Comparison of crop canopy sensors in sugarcane L.R. Amaral, H.J.A. Rosa, G. Portz, F.B. Finazzi, J.P. Molin University of São Paulo, "Luiz de Queiroz" College of Agriculture, Biosystems Engineering Department, Piracicaba, São Paulo, Brazil jpmolin@usp.br Abstract Optimized nitrogen fertilization of sugarcane is still a challenge. Crop canopy reflectance sensors can potentially help in improving N management. However, there is no information available in literature about canopy sensor efficiency as well as about vegetation indices that are suitable for sugarcane production. Therefore, a comparison was undertaken. Two fields were scanned with three canopy sensors (GreenSeeker and CropCircle, models ACS-210 and ACS-430), resulting in five vegetation indices (VI). The resultant spatial data were compared. In order to compare the efficiency of each device, plant samples were taken and analyzed, and then the results were correlated with the sensed data. Both Crop Circle sensors had similar results, regardless of the VI used. Despite the GreenSeeker sensor had the lowest capability to identify sugarcane vigor, based on biomass samples, all sensors are able to identify the variability in crop development. Keywords: proximal sensing, optical sensor, vegetation index, nitrogen fertilization. Introduction Sugarcane (Saccharum spp.) is the most important crop for sugar and ethanol production in tropical and sub-tropical regions. Among the inputs used for this crop, nitrogen (N) is the one that still needs better criteria to be managed, since there are no reliable soil analyses available to determine N availability in tropical soils. One way to optimize N application in sugarcane is through the use of crop canopy reflectance sensors. This approach has already been efficiently used in crops like maize and wheat (Raun et al., 2005; Berntsen et al., 2006; Kitchen et al., 2010). However, in sugarcane there is still a lack of information about their suitability. There have been recent studies with canopy sensors on sugarcane. Molin et al (2010) and Amaral & Molin (2011) worked, respectively, with the canopy sensors Green Seeker (Trimble Navigation, Ltd., Sunny Vale, CA, USA) and CropCircle (model ACS-210, Holland Scientific Inc., Lincoln, NE, EUA), with sugarcane and found high relationships between N rates and NDVI values measured by these devices. In another study, Portz et al. (2011) found that N-Sensor ALS was able to identify biomass and N uptake variability in sugarcane fields. Lofton et al. (2012) obtained a positive relationship between response indices estimated by the GreenSeeker canopy sensor and sugarcane yield. Despite the availability of canopy sensors in the market, choosing which one is more reliable for sugarcane is difficult for users. Other comparison studies in crops like maize and wheat, have demonstrated a good relationship between the readings acquired by canopy sensors GreenSeeker and CropCircle ACS-210 (Tremblay et al., 2008; Sudduth et al., 2010; Shaver et al., 2011). However, no comparison studies between canopy

sensors on sugarcane are documented in the literature. Likewise, there are still some doubts about which vegetation index derived from canopy sensors could be more efficient for this crop. Thus, the objective of the present work was to compare the performance of canopy sensors and vegetation indices for detecting sugarcane variability, aiming the nitrogen fertilization. Materials and Methods Canopy sensors and vegetation indices The sensors used were CropCircle, models ACS-210 and ACS-430 (Holland Scientific Inc., Lincoln, NE, USA), and GreenSeeker (Trimble Navigation, Ltd., Sunny Vale, CA, USA). The wavelengths used and the vegetation indices (VI) calculated in each sensor are described in Table 1. Apart from CI (chlorophyll index), all the VIs were standard output from the devices. The CI was added since Solari et al. (2008), working with maize, pointed out that this VI can be more sensitive to crop variability than NDVI in crops with high leaf area index (LAI) such as sugarcane. Table 1. Canopy sensors used in this study, corresponding wavelengths, and vegetation indices calculated with reflectance in each wavelength. Canopy sensor Wavelenghts (nm) Vegetation index

CropCircle ACS-210 590 and 880 NDVICC210 = (R880 - R590)/(R880 + R590) CICC210 = (R880/R590)-1

CropCircle ACS-430 670, 730 and 780 NDRE CC430 = (R780 - R730)/(R780 + R730) NDVICC430 = (R780 - R670)/(R780 + R670)

GreenSeeker 670 and 760 NDVIGS = (R780 - R670)/(R780 + R670) Field sites and canopy sensors readings Two sugarcane fields with different varieties and crop cycles, both on heavy soils (Oxysol), were chosen. Field 1 (5 ha) was cultivated with the variety CTC2, third ratoon. Field 2 (6.2 ha) was cultivated with variety SP80-1816, seventh ratoon. When the crop stalk height averaged 0.5m, as recommended by Amaral et al. (2012), measurements with the three crop canopy reflectance sensors were undertaken. The sensors were mounted on a high clearance vehicle (Uniport 3000 NPK, Máquinas Agrícolas Jacto, Pompéia, SP, Brazil) and maintained at 0.8 m above the canopy and connected to a DGPS logging at 1.0 Hz to obtain geo-referenced data with a travel speed of ~4.44 m s-1 through the field (Figure 1).

Figure 1. Canopy sensors (a) GreenSeeker, (b) CropCircle ACS-210, and (c) CropCircle

ACS-430 mounted on the pneumatic fertilizer application boom of the Uniport 3000 NPK.

Sample points After obtaining the three sensors readings, interpolated surfaces were created in GIS software (SSToolBox®) and thirty sample points were allocated in each field, taking into account the variability verified by each sensor. Each sample point consisted of four sugarcane rows of 5-m length (i.e., 30 m2 at 1.5-m row width) where the number and the height of stalks were measured. Portable chlorophyll meter readings (SPAD) were acquired on two different leaves, +1 (leaf top visible dewlap) and +3 (leaf third visible dewlap), thirty readings per sample point. Thirty sample leaves (leaf +3) were taken at each point and were sent to a laboratory for nitrogen and sulfur content determination. Above ground fresh biomass was evaluated by cutting 1.5m sub-plots on three rows, totalizing 6.75m2. Data analysis Vegetation indices were kriged in a 5-m grid (Vesper 1.6) and the pixel values were compared through scatter plots (SigmaPlot 10) and correlation analysis. The VI values were normalized to allow the box plot comparison. Relationships between VIs and crop parameters were analyzed by Pearson Correlation. Results According to the pixel correlation from kriged maps, expressed in terms of correlation coefficient in each scatter plot, high relationships between the different VIs were obtained (Figure 2). More data dispersion (less correlation between VIs) was observed when NDVIGS was compared with the VIs from CropCircle devices, due to variability pattern in the field 1 (Figure 3). The VIs in this field showed a small range with not well-defined zones, what affects the sensors ability in identifying crop vigor and reduces the sensors relationships.

NDRE - CC4300.04 0.08 0.12 0.16 0.20 0.24 0.28

ND

VI -

CC

430

0.10

0.20

0.30

0.40

0.50

0.60

0.70

NDRE - CC4300.04 0.08 0.12 0.16 0.20 0.24 0.28

ND

VI -

CC

210

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

NDRE - CC4300.04 0.08 0.12 0.16 0.20 0.24 0.28

CI -

CC

210

0.50

1.00

1.50

2.00

2.50

3.00

3.50

NDRE - CC4300.04 0.08 0.12 0.16 0.20 0.24 0.28

ND

VI -

GS

0.30

0.40

0.50

0.60

0.70

NDVI - GS0.30 0.40 0.50 0.60 0.70

ND

VI -

CC

210

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

CI - CC2100.50 1.00 1.50 2.00 2.50 3.00 3.50

ND

VI -

CC

210

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

r = 0.955 r = 0.946

r = 0.945 r = 0.900

r = 0.991 r = 0.873

Figure 2. Scatter plots showing the pixel relationships of both fields between vegetation

indices, as their respective Pearson Correlation Coefficient.

Figure 3. Interpolated maps from field 1, normalized NDVI from GreenSeeker (left) and

from CropCircle ACS-430 (right). NDVICC210 showed the lowest variability in data, possibly indicating low ability of NDVICC210 to characterize field variability or to differentiate areas with contrasting plant development. The other VIs showed similar data variability appearing to be a more reliable variability indicator in sugarcane fields (Figure 4). Vegetation indices proved to be good indicators of sugarcane vigor, since they had high relationships with biomass and plant stalk height (Figures 5 and 6). A High relationship with number of stalks was found when this parameter had a large influence on biomass (Field 1, Figure 5). Chlorophyll amount was not related to canopy sensor measurements, in agreement with what was reported by Amaral & Molin (2011) evaluating leaf +1, though when the diagnostic leaf was the older one (leaf +3) a higher relationship was found (three/four times higher Figures 5 and 6) .

Figure 4. Box plot of normalized data from all vegetation indices in both fields.

Outmost points indicate 5% and 95% of data dispersion. Leaf nitrogen concentration showed low correlation with VIs (Figure 6) as well as with chlorophyll meter readings (r < 0.1). An undesirable result was the significant negative correlation between leaf sulfur concentration and the VIs from CropCircle ACS-430. Similarly, sulfur availability can interfere with canopy sensor measurements, apart from the GreenSeeker.

Figure 5. Pearson Correlation Coefficient between vegetation indices and parameters

measured in point samples in field 2.

NDRE - CC430.

NDVI - CC430.

NDVI - CC210.

CI - CC210.

NDVI - GS.

NDRE - CC430

NDVI - CC430

NDVI - CC210

CI - CC210

NDVI - GS

VI N

orm

aliz

ed

60

80

100

120

140 Field 1Field 2

Biomass Height Stalks SPAD +1 SPAD +3

Pear

son

Cor

rela

tion

Coe

ficie

nt (r

)

0.0

0.2

0.4

0.6

0.8

1.0

r > 0.361, p = 0.05

NDRE - CC430NDVI - CC430NDVI - CC210CI - CC210NDVI - GS

Figure 6. Pearson Correlation Coefficient between vegetation indices and parameters

measured in point samples in field 1. Discussion All sensors and VIs studied are suitable for determining sugarcane fertilization due to their high relationship with plant biomass. Knowing the nitrogen concentration in plant biomass, the recommendation can be done based on these canopy sensor readings, as proposed by Portz et al. (2011), for an N-Sensor sugarcane fertilization algorithm. CropCircle devices showed similar suitability. However, when comparing them all, it was observed that each one had some specific abilities. The VIs from CC210 were more successful in identifying the number of stalks and chlorophyll content. Otherwise, VIs from CC430 were more successful in identifying leaf nitrogen concentration (according the correlation coefficient in Figures 5 and 6). Against these findings there is the frequently observation that these parameters frequently do not result in yield or reliable nitrogen nutrition information, and for this reason they are not so important as biomass. Both CropCircle sensors were efficient in determining sugarcane biomass and therefore they would be efficient in guiding nitrogen fertilization. The NDVI from the GreenSeeker showed less ability to identify plant development parameters and so it may be less accurate for guiding nitrogen fertilization over sugarcane. This problem strongly happened in field 1 where the GreenSeeker sensor readings identified less variability within the field (Figure 3) and the map was not related to crop vigor (relationship between NDVIGS and biomass in field 1 was only r = 0.43). The positive point of this canopy sensor is that it did not seem susceptible to sulfur interference. For this reason, its suitability in sugarcane should not be discarded, together with the fact that this sensor has already showed efficiency in estimating sugarcane yield (Lofton et al., 2012). Additionally, the similarity of the kriged maps represented by pixel correlation (Figure 2), with the exception of GreenSeeker map from field 1, infers that all sensors can detect variability within sugarcane fields. Sudduth et al. (2010), working with NDVI

Biomass Height Stalks SPAD +1 SPAD +3 Leaf N Leaf S

Pear

son

Cor

rela

tion

Coe

ficie

nt (r

)

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

r > | 0.361 |, p = 0.05

NDRE - CC430NDVI - CC430NDVI - CC210CI - CC210NDVI - GS

from CC210 and GreenSeeker, also found high correlations between both canopy sensors in maize. The expectation of Solari et al. (2008) that CICC210 could be more sensitive than NDVICC210 in high LAI crops were partially reproduced in sugarcane. The CI really showed a large range of data (Figure 4) but it did not identify plant parameters better than the NDVICC210 (Figure 5 and 6). Even so, due to its large range, CI should be preferred for its higher ability to identify variability. Literature has indicated that VIs using red-edge wavelengths were more efficient in identifying N variability in crops (Shiratsuchi et al., 2010). Consequently, NDRE (Normalized Difference Red-Edge) from CC430 (Table 1) should be the best option. However, in the present sugarcane study, all indices from CropCircle device showed similar efficiency. Therefore, more studies have to be conducted to prove these inferences. Conclusion CropCircle devices gave similar results in sugarcane, regardless of the vegetation index used. The NDVI determined by the GreenSeeker sensor was less able to identify sugarcane vigor within the fields than the other vegetation indices analyzed. Canopy sensor measurements do not correlate well with portable chlorophyll meter readings and leaf nitrogen concentration. Canopy sensors are able to identify crop development variability and can be a suitable tool guiding nitrogen fertilization in sugarcane. Acknowledgements All this work would not be possible without the collaboration of São Martinho’s Mill team and Máquinas Agrícolas Jacto. We also acknowledge the Research and Projects Financing (FINEP) from Ministry of Science and Technology through the PROSENSAP project for financial support and São Paulo Research Foundation (FAPESP) for providing the doctorate scholarship to the first author. References Amaral, L.R., and Molin, J.P. 2011. Optical sensor to support nitrogen fertilization

recommendation for sugarcane crops. (In Portuguese, with English abstract.) Pesquisa Agropecuária Brasileira 16: 1633-1642.

Amaral, L.R., Portz, G., Rosa, H.J.A. and Molin, J.P. 2012. Use of active crop canopy reflectance sensor for nitrogen sugarcane fertilization. p. 15. In: Proceedings of 11th International Conference on Precision Agriculture, Indianapolis, USA. CD-ROM

Berntsen, J., Thomsen, A., Schelde, K., Hansen, O.H., Knudsen, L., Broge, N., Hougaard, H., and Hørfarter, R. 2006. Algorithms for sensor-based redistribution of nitrogen fertilizer in winter wheat. Precision Agriculture 7: 65-83.

Kitchen, N.R., Sudduth, K.A., Drummond, S.T., Scharf, P.C., Palm, H.L., Roberts, D.F., and Vories, E.D. 2010. Ground-Based Canopy Reflectance Sensing for Variable-Rate Nitrogen Corn Fertilization. Agronomy Journal 102: 71–84.

Lofton, J., Tubana, B.S., Kanke, Y., Teboh, J., and Viator, H. 2012. Predicting Sugarcane Response to Nitrogen Using a Canopy Reflectance-Based Response Index Value. Agronomy Journal 104: 106–113.

Molin, J.P., Frasson, F.R., Amaral, L.R., Povh, F.P., and Salvi. J.V. 2010. Capability of an optical sensor in verifying the sugarcane response to nitrogen rates. (In Portuguese, with English abstract.) Revista Brasileira de Engenharia Agrícola e Ambiental 14(12): 1345-1349.

Portz, G., Molin, J.P., and Jasper, J. 2011. Active crop sensor to detect variability of nitrogen supply and biomass on sugarcane fields. Precision Agriculture 13: 33–44.

Raun, W., Solie, J., Stone, M., Martin, K., Freeman, K., Mullen, R., Zhang, H., Schepers, J., and Johnson, G. 2005. Optical Sensor-Based Algorithm for Crop Nitrogen Fertilization. Communications in Soil Science and Plant Analysis 36: 2759–2781.

Shaver, T.M., Westfall, D.G., and Khosla, R. 2011. Comparison of two active remote canopy sensors to develop N fertilizer algorithms. In: J. Stafford (Ed.), Proceedings of the 8th European Conference on Precision Agriculture, Wageningen Academic Publishers, The Netherlands, pp. 514–522.

Shiratsuchi, L.S., Ferguson, R.B., Shanahan, J.F., Adamchuk, V.I., and Slater, G.P. 2010. Comparison of spectral indices derived from active crop canopy sensors for assessing nitrogen and water status. In: Proceedings of 10th International Conference on Precision Agriculture, Denver, USA. CD-ROM

Solari, F., Shanahan, J., Ferguson, R., Schepers, J., and Gitelson, A. 2008. Active Sensor Reflectance Measurements of Corn Nitrogen Status and Yield Potential. Agronomy Journal 100: 571–579.

Sudduth, K.A., Kitchen, N.R., Drummond, S.T. 2010. Comparison of three canopy reflectance sensors for variable-rate nitrogen application in corn. In: Proceedings of 10th International Conference on Precision Agriculture, Denver, USA. CD-ROM

Tremblay, N., Wang, Z., Ma, B.L., Belec, C., and Vigneault, P.. 2008. A comparison of crop data measured by two commercial sensors for variable-rate nitrogen application. Precision Agriculture 10: 145–161.