Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane Received for publication: 02 November, 2016. Accepted for publication: 15 March, 2017. Doi: 10.15446/agron.colomb.v35n1.60852 1 Centro de Investigación de la Caña de Azúcar de Colombia (Sugarcane Research Center of Colombia), Cali (Colombia). cegarcia@cenicana.org Agronomía Colombiana 35(1), 82-91, 2017 Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane Evaluación de una cámara NIR para el monitoreo de productividad y efecto del nitrógeno en caña de azúcar César Edwin García1, David Montero1, and Héctor Alberto Chica1 ABSTRACT RESUMEN The main objective of the research carried out in the sugar pro- ductive sector in Colombia is to improve crop productivity of sugarcane. The rise of RPAS, together with the use of multispec- tral cameras, which allows for high spatial resolution images and spectral information outside the visible spectrum, has gen- erated an alternative nondestructive technological approach to monitoring crop sugarcane that must be evaluated and adapted to the specific conditions of Colombiá s sugar productive sector. In this context, this paper assesses the potential of a modified camera (NIR) to discriminate three varieties of sugarcane, as well as three doses of fertilization and estimating the sugarcane yield at an early stage, for the three varieties through multiple vegetation indices. In this study, no significant differences were found by vegetation index between fertilization doses, and only significant differences between varieties were found when the fertilization was normal or high. Likewise, multiple regressions between scores derived from vegetation indices after applying PCA and productivity produced determinations of up to 56%. El principal objetivo de las investigaciones llevadas a cabo en el sector azucarero de Colombia es el de mejorar la productividad del cultivo de la caña de azúcar. El auge de los RPAS y el uso de cámaras multiespectrales ha generado un enfoque tecnológico alternativo para monitorear los cultivos de caña de azúcar de manera no destructiva ya que se pueden obtener imágenes de alta resolución espacial e información espectral fuera del espectro visible. Esta tecnología debe ser evaluada y adaptada a las condiciones específicas del sector azucarero del país. En este contexto, el presente artículo presenta los resultados del potencial de una cámara modificada (NIR) para discriminar tres variedades de caña de azúcar, así como tres dosis de fer- tilización y estimar tempranamente la productividad de tres variedades de caña de azúcar por medio de múltiples índices de vegetación. En este estudio no se encontraron diferencias significativas por índices de vegetación entre dosis de fertil- ización y sólo se encontraron diferencias significativas entre variedades cuando la dosis de fertilización fue normal o alta. Adicionalmente, por cada variedad evaluada, se hizo un análisis de componentes principales entre los índices de vegetación (existió una alta correlación entre los índices), y con los cinco primeros componentes se hizo una regresión múltiple para modelar las toneladas de caña por hectárea obteniéndose de- terminaciones de hasta el 56%. Key words: nitrogen, yield, near infrared, precision agriculture, vegetation index, RPAS. Palabras clave: nitrógeno, productividad, infrarrojo cercano, agricultura de precisión, índice de vegetación, RPAS. The sensors used for spectral and spatial information may be on board satellites or airborne vehicles and their use will depend on the scale of work or detail expected in the products used to monitor the crop (Schmidt et al., 2000); Satellite images have been used on several occasions for studies at a regional level (Mulianga et al., 2013), but in the case of high level detail scales monitoring, the use of Unmanned Aerial Vehicles (UAVs) or Remotely Piloted Aircraft Systems (RPAS) is becoming a great tool for this task. This recent interest in using unmanned aerial vehicles in the study of agricultural crops is largely due to the great Introduction Monitoring sugarcane using orbital remote sensors has been the subject of research around the world. This tech- nique has allowed the measurement of the ref lected and absorbed radiation of the vegetation, obtaining different results that have been used for different purposes in agricul- ture such as discrimination between crops, discrimination based on variety, assessment of water stress, estimating yield, among others (Fernandes et al., 2011; Aguiar et al., 2011; Vieira et al., 2012; Govender et al., 2007). 83García, Montero, and Chica: Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane benefits gained by their implementation: low prices and easy logistics compared to satellite products, high spatial and temporal resolution and the possibility of spectral information in different regions of the electromagnetic spectrum that allow the calculation of various vegetation indices, making them a great tool for monitoring and follow-up in different cultures (Lelong et al., 2008). Multiple vegetation indices derived from visible and infrared spectrum of conventional, multispectral and hyperspectral sensors have been created in order to make assessments on different agronomic crops from these data, including determining the chlorophyll content (Hatfield et al., 2008), leaf area index (LAI) (Hunt et al., 2013) or nitrogen concentration (Miphokasap et al., 2012; Blackmer et al., 1996; Feng et al., 2008; Abdel-Rahman et al., 2010). Many others are equally used for nitrogen treatment dis- crimination (Foster et al., 2012) or varietal discrimination (Johnson et al., 2008; Galvao et al., 2005). More recently, the potential uses of these indices on crop yield from satellite sensors have been evaluated (Morel et al., 2014), as well as time series of other variables such as Fraction of Absorbed Photosynthetically Active radiation (fPAR) (Duveiller et al., 2013). Early estimates have been made using high- resolution airborne sensors of crop yield from vegetation indices (Berger et al., 2013) and even indices obtained with the use of ground sensors (Lofton et al., 2012). In Colombia, the monitoring of sugarcane cultivation is conducted by the Sugarcane Research Center of Colombia (Cenicaña), using satellite images, especially from MODIS- Terra sensor, Landsat TM and ETM+ images and data re- f lectance obtained by field spectroradiometry (Murillo and Carbonell, 2012). However, Colombia, being geographically located within the intertropical front, is subject to perma- nent cloud coverage, which has been a limiting factor for timely and regular acquisition of satellite images of high spatial optical resolution. The use of UAVs or RPAS with multispectral cameras can be a solution to the difficulties encountered in relation to the high percentage of clouds that appears in the satellite images since the f lying height of the RPAS allows imaging without cloud interference. Moreover, having the ability to increase the frequency of shots and produce highly detailed images, thus allowing for a full monitoring of the crop and achieving effective control to improve productivity, which can be affected by various factors that vary both spatially and temporally (GopalaPillai and Tian, 1999). The aim of this study was to evaluate the ability of a mul- tispectral camera to generate vegetation indices and using them to estimate crop yield, between varieties discrimina- tion and between nitrogen levels effect discrimination. Materials and methods Study area The study was conducted by Cenicaña in the experimental lot three of the International Center for Tropical Agricultu- re (CIAT), located in the municipality of Palmira, Valle del Cauca, Colombia (Fig. 1A). The geographical coordinates of the study area are 3°29’43.62’’ N and 76°21’49.51’’ W. The experimental lot houses two experiments in which different sugarcane varieties with different nitrogen doses (Fig. 1C) are evaluated. The average temperature of the area is 23.2°C, with an average radiation of 430 cal cm-2 d-1, an annual rainfall of 886 mm, a soil rich in organic matter, a clay content in a range between 35 and 60% and according to agroecological zoning (Carbonell et al., 2011) the experiment is located in a Typic Haplusterts soil, fine textural family and with a low humidity level (0-200 mm yr-1). RPAS and sensor A multirotor vehicle manufactured by the Advector Com- pany was used. It is a quadricopter type Araknos V2 model. An Agricultural Digital Camera (ADC lite), manufactured by Tetracam Inc. was used for capturing images. The ADC lite is a small 6.55×4.92 mm camera with a weight of 200 g; it generates a record of the position of each of the captured images and has an unique 3.2 megapixel sensor operating in the NIR bands from 520 nm to 920 nm, simulating TM2, TM3 and TM4 bands of Landsat Thematic Mapper sensor (TM) (Tetracam, 2015). The sensor of this camera uses a blocking filter in the blue region of the spectrum and uses the sensitivity curve of the blue filter to measure the res- ponse of the NIR, while green and red filters measure the corresponding response to each of parts of the spectrum. However, sensitivity curves of green and red have a high peak in the NIR perception, similar to the curve of sensi- tivity of the blue, perturbing a pure measurement of these regions of the spectrum (Tetracam, 2015). The f light took place on May 9, 2014, between 11:00 am and 1:00 pm, looking to have the most photosynthetically active radiation (PAR). At the time of the capturing of the images, the crop was 8.5 months old. Experimental design The study was carried out on a two factor experimental de- sign: three varieties of sugarcane in three doses of nitrogen. The levels of each factor are shown in table 1. 84 Agron. Colomb. 35(1) 2017 TABLE 1. Experimental design factors. Varieties Doses (Urea) CC 01-1940 Null (0 kg ha-1) CC 85-92 Normal (200 kg ha-1) CC 93-4418 High (400 kg ha-1) Selected varieties were the ones recommended by Cenicaña Variety Program due to their increase in planted area at a commercial level and their differences in architecture and color, which will permit to assess changes in vegetation and biomass index values. The appropriate N doses were established based on the spatial distribution of organic matter using SEF (Fertiliza- tion Expert System) developed by Cenicaña. A total of 36 field plots divided into nine treatments constructed from the combination of varieties and nitrogen doses were tested in one experiment (Tab. 2). The size of each plot corresponds to 12 furrows with dimensions of 1.65×20 m. In experiment two, only 200 kg ha-1 doses were applied, among the same three varieties of sugar cane for a total of 108 plots. TABLE 2. Treatments in experiment 1. Variety Doses (kg ha-1 of Urea) Treatment CC 01-1940 0 1 CC 85-92 0 2 CC 93-4418 0 3 CC 01-1940 200 4 CC 85-92 200 5 CC 93-4418 200 6 CC 01-1940 400 7 CC 85-92 400 8 CC 93-4418 400 9 Both experiments were harvested between 7 and 8 Sep- tember, 2014. The productivity of the experiments was measured using a cane loader where the weight and the GPS position were registered, the yield was expressed in tons of cane per hectare (t ha-1) for each plot of the two experiments. FIGURE 1. Geographic location of the study area. A. Location of the sugar industry in southwestern Colombia; B. location of the experimental lot at CIAT. C. Experimental lot divided in the two experiments assessed in the study. A B C Coordinate system Magna Sirgas West 85García, Montero, and Chica: Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane Image processing The images obtained by the ADC lite were stored in 8-bit RAW format, obtaining an image per band and converted into TIFF format using PixelWrench2 software, which came with the camera. Later, with the images, an ortho- rectified mosaic of the study area was generated, following the methodology proposed by García et al. (2014), using the X, Y and Z of the metadata of each of the acquired images. Once the images were processed, different vegetation in- dices were evaluated based on the available bands of the ADC Lite camera (Tab. 3). TABLE 3. Indices of vegetation assessed, where N corresponds to the digital levels to near-infrared, G to green digital levels and R to digital levels of red. Index Formula Chlorophyll index - Green Chlorophyll vegetation index Green normalized differ- ence vegetation Index Modified soil adjusted vegetation index Second modified triangular vegetation index Normalized difference vegetation index Normalized green red difference index Optimized soil adjusted vegetation index Ratio vegetation index Soil adjusted vegetation index Triangular vegetation index These vegetation indices respond to different factors, from the impact on quality of vegetation to structural factors such as leaves, stems and orientation. In the case of the quality of vegetation, it absorbs some of the incoming en- ergy to perform its various internal processes, presenting a high absorption in the visible spectrum, highlighting the absorption peaks in the blue and the red, while scat- tering large amounts of energy in the NIR, generating a high ref lectance, which is diminished when the vegetation has some type of complication or is reaching senescence (Ponzoni et al., 2012). To obtain the value of the indices evaluated for each of the plots were created Regions of Interest (ROI) in each of them, taking care to avoid areas that could alter the values obtained (shadows and regions with overturned sugarcane were excluded of the ROI), and extracted the average value of all pixels contained in the respective ROI for each plot. The vegetation indices calculated were evaluated accord- ing to their behavior in relation to the fertilization dose in experiment one, testing its statistical differences by such doses, varieties and their interaction through ANOVA analysis and Bonferroni groups, allowing the comparison of the average of the analyzed factors, after having been subject to ANOVA analysis. Because of the disturbance mentioned earlier in the filters of green and red in the ADC lite camera, the significance level was set at α=0.1 in order to avoid “false negatives” and losing the potential that can reach the assessed camera with the above mentioned limitations. Similarly, indices were related by linear regression against productivity per plot, using data from experiments one and two (36 and 108 plots respectively) evaluating their relation- ship per set of varieties and for each variety individually and determining the best productivity prediction model. Results Agronomic treatment outcomes The experiment showed a high productivity measured in tons of sugar cane per hectare (t ha-1) based on the industry average at the time of its harvest (121.4 t ha-1). The CC 01- 1940 variety with a dose of 400 kg ha-1 of urea (treatment 7), was the one with the highest productivity reaching 250 t ha-1. On average, CC 01-1940 variety, regardless of the dose, presented the highest tonnage with a productivity of 239 t ha-1, followed by varieties CC 93-4418 with a productivity of 225 t ha-1 and CC 85-92, with a productivity of 199 t ha-1 respectively. However, comparing the tonnages between varieties, no significant differences were found due to the interactions between variety and nitrogen doses. Analyzing the tonnage per dose of nitrogen regardless of the variety, the dose of 400 kg ha-1 was the one with the highest yield with 229 t ha-1, followed by the dose of 200 kg ha-1 with 223 t ha-1; the dose of 0 kg ha-1 was the lowest, with a productivity of 211 t ha-1. The statistical differences were found between the highest dose (400 kg ha-1) and the lowest dose (0 kg ha-1). In the results of tonnage by fertilization dose, analyzed by variety, significant differences were found between the 0 CIG = N G –1 (1) CVI = N *R G 2 (2) GNDVI = N – G N +G (3) MSAVI = 0 .5 ( 2 N+1– (2 N+1 )2 – 8( N – R) (4) MTVI 2 = 1.5 (2.5 ( N – G ) – 2.5 ( R – G ) ) ( 2 N +1 ) 2 – 6 N – 5 R – 0 .5 (5) NDVI = N – R N +R (6) NGRDI = G – R G+R (7) OSAVI = 1 .16 ( N – R ) ( N +R +0 .16 ) (8) RVI = N R (9) SAVI = 1.5 ( N – R) ( N+R+0 .5 ) (10) TVI = 0 .5 ( 120 ( N – G ) – 200 ( R – G )) (11) 86 Agron. Colomb. 35(1) 2017 kg ha-1 and the 400 kg ha-1 doses in the CC 01-1940 variety; for other treatments, including CC 93-4418 and CC 85-92 varieties, no significant differences between the doses were found (Fig. 2A). The yield results for varieties analyzed by dose showed that in the 0 kg ha-1 dose there were no significant differences between any of the three varieties, but in doses of 200 kg ha-1 and 400 kg ha-1, significant differences between CC 85-92 and CC 01-1940 varieties were found (Fig. 2B). Separability by indices of fertilization and varieties Analyses of variance were performed to determine the pos- sible statistical differences of the average of the established indices between different fertilization doses, as well as the varieties and their interaction. Table 4 is presented in P- values, obtained from the analyses of variance performed for each index, for the Variety and Dose factors and the interaction between Variety and Doses. There was no experimental evidence that would confirm the existence of significant differences between nitrogen doses and their interaction with the varieties, however, there were two indices, CIG and GNDVI, where it was possible to detect statistical differences between varieties, with a 10% significance. The values of each of the indices of experiment one were averaged for variety and in turn by fertilization dose, ob- taining a similar trend in most of the indices, for which the highest value of each index corresponds to a fertilization dose of 200 kg ha-1 in CC 01-1940 (treatment 4) (Fig. 3), except for the CVI and TVI indices, the first one showing a reverse behavior (Fig. 3C) and the second one obtaining the highest value on a dose of 0 kg ha-1 in CC 93-4418 variety (Fig. 3L); while lower index values correspond to the CC 85-92 variety on the three fertilization doses (treatments 2, 5 and 8). Bonferroni test was used at 10% (which keeps the experi- mental error rate with I and II errors at appropriate levels) to analyze the statistical differences of indices between fertilization dose pairs, varieties pairs (Fig. 3) and between varieties pairs for each fertilization dose (Fig. 4). In figure 3 it can be seen that no index manages to separate between fertilization doses, by variety; whereas in figure 4 it can be seen that two of the evaluated indices, CIG and GNDVI (Fig. 4B and 4D), were able to achieve the difference of CC 01-1940 and CC 93-4418 varieties from CC 85-92 variety, only when these three varieties were fertilized with normal and high doses (200 and 400 kg ha-1). FIGURE 2. A. Differences in productivity between doses for each variety (the letters corresponding to the groups should be read horizontally for each variety) and B. Differences in productivity between varieties per dose (the letters corresponding to the groups should be read vertically for each dose). B A A BA A A A A A A A A A BA A A B BA Yi el d (t h a- 1 ) B Doses 0 200 400 160 180 200 220 240 260 Yi el d (t h a- 1 ) A Doses 0 200 400 160 180 200 220 240 260 CC 93-4418CC 85-92CC 01-1940 TABLE 4. P-value by index for variety, dose factors and the variety-dose interaction, a result of ANOVA variance analysis for both experiments. Index Variety Doses Variety:doses CIG 0.0725* 0.7489 0.5134 CVI 0.3995 0.9360 0.5617 GNDVI 0.0660* 0.7239 0.5256 MSAVI 0.2383 0.8718 0.5131 MTVI2 0.2355 0.8693 0.5183 NDVI 0.2438 0.8881 0.5399 NGRDI 0.3341 0.9343 0.5585 OSAVI 0.2526 0.8893 0.5211 RVI 0.5750 0.8002 0.6481 SAVI 0.2530 0.8905 0.5288 TVI 0.3029 0.7770 0.6217 *Significant differences with a α 87García, Montero, and Chica: Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane B A A BA A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A AT V I L Doses 0 200 400 7000 8000 9000 10000 S A V I K Doses 0 200 400 1.15 1.20 1.25 1.30 R V I J Doses 0 200 400 10 18 16 14 12 20 O S A V I I Doses 0 200 400 0.90 0.94 0.98 1.02 N G R D I H Doses 0 200 400 0.66 0.70 0.74 0.78 N D V I G Doses 0 200 400 0.78 0.82 0.86 M TV I2 F Doses 0 200 400 1.64 1.68 1.72 1.76 M S A V I E Doses 0 200 400 0.87 0.89 0.91 0.93 G N D V I D Doses 0 200 400 0.26 0.28 0.30 0.32 C V I C Doses 0 200 400 0.24 0.28 0.32 0.36 C IG B Doses 0 200 400 0.75 0.80 0.85 0.90 0.90 Yi el d (t h a- 1 ) A Doses 0 200 400 160 180 200 220 240 260 CC 93-4418CC 85-92CC 01-1940 FIGURE 3. Bonferroni grouping of productivity and vegetation indices between doses for each variety (figures should be read horizontally for each of the varieties). 88 Agron. Colomb. 35(1) 2017 A A A A BA B A B BA A A A A BA B A B A A A A A A A A A A A A A A BA B A B A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A AT V I L Doses 0 200 400 7000 8000 9000 10000 S A V I K Doses 0 200 400 1.15 1.20 1.25 1.30 R V I J Doses 0 200 400 10 18 16 14 12 20 O S A V I I Doses 0 200 400 0.90 0.94 0.98 1.02 N G R D I H Doses 0 200 400 0.66 0.70 0.74 0.78 N D V I G Doses 0 200 400 0.78 0.82 0.86 M TV I2 F Doses 0 200 400 1.64 1.68 1.72 1.76 M S A V I E Doses 0 200 400 0.87 0.89 0.91 0.93 G N D V I D Doses 0 200 400 0.26 0.28 0.30 0.32 C V I C Doses 0 200 400 0.24 0.28 0.32 0.36 C IG B Doses 0 200 400 0.75 0.80 0.85 0.90 0.90 Yi el d (t h a- 1 ) A Doses 0 200 400 160 180 200 220 240 260 CC 93-4418CC 85-92CC 01-1940 FIGURE 4. Bonferroni grouping of productivity and vegetation indices between varieties per dose (figures should be read ver tically for each dose). 89García, Montero, and Chica: Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane Relationship between vegetation indices and productivity Structure of correlation between vegetation indices For each variety, a factor analysis was conducted, using the main components as extraction method, to find co- rrelations between the eleven indices measured in the study. Although the nature of the relationship between indices should not be affected by the variety, it was made for each one to numerically compare that the pattern did not change. In all cases (varieties), 100% of the variability structure of the vegetation indices was achieved with five factors, and 98% with the first two. Furthermore, significant correla- tions between components and indices were found in the first two factors, as described in table 5. TABLE 5. Correlations between each index and the first two factors ex- tracted by main components. CC 01-1940 CC 85-92 CC 93-4418 Index Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2 CIG 0.986 -0.115 0.984 -0.133 0.982 -0.118 CVI -0.998 0.001 -0.996 -0.037 -0.996 -0.006 GNDVI 0.986 -0.130 0.983 -0.138 0.980 -0.142 MSAVI 0.994 -0.067 0.997 -0.036 0.993 -0.098 MTVI2 0.994 -0.074 0.996 -0.048 0.992 -0.105 NDVI 0.996 -0.065 0.999 -0.028 0.996 -0.078 NGRDI 0.999 -0.031 0.999 0.002 0.998 -0.022 OSAVI 0.997 -0.062 0.999 -0.030 0.996 -0.078 RVI 0.939 0.108 0.938 0.115 0.914 0.243 SAVI 0.997 -0.058 0.999 -0.024 0.996 -0.076 TVI 0.664 0.745 0.300 0.952 0.750 0.643 From table 5 it is concluded that except the TVI (which is related to factor 2), all indices are strongly correlated because the lowest relationship between them and factor one is 0.93. The CVI index has an inverse relationship with the others. Relationship between vegetation indices and productivity As a result of the high correlation between indices, and to be able to model their relationship to productivity without multicollinearity problems, it was decided to set a multiple linear regression between productivity and the scores of the top five factors of factor analysis (the first five reach a 100% explained variability) for each variety. Table 6 shows that in the varieties CC 85-92 and CC 93-4418 it is possible to explain the variation in productivity in at least 48% and 56% respectively. Also, the average estimation error of the model is 23.6 t ha-1 and 29.5 t ha-1. In variety CC 01-1940 the adjustment range was 20% with an estimation error of 28.4 t ha-1. Table 7 shows the estimates of the model parameters for each variety. TABLE 6. Multiple regression statistics by variety. Variety Anova P-value Standard error (t) Adjusted R2 CC 01-1940 0.0129 28.4 20% CC 85-92 <.0001 23.6 48% CC 93-4418 <.0001 29.5 56% TABLE 7. Parameter estimates with their p-values for each variety. CC 01-1940 CC 85-92 CC 93-4418 Parameter Estimate P-value Estimate P-value Estimate P-value Intercept 229.108 <.0001 174.324 <.0001 174.324 <.0001 Factor 1 11.809 0.007 21.342 <.0001 21.342 <.0001 Factor 2 9.853 0.022 4.330 0.215 4.330 0.215 Factor 3 0.428 0.918 -7.421 0.037 -7.421 0.037 Factor 4 5.275 0.209 -0.602 0.862 -0.602 0.862 Factor 5 4.454 0.288 7.278 0.041 7.278 0.041 Discussion For most indices in experiment one, the highest value ob- tained is the response to a fertilization dose of 200 kg ha-1 on the CC 01-1940 variety, while the lower value obtained is the response of a fertilization dose of 400 kg ha-1 on the CC 85-92 variety; however, no index found significant differences between fertilization doses or Variety-Dose interaction. According to Foster et al. (2012), the indices that are sensitive to the content of chlorophyll in the plant, such as those working in regions of green, red and red edge; take advantage of the ref lectance ratios in these areas to achieve nitrogen treatment discriminations. The indices evaluated in this study did not discriminate between fer- tilization doses. This is explained either by low spectral quality relative to the assessed camera or by the amount of organic matter present in the experiment (3.03%), which allows the release of nitrogen and other nutrients in the soil, yielding a more constant availability of nutrients for the three varieties present. As for varietal discrimination in experiment one, just the CIG and GNDVI indices found significant differences in the dose of 400 kg ha-1 between CC 85-92 variety, which has the lowest index values, and the varieties CC 01-1940 and CC 93-4418, which have the highest index values, and no significant difference between the latter. Similarly, 90 Agron. Colomb. 35(1) 2017 significant differences were found in the dose of 200 kg ha-1 between varieties CC 85-92 and CC 01-1940 while CC 93-4418 variety could not be separated from the first two. This differentiation is due in part to the canopy architecture of the plant in each of the varieties, getting a much higher spectral response in varieties with curved blades (e.g., CC 01-1940) than in varieties with erect leaves (e.g., CC 85-92), as the penetration of sunlight, and hence the ref lectance, are different for each variety (Galvao et al., 2005; Victoria et al., 2013); however, another part of this differentiation is due to the LAI and the amount of biomass produced by each variety, eventually affecting the productivity of each of the varieties (Berger et al., 2013). In contrast to the number of indices found with the ability to perform a statistical difference between varieties (two indices), Hoyos-Villegas and Fritschi (2013), using a camera ADC lite, found sig- nificant differences between two varieties of soybean by nine vegetation indices out of 15 indices evaluated with a significance level of 95%. Conclusions Only vegetation indices CIG and GNDVI show a significant difference between CC 85-92 variety and varieties CC 01- 1940 and CC 93-4418 in doses of 400 kg ha-1 and significant differences between the CC 85-92 and CC 01-1940 varieties in doses of 200 kg ha-1. The CC 85-92 variety is the one that shows the lower value of vegetation index in the majority of evaluated cases, thus allowing for varietal discrimination in two of the evaluated indices. This behavior is associated with the plant structure and a low productivity compared to the productivity of the CC 01-1940 variety. The evaluated indices showed no significant differences between the fertilization doses applied in the nine treat- ments of the experiment, demonstrating that its answer on the evaluated indices is very similar and it will depend both on the variety and its architecture. The relationship between productivity and indices was modeled using a multiple linear regression and the scores obtained after an analysis of principal components (similar to a PLS regression, but made step by step) having found, for varieties CC 93-4418 and CC 85-92, determinations of 48 and 56% respectively. The percentage of variability in productivity for CC 01-1940 variety was 20%. Although productivity depends on many factors, it was possible to approximate the explanation of its variability with the group of evaluated indices, separated in principal components, to avoid multicollinearity. The percentage of explanation found and the standard errors were accept- able and showed that the indices became one more input to predict productivity. The ADC Lite camera proved to be effective in discrimi- nating CC 85-92 variety against varieties CC 01-1940 and CC 93-4418 in doses of 400 kg ha-1, and CC 85-92 variety against CC 01-1940 variety in doses of 200 kg ha-1, however, it does not produce positive results in the discrimination of fertilization doses. Acknowledgments The authors wish to thank the Colombian sugar productive sector contributions to the research being conducted in Cenicaña, as well as the company Advector for the f lights and the taking of the pictures with the ADC lite camera. Literature cited Abdel-Rahman, E. M., F. B. Ahmed, and M. van den Berg. 2010. Estimation of sugarcane leaf nitrogen concentration using in situ spectroscopy. Int. J. Appl. Earth Obs. Geoinf. 12S, S52-S57. Doi: 10.1016/j.jag.2009.11.003 Aguiar, D.A., B.F. Theodor Rudorff, W.F. Silva, M. Adami, and M. Pupin Mello. 2011. Remote sensing images in support of environmental protocol: Monitoring the sugarcane harvest in Sao Paulo State, Brazil. Remote Sens. 3, 2682-2703. Doi: 10.3390/rs3122682 Berger, A.G., D. Gaso, V.S. Ciganda, and A. Otero. 2013. Evaluation of the temporal dynamics of spectral indices and their relation- ship with biophysical variables on wheat for the purpose of yield estimation. 0039-0043. In: Anais XVI Simpósio Brasileiro de Sensoriamento Remoto (SBSR). Foz do Iguaçu, Brazil. Blackmer, T.M., J.S. Schepers, G.E. Varvel, and G. Meyer. 1996. Analysis of aerial photography for nitrogen stress within corn fields. Agron. J. 88, 729-733. Doi: 10.2134/agronj1996.000219 62008800050008x Carbonell, J.A., R. Quintero, J. S. Torres, C.A. Osorio, C.H. Isaacs, and J.I. Victoria. 2011. Zonificación agroecológica para el cultivo de la caña de azúcar en el valle del río Cauca (cuarta aproximación). Principios metodológicos y aplicaciones. Ser- vicio de Cooperación Técnica y Transferencia de Tecnología, Cenicaña. Cali, Colombia. Duveiller, G., R. López-Lozano, and B. Baruth. 2013. Enhanced processing of 1-km spatial resolution Fapar time series for sugarcane yield forecasting and monitoring. Remote Sens. 5, 1091-1116. Doi: 10.3390/rs5031091 Feng, W., X. Yao, Y. Zhu, Y.C. Tian, and W.X. Cao. 2008. Monitoring leaf nitrogen status with hyperspectral ref lectance in wheat. Eur. J. Agron. 28, 394-404. Doi: 10.1016/j.eja.2007.11.005 Fernandes, J.L., J. Vieira, and R.A. Camargo. 2011. Sugarcane y ield est i mates using t i me series a na lysis of spot veg- etation images. Sci. Agric. 68(2), 139-146. Doi: 10.1590/ S0103-90162011000200002 91García, Montero, and Chica: Evaluation of a NIR camera for monitoring yield and nitrogen effect in sugarcane Foster, A.J., V. Gopal Kakani, J. Ge, and J. Mosali. 2012. Discrimina- tion of switchgrass cultivars and nitrogen treatments using pig- ment profiles and hyperspectral leaf ref lectance data. Remote Sens. 4, 2576-2594. Doi: 10.3390/rs4092576 Galvao, L.S., A.R. Formaggio, and D.A. Tisot. 2005. Discrimination of sugarcane varieties in Southeastern Brazil with EO-1 Hy- perion data. Remote Sens. Environ. 94, 523-534. Doi: 10.1016/j. rse.2004.11.012 García, C.E., F.A. Herrera, and E. Erazo. 2014. Metodología básica para la generación de índices de vegetación mediante imágenes multiespectrales aerotransportadas aplicada en cultivos de caña de azúcar. In: Memorias XVI Simposio Internacional SELPER 2014. Sociedad Latinoamericana en Percepción Re- mota y Sistemas de Información Espacial. Medellin, Colombia. GopalaPillai, S. and L. Tian. 1999. In-field variability detection and spatial yield modeling for corn using digital aerial imaging. Trans. ASAE 42(6), 1911-1920. Doi: 10.13031/2013.13356 Govender, M., K. Chetty, and H. Bulcock. 2007. A review of hyper- spectral remote sensing and its application in vegetation and water resource studies. Water SA 33(2), 145-152. Doi: 10.4314/ wsa.v33i2.49049 Hatfield, J.L., A.A. Gitelson, J.S. Schepers, and C.L. Walthall. 2008. Application of spectral remote sensing for agronomic decisions. Agron. J. 100, S-117 - S-131. Doi: 10.2134/agronj2006.0370c Hoyos-Villegas, V. and F.B. Fritschi. 2013. Relationships among vegetation indices derived from aerial photographs and soy- bean growth and yield. Crop Sci. 53, 2631-2642. Doi: 10.2135/ cropsci2013.02.0126 Hunt, E.R. Jr., P.C. Doraiswamy, J.E. McMurtrey, C.S.T. Daughtry, E.M. Perry, and B. Akhmedov. 2013. A visible band index for remote sensing leaf chlorophyll content at the canopy scale. Int. J. Appl. Earth Obs. Geoinf. 21, 103-112. Doi: 10.1016/j. jag.2012.07.020 Johnson, R.M., R.P. Viator, J.C. Veremis, E.P.Jr. Richard, and P.V. Zimba. 2008. Discrimination of sugarcane varieties with pig- ment profiles and high resolution, hyperspectral leaf ref lec- tance data. J. Am. Soc. Sugar Cane Technol. 28, 63-75. Lelong, C.C.D., P. Burger, G. Jubelin, B. Roux, S. Labbé, and F. Baret. 2008. Assessment of unmanned aerial vehicles imagery for quantitative monitoring of wheat crop in small plots. Sensors 8(5), 3557-3585. Doi: 10.3390/s8053557 Lofton, J., B.S. Tubana, Y. Kanke, J. Teboh, H. Viator, and M. Dalen. 2012. Estimating sugarcane yield potential using an in-season determination of normalized difference vegetative index. Sen- sors 12, 7529-7547. Doi: 10.3390/s120607529 Morel, J., P. Todoroff, A. Bégué, A. Bury, J.F. Martiné, and M. Petit. 2014. Toward a satellite-based system of sugarcane yield es- timation and forecasting in smallholder farming conditions: A case study on Reunion Island. Remote Sens. 6, 6620-6635. Doi: 10.3390/rs6076620 Mulianga, B., A. Bégué, M. Simoes, and P. Todoroff. 2013. Forecast- ing regional sugarcane yield based on time integraland spatial aggregation of MODIS NDVI. Remote Sens. 5, 2184-2199. Doi: 10.3390/rs5052184 Murillo, P.J. and J.A. Carbonell. 2012. Principios y aplicaciones de la percepción remota en el cultivo de la caña de azúcar en Colombia. Servicio de Cooperación Técnica y Transferencia de Tecnología, Cenicaña. Cali, Colombia. Ponzoni, F.J., Y.E. Shimabukuro, and T.M. Kuplich. 2012. Senso- riamento remoto da vegetação. 2nd ed. Oficina de Textos, São Paulo, Brazil. Schmidt, E.J., G. Narciso, P. Frost, and C. Gers. 2000. Application of remote sensing technology in the sa sugar industry review of recent research findings. Proc. S. Afr. Sug. Technol. Ass. 74, 192-200. Tetracam. 2015. ADC Lite. Tetracam’s lightweight ADC ideal for un- manned aircraft. In: Tetracam better science through advanced imaging, http://www.tetracam.com/Products-ADC_Lite.htm; consulted: May 2016. Victoria, J.I., C.A. Viveros, F.A. Salazar, J.C. Ángel, A.E. Bustillo, U. Castro, J. López, and C.A. Moreno. 2013. Catálogo de varie- dades de caña de azúcar. 3rd ed. Servicio de Cooperación Téc- nica y Transferencia de Tecnología, Cenicaña. Cali, Colombia. Vieira, M.A., A.R. Formaggio, C.D. Rennó, C. Atzberger, D.A. Aguiar, and M.P. Mello. 2012. Object based image analysis and data mining applied to a remotely sensed Landsat time-series to map sugarcane over large areas. Remote Sens. Environ. 123, 553-562. Doi: 10.1016/j.rse.2012.04.011