic model. The late sown Desi cotton yield is also quadratically correlated with NDVI at 125 (r =0.95) and 170 (r = 0.65) DAS. It shows that in Desi cotton, seed cotton yield of timely sown crop is linearly correlated with RR at peak vegetative growth stage and quadratically correlated with NDVI at initial vegetative and late boll opening stage. While late sown Desi cotton yield is linearly and quadratically correlated at peak vegetative to maturity period with NDVI. Kamat et al 1983 also observed a 40 days period (55-95 DAS) during which spectral data was significantly correlated with biological yield and grain yield of wheat than earlier or late. During this period, the appropriate spectral indices in appropriate model could help to predict the seed cotton yield of American and Desi cotton separately in both timely and late sown crop. These prediction regressions (Ansari et al 1999) were worked out for the cotton and it can be estimated cotton production before the harvest of crop for regional and / or national planning for price control and export.

Conclusion

The agronomic variables are highly correlated with both spectral indices viz., RR and NDVI under American and Desi cotton. The LAI can be predicted by RR by simple linear model while plant height, stem, leaf, total dry matter with NDVI by quadratic model. Number of flower, open boll, total bolls and boll weight per plant are significantly linearly correlated with RR and NDVI and depend on the sowing dates. Timely sown American seed cotton yield can be estimated by RR in quadratic model B from maximum vegetative growth period (95 DAS) and in early vegetative growth period for late sown. Thus plant growth parameters, yield attributes and seed cotton yield can be estimated from RR or NDVI of cotton species.

References
 

1. De by, R., Knippers, R., Sun, Y., Ellis, M ., Kraak M ., Weir, M .,Georgiadou, Y., Radwan , M ., Westen, C., Kainz, W., Sides, E., 2000. "Principles of Geographic Information Systems" , ITC Educational Textbook Series , Enschede, The Netherlands, 230 Pages.Ansari M S and Mahey R K (2002) Effect of sowing dates and nitrogen levels on American (Gossypium hirsutum) and Desi (G. arboreum) cotton species. Agronomy Digest, 2: 101-105.
2. Ansari M S, Mahey R K and Sidhu S S (1999) Cotton yield prediction through spectral parameters. J Indian Soc Remote Sens 27(4): 185-193.
3. Ansari M S, Mahey, R.K., Sidhu, S.S. and Bahl, G.S. 2006. Spectral response of nitrogen fertilization in cotton (Gossypium species). In 13th Australian Society of Agronomy Symposium at Perth, Western Australia on 10-14 September, 2006.
4. Arnon D I (1949) Copper enzyme in isolated chloroplasts-polyphenol oxidase in Beta vulgaris. Pl Physiol 24 : 1-15.
5. Banninger C (1990) Laboratory fluorescence reflectance and chlorophyll measurements of foliage from a stressed spruce forest.User contributions to satellite remote sensing programmes. Proc 9th EAR Sel Symp. pp 328-330. Espoo, Finland.
6. Chang S H and Collins W (1983) Confirmation of the airborne biogeophysical mineral exploration technique using laboratory methods. Econ Geol 78 : 923-36.
7. Clevers J G P W (1986) Application of remote sensing to agricultural field trials. Agricultural University of Wageningen Paper 86-4.
8. Dubey R P, Ajwani N, Kalubarme M H, Sridhar V N, Navalgund R R, Mahey R K, Sidhu S S, Jhorar O P, Cheema S S and Narang R S (1994) Pre-harvest wheat yield and production for the Punjab, India. Int J Remote Sens 15 : 2137-44.
9. Fasheun T A and Balogun E E (1992) Some spectral signatures of grain Amaranthus (Amaranthus Cruentus L.). Int J Remote Sensing 13 : 2009-15.
10. Gill R S (1963) Studies on the effect of spacing in relation to sowing date and nitrogen on the growth and yield of American and Desi cotton in the Punjab M.Sc. Thesis, Punjab Agricultural University, Ludhiana, India
11. Gomez K A and Gomez A A (1983) Statistical procedures for agricultural research. pp 357-423. A willey – Interscience Publication, New York.
12. Hatfield J L (1983) Remote sensing estimators of potential and actual crop yield. Remote Sens Environ 13 : 301-11.
13. Hinzman L D, Bauer M E and Daughtry C S T (1986) Effects of nitrogen fertilization on growth and reflectance characteristics of winter wheat. Remote Sens Environ 19 : 47-61.
14. Kamat D S, Ajai, Shashi Kumar M N, Sinha S K, Chaturvedi G S and Singh A K (1983) Relationship of spectral parameters of wheat crop with yield. Indian J Agric Sci 53 : 89-93.
15. Kleman J and Fagerlund E (1987) Influence of different nitrogen and irrigation treatments on the spectral reflectance of barley. Remote Sens Environ 21 : 1-14.
16. Mahey R K, Sidhu S S, Singh R and Narang R S (1989) Spectral assessment of agronomic parameters and grain yield of Gobhi Sarson (Brassica napus). Proc Nat Symp on Engineering Application on Remote Sensing and Recent Advances. pp 174-78.
17. Patel N K, Singh J P, Sahai B and Patel M S (1985) Spectral response of rice crop and its relations to yield and yield attributes. Int J Remote Sens 6 : 657-64.
18. Qi J, Cabot F, Moran M S, and Dedieu G (1995) Biophysical parameter estimations using multidirectional spectral measurements. Remote Sens Environ 54:71-83.
19. Richardson A J, Wiegand C L, Wanjura D E, Dusek D and Steiner J L (1992) Multispectral analysis of spectral-biophysical data for sorghum. Remote Sens Environ 41 : 71-82.
20. Rao T V, Brahma BC and Singh R K (1997). Spectral response and canopy temperature of rainfed wheat in relation to pre-sown mulches and nitrogen levels. Ann Agric Res 18 : 127-34.
21. Rouse K W, Schell R H J A and Deering D W (1973) Monitoring vegetation systems in the great plains with ERTS. 3rd ERTs Symp NASA SP-351 I : 309-17.
22. Stanhill G, Kafkafi U, Fuchs M and Kagan Y (1972) The effect of fertilizer application on solar reflectance from a wheat crop Israel J Agric Res 22 : 109-18.
23. Thenkabail P S, Smith R B, and Pauw E De (2000) Hyperspectral vegetation indices and their relationships with agricultural crop characteristics. Remote Sens Environ 71:158-182.
24. Tucker C J, Holben B N, Elgin J H Jr and McMurtrey J E Jr.III (1981) Remote sensing of total dry matter accumulation in winter wheat. Remote Sens Environ 11 : 171-89.
25. Walburg G, Bauer M E, Daughtry C S T and Housley T L (1982) Effects of nitrogen nutrition on the growth, yield and reflectance characteristics of corn. Agron J . 74 : 677-83.
26. Wiegand C L and Richardson A J (1984) Leaf area, light interception and yield estimates from spectral component analysis. Agron J . 76 : 543-48.
27. Zhao D H, Huang L M, Li J L, and Qi J G (2007) A comparative analysis of broadband and narrowband derived vegetation indices in predicting LAI and CCD of a cotton canopy. ISPRS J Photogrammetry Remote Sens 62:25-33.