Physics-Based Detection of Subpixel Targets in Hyperspectral Imagery
Hyperspectral and Thermal Imagery for the Detection of Nitrogen … · 2020-04-15 · Hyperspectral...
1
Hyperspectral and Thermal Imagery for the Detection of Nitrogen and Water Stress Tyler Nigon 1 , Carl Rosen 1 , David Mulla 1 , Yafit Cohen 2 , Victor Alchanatis 2 , & Ronit Rud 2 1 Dept. of Soil, Water, & Climate, University of Minnesota, St. Paul, MN 2 Agricultural Research Organization, Volcani Center, Bet-Dagan, Israel INTRODUCTION Potato yield and quality are highly dependent on nitrogen (N) as a fertilizer source; however, excessive N uptake can result in reduced tuber yield and quality. Furthermore, the naturally shallow and poorly developed root system of the potato plant, coupled with a high N requirement and sensitivity to water stress on coarse-textured soils increases the risk for nitrate (NO 3 -N) leaching. Early season N applications have a high potential to be leached below the rooting zone, especially under irrigation (Errebhi et al., 1998). To optimize N uptake and minimize environmental losses, many growers apply post- emergent N fertilizer in split applications implemented with precise irrigation management. To help decide when to apply N and water inputs, growers take tissue samples or use sensors to take point measurements that are indicative of the N or water stress of the crop. However, spatial variability often exists within fields because of differences in soil characteristics or landscape position and a few point measurements are not likely to be an accurate representation of the entire field. Hyperspectral (HS) and thermal aerial imagery is an effective research tool that can be used to develop new techniques to detect N and/or water stress in a crop because of its ability to detect spatial and temporal variability. Modified N and/or water stress methodologies can be applied to active sensors to optimize the efficiency of variable rate application. OBJECTIVES Evaluate the ability of canopy level HS and thermal imagery to detect N and water status of a potato crop Identify specific wavelengths and/or indices that best predict N stress from potato canopy spectral reflectance Examine the effects of variety and growth stage on the indices that best predict N stress in potatoes MATERIALS & METHODS • Two year field study was established in 2010 (only 2010 data presented). 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Coefficient of Determination (R 2 ) Reflectance Derivative 704 nm 751 nm • Variety effect on petiole NO 3 -N & leaf N was significant on each date, so regression analysis was done between varieties (ANOVA not shown). • Particular wavelengths are better able to detect N stress than others. Best correlations between leaf N and canopy reflectance for RB on July 1 occurred from 500-610 nm, around 704 nm (red edge), and from 750-900 nm (red edge & NIR) (Fig. 2, depicted by the maroon line). Indices for comparison were selected if: (1) they utilize wavelengths that are shown to detect N stress based on this data and (2) they showed success in previous studies from literature. Indices compared were: SR8 (Datt, 1998), NDVI (Rouse et al., 1973), GNDVI (Gitelson et al., 1996), NDI1 (Datt, 1999), TCARI & TCARI/OSAVI (Haboudane et al., 2002), TCARI[R751] & TCARI/OSAVI[R751] (Wu et al., 2008), and the first derivative of canopy reflectance (all wavelengths). Best correlations between leaf N and the derivative reflectance for RB on July 1 occurred around 751 nm (Fig. 2, depicted by the gold line). Derivative reflectance at these wavelengths are better correlated to leaf N than raw reflectance was at any wavelength. • Among conventional indices, TCARI[R751] was best on July 1 for RB, while NDI1 was best for AR. On August 6, SR8 was best (Table 1). • The only index that performed the best across multiple scenarios was SR8 (on August 6 it performed best for all scenarios except when petiole NO 3 -N was used as a reference for AR) (Table 1). Table 1. Coefficient of determination (R 2 ) values of tested indices for each tissue sample, each variety, and each flyover date; shaded boxes highlight the R 2 value of the best performing index for each scenario. Fig. 4. First derivative image (at 751 nm) of treatment plots from July 1; photograph is 0.001 0.003 0.004 0.006 Canopy Derivative Reflectance at 751 nm Index/Wavelength Petiole NO 3 -N Leaf N R. Burbank Alpine R. R. Burbank Alpine R. July 1 SR8 0.69 0.46 0.79 0.74 NDI1 0.64 0.41 0.82 0.80 TCARI (R751) 0.73 0.52 0.77 0.70 Derivative (509nm) 0.53 0.39 0.70 0.55 Derivative (751nm) 0.69 0.48 0.85 0.80 Derivative (761nm) 0.59 0.54 0.72 0.73 August 6 SR8 0.63 0.52 0.68 0.72 NDI1 0.49 0.39 0.58 0.63 TCARI (R751) 0.02 0.09 0.07 0.01 Derivative (509nm) 0.45 0.54 0.31 0.71 Derivative (751nm) 0.38 0.33 0.48 0.51 Derivative (761nm) 0.47 0.37 0.54 0.50 • Location & Soil: Sand Plain Research Farm (Becker, MN); Hubbard loamy sand • Experimental design: randomized complete block design with a split-split plot restriction on randomization (4 replications) Two irrigation treatments (whole plot): Non-stressed: water applied according to the checkbook method Stressed: water applied at ~80% of non-stressed rate Five N treatments (sub plot): (In kg ha -1 ): 34 starter only control, 180 split, 270 split, 270 split + soil surfactant, and 270 early Two varieties (sub-sub plot): Russet Burbank (RB) and Alpine Russet (AR) • Images were acquired on July 1 and August 6 with AISA-Eagle VNIR hyperspectral imager (CALMIT, University of Nebraska, Lincoln). • Ground-based data collection included tissue sampling (petiole NO 3 -N and leaf N) and tuber yield. • Linear regression was used to compare the relationship between image reflectance (at every wavelength) and tissue sample (petiole NO 3 –N and leaf N) for each flyover date and variety. • Optimum wavelengths for detecting N stress were selected by comparing coefficient of determination (R 2 ) values (Fig. 2). 40 45 50 55 60 65 70 34 (control) 180 split 270 split 270 split (+ surfactant) 270 early Total Yield (MT/ha) N Rate/Timing (kg N ha -1 ) C B A B AB Fig. 1. Effect of N treatment on total tuber yield. Mean comparisons were done at α = 0.05. 0.0 0.1 0.2 0.3 400 500 600 700 800 900 1000 C Wavelength (nm) 704 nm 554 nm Fig. 2. Coefficient of determination (R 2 ) plotted versus wavelength for the relationship between leaf N and canopy reflectance/canopy reflectance left derivative at all AISA Eagle wavelengths for the Russet Burbank variety on the July 1 flyover. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 400 500 600 700 800 900 1000 Coefficient of Determination (R 2 ) Wavelength (nm) July 1 Russet Burbank July 1 Alpine Russet August 6 Russet Burbank August 6 Alpine Russet 554 nm 573 nm Fig. 3. Coefficient of determination (R 2 ) plotted versus wavelength for the relationship between leaf N and canopy reflectance at all AISA Eagle wavelengths for each variety and each flyover date. • Optimum wavelengths to predict N stress differ depending on variety (Fig. 3, similar colored lines are different varieties on the same date for relationship between reflectance and leaf N). In the visible spectrum for July 1, R 2 values peak at 573 nm for AR, whereas they are relatively constant for RB. • Optimum wavelengths to predict N stress also differ depending on date (Fig. 3, similar shaped lines [solid vs. broken] are similar varieties on different dates for relationship between reflectance and leaf N). In the visible spectrum for RB, R 2 values are considerably lower around 554 nm for August 6 than they are for July 1. CONCLUSIONS • The best index to detect N stress depends on variety, growth stage, and whether petiole NO 3 -N or leaf N was used as a reference. • Overall, reflectance is better correlated with leaf N than petiole NO 3 -N. • Indices formulated from the first derivative of canopy reflectance have potential to increase accuracy when predicting N stress in potatoes. ACKNOWLEDGEMENTS • Funding by: Binational Agricultural Research and Development Fund (BARD) Minnesota Area II Potato Growers Hueg-Harrison Fellowship • Matt McNearny, Ron Faber, Kristin Petersen, Tiffanie Stone, Dan Fischer, & Anina Christensen for assistance with field and laboratory work Fig. 4. First derivative image (at 751 nm) of treatment plots from July 1; photograph is a starter only N control plot and corresponds to plot outlined in the image. RESULTS & DISCUSSION • N rate effect on tuber yield was significant (Fig. 1). 25 30 34 42 Temperature (Degrees Celcius) Non-stressed Irrigation Stressed Irrigation Fig. 5. Thermal image of the treatment plots from July 1. • Overall, the best indicator of N stress was the first derivative at 751 nm on July 1 for RB when leaf N was used as a reference (Table 1, Fig. 4). • Indices applied to thermal image data can help detect water stress in a potato crop. Thermal data processing is in the beginning stages for this study, but canopy temperature differences can be seen in the raw image (Fig. 5). REFERENCES Datt, B. (1998). Remote sensing of chlorophyll a, chlorophyll b, chlorophyll a b, and total carotenoid content in eucalyptus leaves. Remote Sensing of Environment, 66(2), 111-121. Datt, B. (1999). Visible/near infrared reflectance and chlorophyll content in eucalyptus leaves. InternationalJournalof Remote Sensing, 20(14), 2741-2759. Errebhi, M., Rosen, C. J., Gupta, S. C., & Birong, D. E. (1998). Potato yield response and nitrate leaching as influenced by nitrogen management. Agronomy Journal, 90(1), 10-15. Gitelson, A. A., Kaufman, Y. J., & Merzlyak, M. N. (1996). Use of a green channel in remote sensing of global vegetation from EOS-MODIS. Remote Sensing of Environment, 58(3), 289-298. Haboudane, D., Miller, J. R., Tremblay, N., Zarco-Tejada, P. J., & Dextraze, L. (2002). Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture. Remote Sensing of Environment, 81(2-3), 416-426. Rouse, J. W., Haas, R. H., Schell, J. A., & Deering, D. W. (1974). Monitoring vegetation systems in the great plains with ERTS. In N. SP-351, Ed. Third ERTS symposium, vol. 1 (pp. 309-317). Washington: NASA. Wu, C., Niu, Z., Tang, Q., & Huang, W. (2008). Estimating chlorophyll content from hyperspectral vegetation indices: Modeling and validation. Agriculturaland Forest Meteorology,148(8-9), 1230-1241.
Transcript of Hyperspectral and Thermal Imagery for the Detection of Nitrogen … · 2020-04-15 · Hyperspectral...
Hyperspectral and Thermal Imagery for the Detection of Nitrogen and Water StressTyler Nigon1, Carl Rosen1, David Mulla1, Yafit Cohen2, Victor Alchanatis2, & Ronit Rud2
1Dept. of Soil, Water, & Climate, University of Minnesota, St. Paul, MN2Agricultural Research Organization, Volcani Center, Bet-Dagan, Israel
INTRODUCTIONPotato yield and quality are highly dependent on nitrogen (N) as a fertilizersource; however, excessive N uptake can result in reduced tuber yieldandquality. Furthermore, the naturally shallow and poorly developed root systemof the potato plant, coupled with a high N requirement and sensitivity to waterstress on coarse-textured soils increases the risk for nitrate(NO3-N) leaching.Early season N applications have a high potential to be leached below therooting zone, especially under irrigation (Errebhi et al., 1998). To optimize Nuptake and minimize environmental losses, many growers apply post-emergent N fertilizer in split applications implemented with precise irrigationmanagement. To help decide when to apply N and water inputs, growers taketissue samples or use sensors to take point measurements that areindicative ofthe N or water stress of the crop. However, spatial variability often existswithin fields because of differences in soil characteristics orlandscapeposition and a few point measurements are not likely to be an accuraterepresentation of the entire field. Hyperspectral (HS) and thermal aerialimagery is an effective research tool that can be used to developnewtechniques to detect N and/or water stress in a crop because of its ability todetect spatial and temporal variability. Modified N and/or waterstressmethodologies can be applied to active sensors to optimize the efficiency ofvariable rate application.
OBJECTIVES�Evaluate the ability of canopy level HS and thermal imagery to detect N
and water status of a potato crop
� Identify specific wavelengths and/or indices that best predict N stressfrom potato canopy spectral reflectance
� Examine the effects of variety and growth stage on the indices that bestpredict N stress in potatoes
MATERIALS & METHODS•Two yearfield studywasestablishedin 2010(only 2010datapresented). 0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Coe
ffic
ient
of D
eter
min
atio
n (R
2 )
ReflectanceDerivative
704 nm
751 nm
•Variety effect on petiole NO3-N & leaf N was significant on each date, soregression analysis was done between varieties (ANOVA not shown).
•Particular wavelengths are better able to detect N stress than others.
�Best correlations between leaf N and canopy reflectance for RB on July 1occurred from 500-610 nm, around 704 nm (red edge), and from 750-900nm (red edge & NIR) (Fig. 2, depicted by the maroon line).
� Indices for comparison were selected if: (1) they utilize wavelengths thatare shown to detect N stress based on this data and (2) they showedsuccess in previous studies from literature.
�Indices compared were: SR8 (Datt, 1998), NDVI (Rouse et al., 1973),GNDVI (Gitelson et al., 1996), NDI1 (Datt, 1999), TCARI &TCARI/OSAVI (Haboudane et al., 2002), TCARI[R751] &TCARI/OSAVI[R751] (Wu et al., 2008), and the first derivative ofcanopy reflectance (all wavelengths).
�Best correlations between leaf N and the derivative reflectance for RB onJuly 1 occurred around 751 nm (Fig. 2, depicted by the gold line).
�Derivative reflectance at these wavelengths are better correlated to leafN than raw reflectance was at any wavelength.
•Among conventional indices, TCARI[R751] was best on July 1 for RB,while NDI1 was best for AR. On August 6, SR8 was best (Table 1).
•The only index that performed the best across multiple scenarioswas SR8(on August 6 it performed best for all scenarios except when petioleNO3-Nwas used as a reference for AR) (Table 1).
Table 1. Coefficient of determination (R2) values of tested indices for eachtissue sample, each variety, and each flyover date; shaded boxes highlight theR2 value of the best performing index for each scenario.
Fig. 4. First derivative image (at 751 nm) of treatment plots from July 1; photograph is
0.001 0.003 0.004 0.006
Canopy Derivative Reflectance at 751 nm
Index/Wavelength Petiole NO3-N Leaf NR. Burbank Alpine R. R. Burbank Alpine R.
July 1 SR8 0.69 0.46 0.79 0.74NDI1 0.64 0.41 0.82 0.80TCARI (R751) 0.73 0.52 0.77 0.70Derivative (509nm) 0.53 0.39 0.70 0.55Derivative (751nm) 0.69 0.48 0.85 0.80Derivative (761nm) 0.59 0.54 0.72 0.73
August 6 SR8 0.63 0.52 0.68 0.72NDI1 0.49 0.39 0.58 0.63TCARI (R751) 0.02 0.09 0.07 0.01Derivative (509nm) 0.45 0.54 0.31 0.71Derivative (751nm) 0.38 0.33 0.48 0.51Derivative (761nm) 0.47 0.37 0.54 0.50
•Two yearfield studywasestablishedin 2010(only 2010datapresented).
•Location & Soil: Sand Plain Research Farm (Becker, MN); Hubbard loamysand
•Experimental design: randomized complete block design with a split-splitplot restriction on randomization (4 replications)
�Two irrigation treatments (whole plot):�Non-stressed: water applied according to the checkbook method�Stressed: water applied at ~80% of non-stressed rate
�Five N treatments (sub plot):�(In kg ha-1): 34 starter only control, 180 split, 270 split, 270 split + soilsurfactant, and 270 early
�Two varieties (sub-sub plot):�Russet Burbank (RB) and Alpine Russet (AR)
•Images were acquired on July 1 and August 6 with AISA-Eagle VNIRhyperspectral imager (CALMIT, University of Nebraska, Lincoln).
•Ground-based data collection included tissue sampling (petiole NO3-N andleaf N) and tuber yield.
•Linear regression was used to compare the relationship between imagereflectance (at every wavelength) and tissue sample (petiole NO3–N and leafN) for each flyover date and variety.
•Optimum wavelengths for detecting N stress were selected by comparingcoefficient of determination (R2) values (Fig. 2).
40
45
50
55
60
65
70
34 (control) 180 split 270 split 270 split (+ surfactant)
270 early
Tota
l Yie
ld (
MT
/ha)
N Rate/Timing (kg N ha-1)
C
BA
BAB
Fig. 1. Effect of N treatment on total tuber yield. Mean comparisonswere done at α = 0.05.
0.0
0.1
0.2
0.3
400 500 600 700 800 900 1000
Coe
ffic
ient
of D
eter
min
atio
n (
Wavelength (nm)
704 nm554 nm
Fig. 2. Coefficient of determination (R2) plotted versus wavelength for the relationshipbetween leaf N and canopy reflectance/canopy reflectance left derivative at all AISAEagle wavelengths for the Russet Burbank variety on the July 1 flyover.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
400 500 600 700 800 900 1000
Coe
ffic
ient
of
Det
erm
inat
ion
(R2 )
Wavelength (nm)
July 1 Russet Burbank July 1 Alpine RussetAugust 6 Russet Burbank August 6 Alpine Russet
554 nm
573 nm
Fig. 3. Coefficient of determination (R2) plotted versus wavelength for the relationshipbetween leaf N and canopy reflectance at all AISA Eagle wavelengths for each varietyand each flyover date.
•Optimum wavelengths to predict N stress differ depending on variety (Fig.3, similar colored lines are different varieties on the same date forrelationship between reflectance and leaf N).
� In the visible spectrum for July 1, R2 values peak at 573 nm for AR,whereas they are relatively constant for RB.
•Optimum wavelengths to predict N stress also differ depending on date (Fig.3, similar shaped lines [solid vs. broken] are similar varietieson differentdates for relationship between reflectance and leaf N).
� In the visible spectrum for RB, R2 values are considerably lower around554 nm for August 6 than they are for July 1. CONCLUSIONS
•The best index to detect N stress depends on variety, growth stage, andwhether petiole NO3-N or leaf N was used as a reference.
•Overall, reflectance is better correlated with leaf N than petiole NO3-N.
•Indices formulated from the first derivative of canopy reflectance havepotential to increase accuracy when predicting N stress in potatoes.
ACKNOWLEDGEMENTS• Funding by:
� Binational Agricultural Research and Development Fund (BARD)
� Minnesota Area II Potato Growers
� Hueg-Harrison Fellowship
• Matt McNearny, Ron Faber, Kristin Petersen, Tiffanie Stone, Dan Fischer, & Anina Christensen for assistance withfield and laboratory work
Fig. 4. First derivative image (at 751 nm) of treatment plots from July 1; photograph isa starter only N control plot and corresponds to plot outlined in the image.
RESULTS & DISCUSSION•N rate effect on tuber yield was significant (Fig. 1).
25 30 34 42
Temperature (Degrees Celcius)
Non-stressed Irrigation
Stressed Irrigation
Fig. 5. Thermal image of the treatment plots from July 1.
•Overall, the best indicator of N stress was the first derivative at 751 nm onJuly 1 for RB when leaf N was used as a reference (Table 1, Fig. 4).
•Indices applied to thermal image data can help detect water stress in a potatocrop. Thermal data processing is in the beginning stages for this study, butcanopy temperature differences can be seen in the raw image (Fig. 5).
REFERENCESDatt, B. (1998). Remote sensing of chlorophyll a, chlorophyll b, chlorophyll a b, and total carotenoid content in eucalyptus leaves.Remote Sensing of
Environment, 66(2), 111-121.Datt, B. (1999). Visible/near infrared reflectance and chlorophyll content in eucalyptus leaves.InternationalJournalof Remote Sensing, 20(14), 2741-2759.Errebhi, M., Rosen, C. J., Gupta, S. C., & Birong, D. E. (1998). Potato yield response and nitrate leaching as influenced by nitrogen management.Agronomy
Journal, 90(1), 10-15.Gitelson, A. A., Kaufman, Y. J., & Merzlyak, M. N. (1996). Useof a green channel in remote sensing of global vegetation from EOS-MODIS.Remote
Sensing of Environment, 58(3), 289-298.Haboudane, D., Miller, J. R., Tremblay, N., Zarco-Tejada, P. J., & Dextraze, L. (2002). Integrated narrow-band vegetation indices for prediction of crop
chlorophyll content for application to precision agriculture. Remote Sensing of Environment, 81(2-3), 416-426.Rouse, J. W., Haas, R. H., Schell, J. A., & Deering, D. W. (1974). Monitoring vegetation systems in the great plains with ERTS. In N. SP-351, Ed. Third
ERTS symposium, vol. 1 (pp. 309-317). Washington: NASA.Wu, C., Niu, Z., Tang, Q., & Huang, W. (2008). Estimating chlorophyll content from hyperspectral vegetation indices: Modeling and validation.
Agriculturaland Forest Meteorology, 148(8-9), 1230-1241.
