Nadir and Oblique Canopy Reflectance Sensing for N Application in Corn

Transcription

1 An ASABE Meeting Presentation Paper Number: 26 Nadir and Oblique Canopy Reflectance Sensing for N Application in Corn Kenneth A. Sudduth, Agricultural Engineer Newell R. Kitchen, Soil Scientist Scott T. Drummond, IT Specialist USDA-ARS Cropping Systems and Water Quality Research Unit, 269 Agricultural Engineering Bldg., University of Missouri, Columbia, MO 652; contact Ken.Sudduth@ars.usda.gov Written for presentation at the 20 ASABE Annual International Meeting Sponsored by ASABE Gault House Louisville, Kentucky August 7 0, 20 Abstract. Canopy reflectance sensing can be used to assess in-season crop nitrogen (N) health for subsequent control of N fertilization. The several sensor systems that are now commercially available have design and operational differences, including sensed wavelengths, size of the sensed area, and nadir vs. oblique sensing orientation. Data comparing the different sensor designs is lacking. Thus, the objective of this research was to evaluate three different commercial canopy reflectance sensors for N fertilization control in corn. Two units of each of three commercial sensors GreenSeeker, Crop Circle, and CropSpec were mounted to a high-clearance applicator for field data collection. Data were collected multiple times on plots with eight different N rates 0 to 235 kg N ha -. Relative NDVI data from GreenSeeker and Crop Circle sensors were most highly correlated, while data from those two sensors were less strongly related to CropSpec data. CropSpec NDVI was more strongly related to SPAD readings as an indication of leaf N, while the other two sensors were more affected by crop height variations. For multiple data collection runs in a single day, somewhat less run-to-run variation was seen with the CropSpec. For best results, users need to take into account the differences among these commercial sensors, particularly between the two smallfootprint nadir sensors (Crop Circle and GreenSeeker) and the large-footprint, oblique sensor (CropSpec). Keywords. Variable-rate application, Nitrogen, Crop reflectance, Crop canopy sensor The authors are solely responsible for the content of this technical presentation. The technical presentation does not necessarily reflect the official position of the American Society of Agricultural and Biological Engineers (ASABE), and its printing and distribution does not constitute an endorsement of views which may be expressed. Technical presentations are not subject to the formal peer review process by ASABE editorial committees; therefore, they are not to be presented as refereed publications. Citation of this work should state that it is from an ASABE meeting paper. EXAMPLE: Author's Last Name, Initials. 20. Title of Presentation. ASABE Paper No St. Joseph, Mich.: ASABE. For information about securing permission to reprint or reproduce a technical presentation, please contact ASABE at rutter@asabe.org or (2950 Niles Road, St. Joseph, MI USA).

2 Introduction Crop canopy reflectance sensing for nitrogen (N) status assessment and subsequent control of N fertilization has been widely researched and is becoming an accepted practice. Farm magazines describe producer experiences with this technology and in at least one US state (Missouri), farmers can receive government payments for using canopy sensor-controlled N application (USDA-NRCS, 2009). Initial systems, both in the US (Stone et al., 996) and abroad (Heege and Reusch, 996) used passive radiometers that were dependent on ambient sunlight for illumination. Difficulties in compensating for spatiotemporal variation in ambient illumination (Souza et al., 200) led to the development of active sensors with their own illumination source, designed so that the sensor would respond only to reflectance based on the active illumination and not on sunlight. Several active canopy reflectance sensors designed for N application control are commercially available. Considerable research has been directed toward developing algorithms to translate commercial sensor output into N-rate control decisions for crops such as wheat (Raun et al., 2002), corn (Kitchen et al., 200; Solari et al., 2008) and cotton (Olivera, 2008). These efforts have generally been sensor-dependent in that one particular sensor has been used in the research and the findings are specific to that sensor. Some sensor-comparison work has been done (Solari, 2006; Olivera, 2008; Tremblay et al., 2009; Shaver et al., 20) and algorithm conversions have been established in some cases (USDA-NRCS, 2009). However, additional documentation and comparison of the operational characteristics of active crop canopy sensors is desirable. In a 2009 experiment (Sudduth et al., 200), we investigated the performance of three active canopy reflectance sensors available in the US. These sensors were the Holland Scientific Crop Circle ACS-20 (Holland Scientific, Lincoln, NE), N-Tech GreenSeeker Model 505 (Trimble Navigation, Sunnyvale, CA), and Topcon CropSpec (Topcon Precision Agriculture, Mawson Lakes, SA, Australia). In that experiment we found larger-than-expected differences among the three sensors in terms of differences in crop reflectance, the relationship of sensor readings to crop variables, and temporal stability. We suspected that the effects of high variability in plant spacing, disruption of N placement due to excessive rainfall, and small plot sizes may have led to results that were unrepresentative of typical field conditions. Therefore, in 200 we repeated the experiment with larger plot sizes. This paper reports the results of that 200 study. Objective The overall objective of this research was to evaluate the operation of three commercial canopy reflectance sensors during applicator-based data collection on a corn crop. Specific objectives were to determine: Differences in crop reflectance, measured as NDVI, among the three different sensors, The relationship of sensor readings to crop variables, and Temporal variation in sensor readings. 2

3 Materials and Methods Field Site Data were collected in 200 at the University of Missouri South Farm near Columbia, MO where four blocks of sensor test plots received between-row N application soon after planting. Each block consisted of 8 randomly assigned N treatments from 0 to 235 kg N ha - on 34 kg N ha - increments. The 8 plots within each block were 2 rows wide (9. m on a 76 cm row spacing) and 5 m long. Corn was planted at a target rate of 64,250 seeds/ha on 7 June 200. Nitrogen was applied on 24 June 200 with a modified AGCO Spra-Coupe, as described by Kitchen et al. (200). The applicator width was 6 rows; thus two passes were required to fertilize the full width of each plot. Incorporating rainfall was received within 2 days of application. Canopy Reflectance Sensors Data were collected with three commercial sensors Holland Scientific Crop Circle ACS-20, N-Tech GreenSeeker Model 505, and Topcon CropSpec. All three are active sensors, meaning that they have their own light source(s) and use detection technology that minimizes the effect of changes in ambient light on sensor readings. Each sensor emits and measures reflected light at two different wavelengths, one in the visible spectrum and one in the near-infrared spectrum. Specific wavelengths vary among the sensors (table ). The Crop Circle sensor uses a single polychromatic light-emitting diode (LED) for illumination and two separate detectors. Designed for nadir sensing, it has a field of view proportional to height above target (table ), and response has been described as relatively constant over the field of view (Holland et al., 2005). The GreenSeeker sensor, developed based on initial work by Stone et al. (996) uses a separate LED for each wavelength and a single detector. The GreenSeeker field of view is approximately constant over its operating height range; however, Solari (2006) showed that a larger portion of its response comes from the center of the field of view. The CropSpec sensor, based on initial research by Heege and Reusch (996), is designed to view a larger area of crop obliquely (table ). The CropSpec sensor uses two pulsed laser diodes for illumination and a single detector. In contrast to the other two sensors that sense red or amber light in the visible range, the visible channel of the CropSpec senses the red edge of the spectrum, where the reflectance of green vegetation transitions from low (~ 5%) in the visible range to relatively high (~ 30 to 50%) in the near-infrared range. Table. Manufacturer s stated operational characteristics of the crop canopy sensors used in this study. Holland Scientific Crop Circle ACS-20 NTech Industries GreenSeeker Model 505 Topcon CropSpec Visible wavelength 590 ± 5.5 nm 660 ± 5 nm 735 ± 5 nm NIR wavelength 880 ± 0 nm 770 ± 5 nm 805 ± 5 nm Height above target 5 to 2. m to.6 m 2 to 4 m View direction Nadir Nadir Oblique, 45 to 55º Field of view / sensing footprint 32º x 6º 6 x.5 cm (~ constant over height range) 2 to 4 m wide (~ proportional to height from target) 3

4 The GreenSeeker and Crop Circle sensors were mounted to a frame on the front of the applicator above rows 2 and 5 of the six-row application width (fig. ). This frame allowed adjustment of the height of the sensors to maintain their position relative to the crop. The two CropSpec sensors were affixed to the top of the Spra-Coupe cab (fig. ), with one positioned to view each adjoining 6-row data pass. Holland Scientific Crop Circle ACS-20 Topcon CropSpec NTech GreenSeeker Figure. Crop canopy sensors mounted to Spra-Coupe for field data collection. Note that two sensors of each type were used, one on each side of the applicator. Data Collection Data were collected on 6 July and 23 July 200 (table 2) as the Spra-Coupe drove through the plots at approximately 2 to 3 m/s. Multiple data collection runs were attempted on both dates; however, on the first date operational issues meant that only one valid dataset was obtained. Data from the Greenseeker and Crop Circle sensors were recorded on a tablet computer at 0 Hz for further processing. CropSpec data were collected at Hz and GPS position data were obtained at 2 Hz. All sensor data were post-processed to account for sensor and GPS antenna offset, and for the CropSpec, to align with the proper swath. Sensor data from the center 7.5 m of each 5-m-long plot were averaged to represent the plot. Data from each pass of the sensor platform were kept separate, resulting in 64 experimental units (32 plots, two passes per plot). Additional data were obtained at each measurement date for comparison with sensor data. Leaf chlorophyll content was quantified with a Minolta 502 SPAD meter (Konica Minolta Sensing Inc., Osaka, Japan). The SPAD meter was clamped onto the most recently collared leaf, mid-way along the blade of 0 randomly selected plants in each of rows 2 and 5, the same rows sensed by GreenSeeker and Crop Circle, and averaged to get a value per each row. Crop height was calculated as the mean of height to the whorl of 3 randomly selected plants in each of the same two rows. 4

5 Table 2. Data collection information for 200 field tests. Date Time of day Days after planting Plant height, mean and (std. dev.) (m) SPAD reading, mean and (std. dev.) Sensor height above ground (m)* 6 July (0.5) 5.5 (5.7) July July July July (6) 46.5 (6.) 2.34 * Height of Crop Circle and GreenSeeker sensors. CropSpec sensor height was 3.5 m. Results and Discussion NDVI Differences among Sensors Response of the different sensors was compared on the basis of relative normalized difference vegetative index (NDVI), an index commonly used in sensor-based variable N algorithms. In this calculation, each NDVI value was ratioed to the mean NDVI obtained from the high-n (235 kg N ha - ) plot areas during that measurement run. Other algorithms use the simple ratio (SR) of the NIR to the visible channel, or the inverse simple ratio (ISR). These can be easily calculated from NDVI, as ISR = ( NDVI)/( + NDVI). Means of the two sensors of each type were used in the comparison, after pre-screening of individual sensor data streams to eliminate any spurious, out-of range data (<< % of data points were eliminated). Previous research (Roberts et al., 2009) showed an advantage to using mean data from multiple sensors to guide a single application rate in contrast to controlling multiple boom sections individually, each based on a single sensor reading. NDVI comparisons were done separately for each data collection pass, resulting in a maximum of 64 comparisons (two passes per plot and 32 plots). Gain setting problems with the CropSpec sensor resulted in fewer comparisons on 23 July. Data are shown (fig. 2) from the 0944 measurement run on 23 July; results were representative of all runs on that date. As might be expected, readings from the two nadir-looking, small-footprint sensors (GreenSeeker and Crop Circle) were most closely related (table 3), with r 0.9, while a weaker relationship was seen with either of the nadir sensors and the oblique, large-footprint sensor (CropSpec). In most cases, relationships among the different sensors were stronger (higher r) and more consistent (slope closer to one) for 23 July, when canopy closure was more complete (table 3). Varying effects of a mixed soil-crop view on the different sensors likely contributed to the higher variability on 6 July. Differences in signal-to-noise ratio among the sensors (Solari, 2006) may have been another contributor. Variation among sensors was considerably lower in this dataset than it was in our similar 2009 experiment (Sudduth et al., 200). We attribute this improvement to the following factors: () a highly variable corn stand in 2009 may have affected the different sensor types somewhat differently, (2) excessive rainfall in 2009 blurred the boundaries between N treatments, and (3) larger plots used in 200 provided more sensor data and therefore a more robust plot average. 5

6 .2 6 July, 450 Data Run 23 July, 944 Data Run Y = * X - 7 R2 =4.2 Y =.27 * X n=55 R2 =0.73 NDVI CropSpec NDVI CropSpec.2 NDVI Crop Circle.2 NDVI Crop Circle.2.2 Y =.38 * X R2 =0.90 Y =. * X R2 =0.99 NDVI GreenSeeker NDVI GreenSeeker.2 NDVI Crop Circle.2 NDVI Crop Circle.2.2 Y = * X R2 =0.55 Y =.4 * X - 26 n=55 R2 =0.73 NDVI CropSpec NDVI CropSpec.2 NDVI GreenSeeker.2 NDVI GreenSeeker Figure 2. Relative NDVI relationships among the three canopy sensors. Relative NDVI was calculated as the ratio to the high-n reference plot average NDVI for each measurement run. 6

7 Table 3. Pearson correlation coefficients between sensor NDVI readings for individual datasets. Dataset CropSpec vs. Crop Circle Correlation coefficient (r) Crop Circle vs. GreenSeeker CropSpec vs. GreenSeeker 6 July July Sensor Data vs. Chlorophyll and Crop Height On 23 July, at 46 days after planting (DAP), data from all three sensors were related to variations in both chlorophyll, as measured by SPAD meter, and crop height (fig. 3). Data from the small-footprint, nadir-looking sensors were more strongly related to crop height, while data from the large-footprint, oblique-looking sensor were more strongly related to SPAD reading. These findings were consistent across both measurement dates (table 4) with correlation of sensor data to plant biophysical variables stronger at the later measurement date. The finding that nadir-looking crop sensors respond to both chlorophyll and crop height is consistent with prior research, including our 2009 study (Sudduth et al., 200). Solari (2006) reported GreenSeeker and Crop Circle data were highly affected by biomass and crop height. Jones et al. (2007) reported chlorophyll estimates from GreenSeeker NDVI were improved when height measured by an ultrasonic distance sensor was included as a second term in a linear regression. As corn crop height is strongly related to biomass during vegetative growth stages (Freeman et al., 2007) and chlorophyll content measured as SPAD is a good estimator of corn leaf N concentration (e.g., Blackmer et al., 994, r 2 = 4; Sudduth et al., 200, r 2 = 4), the product of crop height and SPAD could be reasonably expected to relate to total plant N content. Crop Circle and GreenSeeker data were more predictive of this product than they were of either parameter alone (table 3). However, CropSpec data was more strongly related to chlorophyll content than to the product of chlorophyll and height. Although these differences in sensitivity among the sensors were relatively small in this experiment, they point out the need for further research to evaluate whether existing algorithms developed for Crop Circle or GreenSeeker data could be used directly with the CropSpec sensor. There are several possible explanations for why data from the CropSpec sensor was more strongly related to leaf chlorophyll concentration (or SPAD) and less related to crop height than data from the other sensors. It may be that unique operational characteristics of the CropSpec (e.g., large sensing footprint, oblique viewing angle) may have made data from this instrument less affected by variation in crop size or height, as little if any soil was viewed by this sensor. Additionally, the oblique orientation may have allowed the CropSpec to better sense leaves lower on the plant, which will generally be more responsive to signs of N stress. Alternatively, it may be that inclusion of the red edge (735 nm) data made this sensor more sensitive to chlorophyll. Hatfield et al. (2008) noted that a reflectance index combining NIR and red edge data had the lowest error in estimating leaf chlorophyll content compared to all other visible 7

8 wavelengths. It is worth noting that these results were with the chlorophyll index (CI = NIR/VIS - ), which has been shown to be superior to NDVI using the same wavelengths for chlorophyll estimation (Solari et al., 2008). Comparison of sensor results on the basis of CI rather than NDVI would be desirable. Table 4. Pearson correlation coefficients between sensor NDVI readings and SPAD, corn height, and product of SPAD x height. SPAD Height (m) SPAD x height 6 July GreenSeeker Crop Circle CropSpec July GreenSeeker Crop Circle CropSpec Temporal Stability The four sets of data collected on 23 July (table 2) allowed assessment of the temporal variation in sensor readings (fig. 4). When using the second run of the day as a reference, the three other runs fell near the : line for the CropSpec. Only two comparison runs were available for GreenSeeker and Crop Circle because a non-standard sensor orientation was used during the 0903 data run; for these two comparisons, data also fell near the : line (fig. 4). Correlations between individual runs ranged between 0.96 and 0.99 for the CropSpec, between 0.92 and 0.99 for the GreenSeeker, and between 0.96 and 0.99 for the Crop Circle. In all cases the data runs furthest apart in time (i.e., 0944 and 452) were the most different. Differences between runs for the CropSpec sensor with the larger sensing footprint would likely have been less affected by run-to-run variations in driving position relative to the crop rows, plant movement due to wind, and other potential sources of variability. Differences from run to run were much less than in a previous, similar experiment (Sudduth et al., 200), particularly for the smallfootprint sensors. The plant stand and N application issues noted earlier for that previous experiment were likely causes for the greater temporal variation seen in that data. It is well known that vegetative indices from passive (i.e., based on ambient light) sensors can vary widely over the course of a day, even if the sensor is stationary above the canopy (e.g., Souza et al., 200). Temporal variation in active sensor (Crop Circle and GreenSeeker) vegetation indices has also been reported (Olivera, 2008). It is unknown if the source of variation is the sensor (e.g., influence of variations in ambient light), external plant considerations (e.g., leaf surface moisture, plant movement), physiological changes in the plant itself, or some combination thereof. Although the temporal variation in relative NDVI was small in this experiment, it is possible that the different active sensors in this study may have been affected by these issues, and possibly to different degrees. Further research directed toward understanding the relative effect of various error sources on mobilized sensor data collection would be warranted. 8

9 .2 Y = 0.05 * X + 07 R2 = Y = 0.38 * X R2 =2 NDVI Crop Circle NDVI Crop Circle SPAD Reading Corn Height (m).2 Y = * X R2 = Y = 72 * X R2 =0.54 NDVI CropSpec NDVI CropSpec SPAD Reading Corn Height (m).2 Y = * X R2 = Y=30*X+89 R2 =3 NDVI GreenSeeker NDVI GreenSeeker SPAD Reading Corn Height (m) Figure 3. Relationship of SPAD reading (left) and corn height (right) to NDVI measured on 23 July 200 using GreenSeeker, Crop Circle, and CropSpec canopy sensors. 9

10 .2.2 GreenSeeker NDVI : line CropSpec NDVI : line.2.2 GreenSeeker NDVI, 944 run.2 CropSpec NDVI, 944 run Crop Circle NDVI : line 27 run 452 run 903 run.2 Crop Circle NDVI, 944 run Figure 4. Temporal stability of plot-average canopy sensor readings. Other data runs on 23 July are shown as a function of second (0944) run. There were only two comparisons for GreenSeeker and Crop Circle due to non-standard sensor orientations used in the 0903 run. Conclusion In this study, three different commercial crop canopy sensors were evaluated for their ability to discriminate reflectance differences related to corn N health. Reflectance data were processed to average (~ 3.6 x 7.5 m) NDVI values for analysis. Data from GreenSeeker and Crop Circle sensors were highly correlated (r 0.9), while data from those two sensors were less strongly related to CropSpec data (0.74 r 0.90). While substitution between GreenSeeker and Crop Circle data is feasible (and has been done) in N recommendation algorithms, it may be less reliable to make a similar substitution with CropSpec data. CropSpec NDVI was more strongly related to SPAD as an indication of leaf N content, while the other two sensors were more related to crop height variations. The possibility of improving these relationships through inclusion of crop height data (e.g., from an ultrasonic distance sensor) should be investigated. For multiple data collection runs in a single day, somewhat less run-to-run variation was seen 0

11 with the CropSpec. This was likely due to the averaging effects of its larger sensing footprint which would be less sensitive to driving inaccuracies, crop movement, and small-scale variability. For best results, users need to take into account the differences among these three commercial sensors, particularly between the two small-footprint nadir sensors (Crop Circle and GreenSeeker) and the large-footprint, oblique sensor (CropSpec). Acknowledgements and Disclaimer We acknowledge the following for assistance in data collection: Bob Mahurin, Kurt Holiman, Matt Volkmann, Eric Allphin, Alec Sheridan, and Anna Hodge. Mention of trade names or commercial products in this paper is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the United States Department of Agriculture. References Blackmer, T.M., J.S. Schepers, and G.E. Varvel Light reflectance compared with other nitrogen stress measurements in corn leaves. Agron. J. 86: Freeman, K.W., K. Girma, D.B. Arnall, R.W. Mullen, K.L. Martin, R.K. Teal, and W.R. Raun By-plant prediction of corn forage biomass and nitrogen uptake at various growth stages using remote sensing and plant height. Agron. J. 99: Hatfield, J.L., A.A. Gitelson, J.S. Schepers, and C.L. Walthall Application of spectral remote sensing for agronomic decisions. Agron. J. S7-S3. Heege, H.J. and S. Reusch Sensor for on the go control of site specific nitrogen top dressing. ASAE Paper No St. Joseph, Mich.: ASABE. Holland, K.H., J.S. Schepers, J.F. Shanahan, and G.L. Horst Plant canopy sensor with modulated polychromatic light source. In Proc. 7th Intl. Conf. on Precision Agriculture. D.J. Mulla, ed. Minneapolis, Minn.: Precision Agriculture Center, University of Minnesota. Jones, C.L., N.O. Maness, M.L. Stone, and R. Jayasekara Chlorophyll estimation using multispectral reflectance and height sensing. Trans. ASABE 50(5): Kitchen, N.R., K.A. Sudduth, S.T. Drummond, P.C. Scharf, H.L. Palm, D.F. Roberts, and E.D. Vories Ground-based canopy reflectance sensing for variable-rate nitrogen corn fertilization. Agron J. 02:7-84. Olivera, L.F Reflectance sensors to predict mid-season nitrogen need of cotton. MS thesis. Columbia, Mo.: University of Missouri. Raun, W.R., J.B. Solie, G.V. Johnson, M.L. Stone, R.W. Mullen, K.W. Freeman, W.E. Thomason, and E.V. Lukina Improving nitrogen use efficiency in cereal grain production with optical sensing and variable rate application. Agron. J. 94: Roberts, D.F., V.I. Adamchuk, J.F. Shanahan, R.B. Ferguson, and J.S. Schepers Optimization of crop canopy sensor placement for measuring nitrogen status in corn. Agron. J. 0: Shaver, T.M., R. Khosla, and D.G. Westfall. 20. Evaluation of two crop canopy sensors for nitrogen variability determination in irrigated maize. Precision Agric. DOI: 0.007/s Solari, F Developing a crop based strategy for on-the-go nitrogen management in irrigated cornfields. PhD diss. Lincoln, Neb.: University of Nebraska.

12 Solari, F., J. Shanahan, R. Ferguson, J. Schepers, and A. Gitelson Active sensor reflectance measurements of corn nitrogen status and yield potential. Agron. J. 00: Souza, E.G., P.C. Scharf, and K.A. Sudduth The influence of sun position and clouds on reflectance and vegetation indices of greenhouse-grown corn. Agron. J. 02: Stone, M.L., J.B. Solie, W.R. Raun, R.W. Whitney, S.L. Taylor, and J.D. Ringer Use of spectral radiance for correcting in-season fertilizer nitrogen deficiencies in winter wheat. Trans. ASAE 39: Sudduth, K.A., N.R. Kitchen, and S.T. Drummond 200. Comparison of three canopy reflectance sensors for variable-rate nitrogen application in corn. In Proc. 0th Intl. Conf. on Precision Agriculture. Ft. Collins, Colo.: Colorado State University. Tremblay, N., Z. Wang, B.-L. Ma, C. Belec, and P. Vigneault A comparison of crop data measured by two commercial sensors for variable-rate nitrogen application. Precision Agric. 0: USDA-NRCS Variable-Rate Nitrogen Fertilizer Application in Corn Using In-field Sensing of Leaves or Canopy. Missouri NRCS Agronomy Tech Note 35. Available at: %20MO-35.pdf. 2