- Email: [email protected]
Estimating Triticale Dry Matter Yield in Parcel Plot Trials From Aerial and Ground Based Spectral Measurements
5th 5th IFAC IFAC Conference Conference on on Sensing, Sensing, Control Control and and Automation Automation for for Agriculture 5th IFAC Conference on Sensing, Control and Automation for Agriculture 5th IFAC Conference on Sensing, Control and Automation for Available August 14-17, 14-17, 2016. 2016. Seattle, Seattle, Washington, Washington, USA online at www.sciencedirect.com Agriculture August USA Agriculture August 14-17, 2016. Seattle, Washington, USA August 14-17, 2016. Seattle, Washington, USA
ScienceDirect
IFAC-PapersOnLine 49-16 (2016) 404–408
Estimating Triticale Dry Matter Yield in Parcel Plot Trials Estimating Triticale Dry Yield in Plot Estimating Triticale Dry Matter Matter in Parcel Parcel Plot Trials Trials From Aerial and Ground BasedYield Spectral Measurements From Aerial and Ground Based Spectral Measurements From Aerial and Ground Based Spectral Measurements P.O. Noack* Noack* P.O. P.O. Noack* P.O. Noack* *University *University of of Applied Applied Sciences Sciences Weihenstephan-Triesdorf, Weihenstephan-Triesdorf, *University of Applied Markgrafenstr. 16 ,, Sciences D-91746 Weihenstephan-Triesdorf, Weidenbach, Germany Germany Markgrafenstr. 16 D-91746 Weidenbach, *University of Applied Sciences Weihenstephan-Triesdorf, Markgrafenstr. 16 , D-91746 Germany (Tel: +49-9826-654242; e-mail:Weidenbach, [email protected]) (Tel: +49-9826-654242; e-mail: [email protected]) Markgrafenstr. 16 , D-91746 Weidenbach, Germany (Tel: +49-9826-654242; e-mail: [email protected]) (Tel: +49-9826-654242; e-mail: [email protected])
Abstract: Spectral Spectral measurement measurement devices devices are are aa useful useful tool tool for for quantifying quantifying biomass biomass and and the the nitrogen nitrogen Abstract: Abstract: Spectral measurement devices arespecific a useful tool for quantifying biomass and the uptake of plants. plants. They are employed employed for site site specific nitrogen application in arable arable farming and nitrogen for crop crop uptake of They are for nitrogen application in farming and for Abstract: Spectral measurement devices are a useful tool for quantifying biomass and the nitrogen uptake ofin plants. They are employed for site specific nitrogen application in arable farming andhas for been crop scouting parcel plot trials. In 2014 aa plot trial with varying fertilizer types and levels scouting in parcel plot trials. In 2014 plot trial with varying fertilizer types and levels has been uptake of plants. They are employed for site specific nitrogen application in arable farming and for crop scouting in parcel plot trials. In 2014 a plot trial with varying fertilizer types and levels has been monitored with a low-cost handheld NDVI sensor and an UAV carrying a multispectral camera. The monitored a low-cost handheld NDVI sensor UAV carrying multispectral camera. The scouting inwith parcel plot trials. In 2014 a plot trialand withanvarying fertilizera types and levels has been monitored with a low-cost handheld NDVI sensor and an UAV carrying a multispectral camera. The results show that data from both spectral measurement devices was fit to estimate triticale dry matter results show thata data fromhandheld both spectral measurement fit to aestimate triticalecamera. dry matter monitored with low-cost NDVI sensor and devices an UAVwas carrying multispectral The results show that data from both spectral measurement devices was fit to estimate triticale dry matter yield with confidence levels exceeding 95%. yield with confidence 95%.measurement devices was fit to estimate triticale dry matter results show that datalevels from exceeding both spectral yield with confidence levels exceeding 95%. yield with confidence levels exceeding 95%. © 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogenLtd. All rights reserved. Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen 1. INTRODUCTION INTRODUCTION 1. 1. INTRODUCTION 1. INTRODUCTION Establishing parcel parcel plots plots is aa common common method method for for comparing comparing Establishing is Establishing parcel plots commontreatments method for comparing different breeds breeds of of plants plants is oradifferent different treatments under ceteris different or under ceteris Establishing parcel plots is a common method for comparing different breeds of plants or different treatments under ceteris paribus conditions. They consist of several rectangular plots paribus conditions. They consist of several rectangular plots different breeds of plants or different treatments under ceteris paribus conditions. They consist of several rectangular plots arranged in rows and columns. Traditionally, the arranged in rowsThey and columns. Traditionally, the paribus conditions. consist of several rectangular plots arranged in rows and columns. Traditionally, the development of plants from seeding to harvest is being development plantsand from columns. seeding toTraditionally, harvest is being arranged in ofrows the development of plants from seeding to harvest is being monitored by human perception based on regular crop monitored byofhuman based on regular crop development plants perception from seeding to harvest is being monitored by human perception based regular set crop scouting resulting in for selected of scouting missions missions resulting in scores scores for aaon of monitored by human perception based onselected regular setcrop scouting missions resulting in scores for a selected set of target features. target features. scouting missions resulting in scores for a selected set of target features. target Plant features. phenotyping is is an an emerging emerging approach approach for for assessing assessing Plant phenotyping Plant phenotyping an emerging approach the for subjective assessing plant traits traits based on onissensor sensor values eliminating eliminating the subjective plant based values Plant phenotyping is an emerging approach for assessing plant traits based on sensor values eliminating the subjective nature of human judgement. It aims at achieving highly nature of human It aims at achieving highly plant traits based onjudgement. sensor values eliminating the subjective nature of human judgement. It aims at achieving highly reproducible measurements of plant properties. At the same reproducible measurements of plant properties. At the same nature of human judgement. It aims at achieving highly reproducible measurements of plant properties. At the same time a sensor-based approach allows for high-throughput time a sensor-based approach allows for high-throughput reproducible measurements of plant properties. At the same time a sensor-based approach allows for high-throughput measurements increasing data density both in the time as measurements increasing data density both the time as well well time a sensor-based approach allows forin high-throughput measurements increasing data density both in thehyperspectral time as well as in the spatial domain. Multiand as in the spatial domain. Multimeasurements increasing data density bothand in thehyperspectral time as well as in the devices spatial are domain. Multiandaa high hyperspectral measurement perceived to potential measurement perceived to have have potential as in the devices spatial are domain. Multiand high hyperspectral measurement are perceivedas have a high of potential for monitoring monitoringdevices plant characteristics characteristics astothe the reflection of visible for plant reflection visible measurement devices are perceived to have a high potential for monitoring plant characteristics as the reflection of visible and near-infrared near-infrared light light is is closely closely correlated correlated to to properties properties of and of for monitoring plant characteristics as the reflection of visible and near-infrared light is closely correlated to properties of plants relevant to breeding and production. plants relevant to breeding and production. and near-infrared light is closely correlated to properties of plants relevant to breeding and production. plants relevant to breeding and production. Gates Gates et et al. al. (1965) (1965) report report that that plants plants absorb, absorb, transmit transmit and and Gates et al. (1965) report that plants absorb, transmit and reflect light dependant on species and the thickness of reflect et light on species and theabsorb, thickness of leaves. leaves. Gates al.dependant (1965) report that plants transmit and reflect light dependant on species and the thickness of leaves. Schmidhalter et al. (2003) have investigated the relationship Schmidhalter et al. (2003) have investigated the relationship reflect light dependant on species and the thickness of leaves. Schmidhalter et al. spectral (2003) have investigated the relationship between indices and nitrogen between different different indices and biomass, biomass, nitrogen Schmidhalter et al. spectral (2003) have investigated the relationship between different spectral indices and biomass, nitrogen uptake and yield of wheat (Triticum aestivum). They found uptake and yield of wheat (Triticum aestivum). They found between different spectral indices and biomass, nitrogen uptake andof of 94% wheatwhen (Triticum aestivum). They found -values ofyield up to to 94% when applying quadratic fits. All R22-values up applying quadratic fits. All R uptake and yield of wheat (Triticum aestivum). They found 2 -values of up to 94% when applying quadratic fits. All R indices under investigation investigation had had aa coefficient coefficient of of determination determination 2 indices under R -values of up to 94% when applying quadratic fits. All indices under investigation had a coefficient of determination above 0.7 when related to grain yield. Bach (1998) was able above when related to grain Bach of (1998) was able indices0.7 under investigation had ayield. coefficient determination above 0.7 when related yield. Bach (1998) was able to predict corn in the Rhine Valley from satellite-born to predict corn yield yield in to thegrain Rhine Valley from satellite-born above 0.7 when related to grain yield. Bach (1998) was able to predict corn yield in the Rhine Valley from satellite-born spectral data with “a high degree of accuracy”. Carter spectral with “a in high accuracy”. Carter et et al. al. to predictdata corn yield thedegree Rhine of Valley from satellite-born spectral data with “a high degree of accuracy”. Carter et al. (2001) found that reflectance of light in the far-red and green(2001) found of light the far-redCarter and greenspectral data that withreflectance “a high degree of in accuracy”. et al. (2001) that is reflectance of lightrelated in the far-red and greenyellow spectra consistently to stress. yellow found spectra consistently to plant plant stress. (2001) found that is reflectance of lightrelated in the far-red and greenyellow spectra is consistently related to plant stress. Lilienthal (2011) has presented a tractor-mounted Lilienthal (2011) has presented yellow spectra is consistently relateda totractor-mounted plant stress. Lilienthal (2011) a crop tractor-mounted hyperspectral sensing has devicepresented for estimating estimating crop parameters hyperspectral sensing device for parameters Lilienthal (2011) has presented a tractor-mounted hyperspectral sensing device for estimating crop parameters in parcel parcel plot plot trials. trials. in hyperspectral sensing device for estimating crop parameters in parcel plot trials. in parcel plot trials.
The The reflectance reflectance of of light light from from plants plants in in parcel parcel plot plot trials trials can can be be The reflectance of light from plants in parcel plot trials canthe be captured with different different approaches one aspect aspect being captured with approaches one being The reflectance of light from plants in parcel plot trials canthe be captured with different approaches one aspect being the carrier platform. platform. Lately, Lately, unmanned unmanned aerial vehicles vehicles (UAV) (UAV) carrier captured with different approaches aerial one aspect being the carrier platform. Lately,cameras unmanned (UAV) carrying multispectral have been proposed and carrying multispectral haveaerial beenvehicles proposed and carrier platform. Lately,cameras unmanned aerial vehicles (UAV) carrying multispectral cameras have been proposed and utilized for capturing data in the field. Tractor-based sensor utilized for capturing data in the field. carrying multispectral cameras have Tractor-based been proposedsensor and utilized for capturing data in the field. Tractor-based sensor systems like the Trimble Greenseeker system or the YARA systems likecapturing the Trimble or the YARA utilized for dataGreenseeker in the field. system Tractor-based sensor systems system or are the already YARA N-Sensor for site specific nitrogen N-Sensorlike forthe siteTrimble specificGreenseeker nitrogen fertilization fertilization systems like the Trimble Greenseeker system or are the already YARA N-Sensor for site specific nitrogen fertilization are already applied in arable farming and may be suitable for assessing applied in for arable may be suitable for N-Sensor site farming specificand nitrogen fertilization areassessing already applied in arable farming andplot maytrials. be suitable assessing plant characteristics in Sensors may also plant characteristics in parcel parcel Sensorsfor may also be be applied in arable farming andplot maytrials. be suitable for assessing plant parcel plot Sensors may also be carriedcharacteristics by humans humans in triggering thetrials. measurements manually carried by triggering the measurements manually plant characteristics in parcel plot trials. Sensors may also be carried byspectrometers). humans triggering the measurements manually (handheld (handheld carried byspectrometers). humans triggering the measurements manually (handheld spectrometers). (handheld spectrometers). Another aspect aspect of of measuring measuring the the reflectance reflectance of of light light are are Another Another aspectand of resolution. measuring Basic the reflectance of light are spectral range sensor systems directly spectral range and resolution. Basic sensor systems directly Another aspect of measuring the reflectance of light are spectral range and resolution. Basic sensor systems directly compute spectral indices (e.g. NDVI) without revealing the compute spectral (e.g. Basic NDVI)sensor without revealing the spectral range andindices resolution. systems directly compute spectral indices (e.g. NDVI) without revealing the underlying reflectances. Multispectral sensors or cameras underlyingspectral reflectances. sensorsrevealing or cameras compute indices Multispectral (e.g. NDVI) without the underlying reflectances. Multispectral sensors or cameras acquire the reflection of several distinct, nonadjacent spectra acquire the reflection of several distinct, sensors nonadjacent spectra underlying reflectances. Multispectral or cameras acquire reflection of sensor several systems distinct, nonadjacent spectra whereas hyperspectral seamlessly whereasthe hyperspectral seamlessly acquire acquire acquire the reflection of sensor several systems distinct, nonadjacent spectra whereas systems seamlessly acquire reflectionshyperspectral over aa given givensensor spectrum. Passive sensor systems systems reflections over spectrum. Passive sensor whereas hyperspectral sensor systems seamlessly acquire reflections overthe only measure measure thea given energyspectrum. of light light Passive reflectedsensor from systems body only energy of reflected from aa body reflections over a given spectrum. Passive sensor systems only measure the energy of light reflected from a body whereas active sensors emit light with a defined energy level whereas active sensors emit light with a defined energy level only measure the energy of light reflected from a body whereas active sensors emit light with a defined energy enabling them to directly calculate the reflectance being the enabling them to directly calculate theareflectance beinglevel the whereas active sensors emit light with defined energy level enabling them to directly calculate the reflectance being the ratio of reflected light and the total amount of radiation. ratio of reflected and the total amount of radiation. enabling them tolight directly calculate the reflectance being the ratio of reflected light and the total amount of radiation. ratio of reflected light and the total is amount compare of radiation. The The objective objective of of this this paper paper is to to compare spectral spectral The objective from of aathis paper handheld is to compare spectral measurements low-cost device data measurements low-cost device with with data The objective from of this paper handheld is to compare spectral measurements from a low-cost handheld device with from a high-end multispectral camera carried by an from a high-end camera device carriedwith by data an measurements from multispectral a low-cost handheld data from a high-end multispectral camera an unmanned aerial vehicle vehicle (UAS) with with respectcarried to their their by fitness unmanned aerial (UAS) respect to fitness from a high-end multispectral camera carried by an unmanned aerial vehicle (UAS) with respect to their fitness for anticipating anticipating the biomass biomass of of small small grains. grains. for unmanned aerialthevehicle (UAS) with respect to their fitness for anticipating the biomass of small grains. for anticipating the biomass of small grains. 2. MATERIALS MATERIALS AND AND METHODS METHODS 2. 2. MATERIALS AND METHODS 2. MATERIALS AND METHODS After After the the introduction introduction of of the the Renewable Renewable Energy Energy Act Act (EEG) (EEG) in in After the introduction of the Renewable Energy Act (EEG) the year 2000 the number and the size of biogas plants the year the number the sizeEnergy of biogas plants in After the 2000 introduction of the and Renewable Act (EEG) in the year 2000 the number and the size of biogas plants in Germany has been Biogas plants Germany has steadily steadily been growing. growing. Biogas plants primarily primarily the year 2000 the number and the size of biogas plants in Germany steadily of been growing. BiogasAt plants primarily serve production electrical energy. the time serve the the has production electrical energy. the same same time Germany has steadily of been growing. BiogasAtplants primarily serve the production of electrical energy. At the same time they produce substantial amounts of biogas slurry serving as they produce substantial amounts of biogasAt slurry serving as serve the production of electrical energy. the same time they produce substantial of biogas slurry serving as organic fertilizer in plant plant amounts production. organic fertilizer in production. they produce substantial amounts of biogas slurry serving as organic fertilizer in plant production. organic fertilizer in plant production.
Copyright © 2016 409 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2016, 2016 IFAC IFAC 409Hosting by Elsevier Ltd. All rights reserved. Copyright 2016 responsibility IFAC 409Control. Peer review©under of International Federation of Automatic Copyright © 2016 IFAC 409 10.1016/j.ifacol.2016.10.074
IFAC AGRICONTROL 2016 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
2.1 Trial Description
triggered. The resulting NDVI is calculated from the reflectances (reflection/radiation) according to the formula below:
The fertilizing effect of animal slurry has been well investigated whereas the effect of biogas slurry is not yet fully understood. The Agricultural Academy in Triesdorf (Landwirtschaftliche Lehranstalten, LLA) has established a long-term trial (2011-2014) with four different crop rotations and fertilizing regimes.
NDVI =
Corn-Rye
TriticaleSorghum
Corn
CornWheatCanola
Fertilization
100% biogas slurry
50% biogas slurry/ mineral fertilizer
100% mineral fertilizer
No fertilization
R770 nm − R650 nm R770 nm + R650 nm
In addition, a Tetracam Mini MCA6 with ILS (Incident Light Sensor) carried by a geo-XR6 hexacopter UAV was operated over the trial on May 5, 2014, May 23, 2014 and June 30,2014. The Tetracam multispectral camera was configured to measure the reflection of light at 530 nm, 670 nm, 700 nm, 730 nm and 780 nm. Due to the light sensor compensating the spectral reflections for solar radiation the output is again represented as reflectances. The pictures were processed by geo-konzept GmbH, Adelschlag, Germany, resulting in multispectral digital orthophotos (DOP) and digital terrain models (DTM) extracted from pictures taken with a RGB camera (Sony a5000 alpha).
Table 1. Crop rotations and fertilizing regimes Rotation
405
The crop rotations under investigation are typical for farms operating biogas plants in the region.
Plant heights were derived from the digital terrain model by following an approach suggested by Bendig et al. (2015). A terrain model of the soil surface was created by neglecting the areas covered by plants. After subtracting the original DTM from the soil surface model the resulting dataset contains the plant heights (Crop Surface Model, CSM).
Triesdorf is located in Bavaria, Germany (49.210 N, 10.590 E) in the county of Ansbach which has the highest density of biogas plants in Germany. The average annual temperature is 7,7 0C. Average annual precipitation between 1960 and 1990 was 630 mm, in 2014 the annual precipitation was 661 mm.
After the parcel boundaries had been surveyed with a RTKGNSS receiver (Trimble Integrated FmX Display applying Ntrip corrections), median plant heights and median reflectances for all wavelengths captured with the multispectral camera have been assigned to the parcels using Quantum GIS.
This paper focusses on Triticale which was sown on October 1, 2013. Four parcel blocks with randomly arranged variations with respect to fertilization were established. Each parcel plot had a length of 8.20 m and was three parcels or 4.50 m wide. Yield was determined by harvesting only the center parcel (1.50 m wide).
The parcels were harvested on July 1, 2014. The dry matter yield was determined by drying and related to the area harvested resulting in dry matter yield in dt/ha (decitonnes per hectare).
The nitrogen fertilization was carried out on February 23, 2014 and March 24/25, 2014. Parcels treated solely with biogas slurry received 98 kg N /ha and 42 kg N /ha respectively. Parcels receiving both organic and mineral fertilizer received 70 kg N/ha in the form of biogas slurry as a starter dressing and a subsequent fertilizer dose of 70 kg N/ha Hydrosulfan. Parcels solely treated with mineral fertilizer received two fertilizer dressings with a dosage of 68 kg N/ha Hydrosulfan.
3. RESULTS 3.1 Dry Matter Yield As expected dry matter yield in parcels with no fertilization clearly underperformed (Ø 72 dt/ha) when compared to the parcels which received N-fertilization (Ø 152 dt/ha). The relative ranking within the repetitions revealed that parcels treated with mineral fertilizer performed best, followed by the parcels treated with biogas slurry only (Table 2).
All parcels have been treated with the herbicide Broadway on March 31, 2014. Triticale was harvested on July 1, 2014. 2.2 Measurement Devices and Data Sampling
Table 2. Relative Treatment Ranking The NDVI (Normalized Difference Vegetation Index) in the parcel plots under investigation has been measured with a Trimble GreenSeeker handheld device in the period between April 07, 2014 and June 30, 2014 on a weekly basis with four repetitions per parcel plot (576 samples). According to Wilkinson (2011) the sensor measures the reflection from the plant canopy at 650 nm (red) and 770 nm (near-infrared). The device features LEDs (Light emitting diodes) radiating light at the same wavelengths whenever a measurement is
Treatment
Fertilization
Rank 1 100% biogas slurry 2.0 2 50% biogas slurry/mineral fertilizer 2.8 3 100% mineral fertilizer 1.3 4 No fertilization 4.0 A comparison of the repetitions revealed that parcels which received fertilization in repetition 1 significantly produced 410
IFAC AGRICONTROL 2016 406 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
less biomass (130 – 151 dt/ha, 95% CI) than the parcels in repetition 4 (157 – 173 dt/ha, 95% CI). Repetitions 2 and 3 did not significantly differ from the other repetitions.
Table 3. Correlation between yield and NDVI measurements with GreenSeeker handheld 04/07
04/14
04/23
04/28
05/04
05/19
05/26
06/18
06/30
As expected, there is a close correlation between the NDVI calculated from the data collected with the Tetracam sensor and the NDVI measured with the GreenSeeker handheld (R2>0.9, Figure 1). This is true for the measurements taken with the GreenSeeker on May 5 and the pictures acquired with the Tetracam sensor on May 4 as well as for the data acquired on May 23 and May 26. In opposition, the NDVI from both sensor systems measured on June 30 does not correlate at all (R2>0.01).
Treatment
3.2 NDVI GreenSeeker vs. Tetracam Mini MCA6
1 2 3 4 All
1.00 0.69 0.97 -0.25 0.98
0.96 0.31 0.98 0.17 0.97
0.67 0.77 0.89 0.53 0.98
0.56 0.75 0.74 0.42 0.97
0.84 0.65 1.00 0.26 0.96
0.75 0.48 0.78 0.67 0.97
0.91 0.68 0.65 0.44 0.98
0.82 0.55 0.84 0.68 0.87
-0.20 0.28 -0.08 -0.31 0.01
Figure 2 shows that even early measurements of the NDVI (April 7) closely correlate with dry matter yield.
Fig. 1. Correlation between NDVI from Tetracam and NDVI from GreenSeeker handheld Fig. 2. Correlation between Triticale Dry Matter Yield and NDVI from measurements with a GreenSeeker handheld on April 7, 2014
Figure 1 shows that the NDVI acquired with the two sensor systems in the beginning of May significantly differ in spectral behavior. The slope of the regression term is approximately 2 where a value of 1 was expected. Also, the intercept has a very high value in the range of 1. The relation of data collected on May 23/May 26 is much closer to the expected range (slope 0.99, intercept 0.18).
However, it also shows that the high correlation coefficients are mainly due to the skew distribution of the data. The correlation between yield in fertilized parcels only and NDVI measured with the GreenSeeker handheld is much lower (R2= 0.70).
3.3 NDVI GreenSeeker vs. Dry Matter Yield The NDVI measurements with the GreenSeeker handheld device were sampled with four repetitions per parcel plot. The values obtained from single parcels were very consistent showing a low variation (mean standard deviation 0.02). The measurements taken in April and May were closely correlated to dry matter yield (R > 0.95, Table 2). In mid June (June 18) the correlation between dry matter yield and NDVI started decreasing with progressing senescence. The NDVI measurements taken on June 30 do not correlate with the dry matter yield harvested on the following day at all (R=0.01). Fig. 3. Correlation between Triticale Dry Matter Yield and NDVI from measurements with a GreenSeeker handheld on April 7, 2014, fertilized parcels only
It is noticeable that in the beginning of April the relation on treatment level is close for plots either dressed with mineral fertilizer or biogas slurry whereas the plots with mixed fertilizer treatment show a significantly lower correlation
Figure 3 shows that a very high correlation between dry matter yield and NDVI measurements with the GreenSeeker handheld can be established when grouping the data. Group1 represents data from parcels fertilized with biogas slurry only 411
IFAC AGRICONTROL 2016 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
and repetitions 1 and 2 from parcels fertilized with both fertilizers. Group2 contains data from parcels dressed with mineral fertilizer only and repetitions 3 and 4 from parcels fertilized with both fertilizers. In both groups yield is very closely correlated (R2= 0.98) to the NDVI measured on April 7.
407
yield is very close, despite one outlier in a parcel fertilized with both fertilizers.
3.4 NDVI Tetracam vs. Dry Matter Yield The median reflectances at 670 nm und 780 nm obtained from the Tetracam camera on May 5 and May 23 were used to calculate the NDVI and related to the dry matter yield (Figure 4). The close correlation (R2> 0.9) is again mainly due to the skew distribution of data. The NDVI measured in fertilized parcels with higher yields shows a very low response to yield.
Fig. 5. Triticale Dry Matter Yield versus reflectance at 780 nm measured with Tetracam on May 23, 2014 3.5 Crop Height Model vs. Dry Matter Yield The mean crop heights resulting from the digital terrain model derived from RGB pictures taken on May 5 show an extremely high correlation to dry matter yield (R2= 0.98).
Fig. 4. Triticale Dry Matter Yield versus NDVI from Tetracam on May 5 and May 23, 2014 The Tetracam camera collects reflectances at different wavelength, which can individually be related to yield without prior calculation of indices. Table 4 summarizes the correlation coefficients grouped by sampling dates and wavelengths. Both wavelengths related to NDVI show a close correlation to dry matter yield. The red channel (670 nm) is negatively correlated to yield on both dates (R< -0.9) while the infrared channel show a consistent positive correlation with yield (R = 0.98).
Fig. 6. Mean Crop Plant Height versus Triticale Dry Matter Yield May 5, 2014 The close relation is not only based on the uneven distribution of data. The value for Spearmen’s rank correlation between crop height and dry matter yield is also well above 0.9 (RSpearman = 0.93).
Table 4. Correlation between yield and spectral reflectance at different wavelength captured with Tetracam Mini MCA6
Date
Wavelength
7. DISCUSSION
530
670
700
730
780
May 5
-0.96
-0.95
-0.96
0.10
0.98
May 23
-0.88
-0.95
-0.91
0.87
0.98
In the field trial under investigation dry matter yield of Triticale varied substantially depending on fertilization and location. Parcels dressed with mineral fertilizer performed best when considering all repetitions. On the other hand, the yield in all fertilized parcels in repetition 1 was lower than the yield of corresponding plots in repetition 4. This indicates that the effect of soil variation may overrule the effect of fertilization even on a small scale (< 100 m).
Also the green channel (530 nm) and the second red channel (730 nm) are closely related to dry matter yield. On May5, the relation is even a bit closer than that of the red channel regarded for the NDVI. In Figure 4 dry matter yield is plotted against the reflectance at 780 nm. The coefficient of determination (R2) is 0.97. The relation between the reflectances measured and dry matter
The relation between the NDVI obtained from the GreenSeeker handheld and the Tetracam camera was looser than expected. Despite high coefficients of determination, 412
IFAC AGRICONTROL 2016 408 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
the slope of regression and the intercept indicated substantial differences between the NDVI measurements. The time lag of 1 resp. 3 days between the measurements may have had an effect, but cannot appropriately explain the large deviations. The difference may partly be accounted to the different spectra covered for red (650 nm vs. 670 nm) and infrared (770 nm vs. 780 nm). It is also likely that the spatial resolution and the number of samples taken by the handheld sensor and the camera have affected the results.
better understand the spectral and the time domain of plant feature responses. 7. ACKNOWLEDGEMENTS The author would like to thank Markus Heinz from LLA Triesdorf for providing access to his trial fields and data collected by his staff. geo-konzept GmbH has supported this study with performing two UAS missions and processing the resulting data. Florian Baus, Dominik Weizmann, and Christian Schnitzlein deserve appreciation for very reliably collecting data with a handheld sensor on a weekly basis.
The relation between dry matter yield and the NDVI measurements taken with the GreenSeeker handheld were extremely close. This was even true for data collected at an early stage (April 7). An even closer relation could be established by arranging the data in groups. The results indicate that plots fertilized with mineral fertilizer generally produced higher NDVI values at the same level of dry matter yield when compared to plots fertilized with biogas slurry. Repetition 1 and 2 of those plots with mixed fertilisation align with the plots with biogas slurry treatment whereas repetition 3 and 4 were in line with the plots with mineral fertilization. This would again indicate that differences in soil property affect the behaviour of nitrogen uptake from mineral and organic sources.
REFERENCES Bach, H. (1998). Yield estimation of corn based on multitemporal LANDSAT-TM data as input for an agrometeorological model. Pure Appl. Opt. 7 (1998), pp. 809–825 Bendig,J., Yu, K., Aasen, H., Bolten, A., Bennertz, S., , Broscheit, J., Gnyp, M.L., Bareth, G. (2015). Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley. International Journal of Applied Earth Observation and Geoinformation 39 (2015), pp 79–87 Carter, G.A., and Knapp, A.K. (2001). Leaf optical properties in higher plants: linking spectral characteristics to stress and chlorophyll concentration. American Journal of Botany, April 2001, Vol. 88, No. 4, pp. 677-684 Gates, D.M., Keegan, H.J., Schleter, J.C. and Weidner, Victor R. (1965). Spectral Properties of Plants. Applied Optics, Vol. 4, No. 1, January 1965 Carter, G.A., and Knapp, A.K. (2001). Leaf optical properties in higher plants: linking spectral characteristics to stress and chlorophyll concentration. American Journal of Botany, April 2001, Vol. 88, No. 4, pp. 677-684 Lilienthal, H. (2011). Estimating Crop Parameters by processing hyperspectral canopy reflectance data. International Symposium on Sensing in Agriculture. February 21 - 24, 2011. Haifa Santin-Janin, H., Garel, M., Chapuis, J.-L, Pontier, D. (2009) Assessing the performance of NDVI as a proxy for plant biomass using non-linear models: a case study on the Kerguelen archipelago, Polar Biol (2009) 32:pp 861–871 Schmidhalter, U., Jungert, S., Bredemeier, C. , Gutser, R., Manhart, R., Mistele, B., and Gerl, G. (2003). Field-scale validation of a tractor based multispectral crop scanner to determine biomass and nitrogen uptake of winter wheat. Stafford, J.,Werner, A.(eds): Precision Agriculture. Wageningen Academic Publishers, pp 615-619 Wilkerson, J. (2011): Real-Time Sensor-Based Nutrient Management,http://www.utcrops.com/Presentations/Sens ors%20and%20Nutrients%20-%20Wilkerson.pdf, last visited February 29, 2016
The relation between the NDVI calculated from the reflectances captured with the Tetracam camera and dry matter yield was looser than expected. It is mainly the uneven distribution of data that is accountable for the high coefficients of determination. Especially in the fertilized plots the NDVI response on increasing yield was very low and inconsistent. This may be due to the saturation of NDVI when it comes to high biomass values as suggested by Santin-Janin et al. (2009). On the contrary, the NDVI measurements obtained with the GreenSeeker handheld were closely correlated despite high biomasses. Reflectances measured with the Tetracam camera at single wavelengths were partly much closer related to dry matter yield than the NDVI. The reflectances measured at 780 nm on May 25 were very much in line with dry matter yield, except from one outlier. This result supports the approach of Lilienthal (2011) who suggests applying complex modelling techniques (e.g. SVM) to spectral reflectance measurements at different wavelength rather than calculating indices. The relation between mean crop height of Triticale derived from a digital terrain model on the one hand and dry matter yield on the other hand was much closer (R2=0.98) than the relation between yield and the spectral data from the Tetracam camera. In case of Triticale, plant height seems to be the best proxy for dry matter yield when compared to spectral measurements. This study is limited to the measurement of biomass of Triticale. The results do not allow for the prediction of dry matter yield of other plants or even grain yield. However, the results show that there is high potential in spectral measurements with low-cost sensors (GreenSeeker) and high end technology systems like UAV for predicting plant properties such as biomass. Further research is necessary to
413
ScienceDirect
IFAC-PapersOnLine 49-16 (2016) 404–408
Estimating Triticale Dry Matter Yield in Parcel Plot Trials Estimating Triticale Dry Yield in Plot Estimating Triticale Dry Matter Matter in Parcel Parcel Plot Trials Trials From Aerial and Ground BasedYield Spectral Measurements From Aerial and Ground Based Spectral Measurements From Aerial and Ground Based Spectral Measurements P.O. Noack* Noack* P.O. P.O. Noack* P.O. Noack* *University *University of of Applied Applied Sciences Sciences Weihenstephan-Triesdorf, Weihenstephan-Triesdorf, *University of Applied Markgrafenstr. 16 ,, Sciences D-91746 Weihenstephan-Triesdorf, Weidenbach, Germany Germany Markgrafenstr. 16 D-91746 Weidenbach, *University of Applied Sciences Weihenstephan-Triesdorf, Markgrafenstr. 16 , D-91746 Germany (Tel: +49-9826-654242; e-mail:Weidenbach, [email protected]) (Tel: +49-9826-654242; e-mail: [email protected]) Markgrafenstr. 16 , D-91746 Weidenbach, Germany (Tel: +49-9826-654242; e-mail: [email protected]) (Tel: +49-9826-654242; e-mail: [email protected])
Abstract: Spectral Spectral measurement measurement devices devices are are aa useful useful tool tool for for quantifying quantifying biomass biomass and and the the nitrogen nitrogen Abstract: Abstract: Spectral measurement devices arespecific a useful tool for quantifying biomass and the uptake of plants. plants. They are employed employed for site site specific nitrogen application in arable arable farming and nitrogen for crop crop uptake of They are for nitrogen application in farming and for Abstract: Spectral measurement devices are a useful tool for quantifying biomass and the nitrogen uptake ofin plants. They are employed for site specific nitrogen application in arable farming andhas for been crop scouting parcel plot trials. In 2014 aa plot trial with varying fertilizer types and levels scouting in parcel plot trials. In 2014 plot trial with varying fertilizer types and levels has been uptake of plants. They are employed for site specific nitrogen application in arable farming and for crop scouting in parcel plot trials. In 2014 a plot trial with varying fertilizer types and levels has been monitored with a low-cost handheld NDVI sensor and an UAV carrying a multispectral camera. The monitored a low-cost handheld NDVI sensor UAV carrying multispectral camera. The scouting inwith parcel plot trials. In 2014 a plot trialand withanvarying fertilizera types and levels has been monitored with a low-cost handheld NDVI sensor and an UAV carrying a multispectral camera. The results show that data from both spectral measurement devices was fit to estimate triticale dry matter results show thata data fromhandheld both spectral measurement fit to aestimate triticalecamera. dry matter monitored with low-cost NDVI sensor and devices an UAVwas carrying multispectral The results show that data from both spectral measurement devices was fit to estimate triticale dry matter yield with confidence levels exceeding 95%. yield with confidence 95%.measurement devices was fit to estimate triticale dry matter results show that datalevels from exceeding both spectral yield with confidence levels exceeding 95%. yield with confidence levels exceeding 95%. © 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogenLtd. All rights reserved. Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen Keywords: spectroscopy, plant nutrition, yield estimation, UAS, biogas, nitrogen 1. INTRODUCTION INTRODUCTION 1. 1. INTRODUCTION 1. INTRODUCTION Establishing parcel parcel plots plots is aa common common method method for for comparing comparing Establishing is Establishing parcel plots commontreatments method for comparing different breeds breeds of of plants plants is oradifferent different treatments under ceteris different or under ceteris Establishing parcel plots is a common method for comparing different breeds of plants or different treatments under ceteris paribus conditions. They consist of several rectangular plots paribus conditions. They consist of several rectangular plots different breeds of plants or different treatments under ceteris paribus conditions. They consist of several rectangular plots arranged in rows and columns. Traditionally, the arranged in rowsThey and columns. Traditionally, the paribus conditions. consist of several rectangular plots arranged in rows and columns. Traditionally, the development of plants from seeding to harvest is being development plantsand from columns. seeding toTraditionally, harvest is being arranged in ofrows the development of plants from seeding to harvest is being monitored by human perception based on regular crop monitored byofhuman based on regular crop development plants perception from seeding to harvest is being monitored by human perception based regular set crop scouting resulting in for selected of scouting missions missions resulting in scores scores for aaon of monitored by human perception based onselected regular setcrop scouting missions resulting in scores for a selected set of target features. target features. scouting missions resulting in scores for a selected set of target features. target Plant features. phenotyping is is an an emerging emerging approach approach for for assessing assessing Plant phenotyping Plant phenotyping an emerging approach the for subjective assessing plant traits traits based on onissensor sensor values eliminating eliminating the subjective plant based values Plant phenotyping is an emerging approach for assessing plant traits based on sensor values eliminating the subjective nature of human judgement. It aims at achieving highly nature of human It aims at achieving highly plant traits based onjudgement. sensor values eliminating the subjective nature of human judgement. It aims at achieving highly reproducible measurements of plant properties. At the same reproducible measurements of plant properties. At the same nature of human judgement. It aims at achieving highly reproducible measurements of plant properties. At the same time a sensor-based approach allows for high-throughput time a sensor-based approach allows for high-throughput reproducible measurements of plant properties. At the same time a sensor-based approach allows for high-throughput measurements increasing data density both in the time as measurements increasing data density both the time as well well time a sensor-based approach allows forin high-throughput measurements increasing data density both in thehyperspectral time as well as in the spatial domain. Multiand as in the spatial domain. Multimeasurements increasing data density bothand in thehyperspectral time as well as in the devices spatial are domain. Multiandaa high hyperspectral measurement perceived to potential measurement perceived to have have potential as in the devices spatial are domain. Multiand high hyperspectral measurement are perceivedas have a high of potential for monitoring monitoringdevices plant characteristics characteristics astothe the reflection of visible for plant reflection visible measurement devices are perceived to have a high potential for monitoring plant characteristics as the reflection of visible and near-infrared near-infrared light light is is closely closely correlated correlated to to properties properties of and of for monitoring plant characteristics as the reflection of visible and near-infrared light is closely correlated to properties of plants relevant to breeding and production. plants relevant to breeding and production. and near-infrared light is closely correlated to properties of plants relevant to breeding and production. plants relevant to breeding and production. Gates Gates et et al. al. (1965) (1965) report report that that plants plants absorb, absorb, transmit transmit and and Gates et al. (1965) report that plants absorb, transmit and reflect light dependant on species and the thickness of reflect et light on species and theabsorb, thickness of leaves. leaves. Gates al.dependant (1965) report that plants transmit and reflect light dependant on species and the thickness of leaves. Schmidhalter et al. (2003) have investigated the relationship Schmidhalter et al. (2003) have investigated the relationship reflect light dependant on species and the thickness of leaves. Schmidhalter et al. spectral (2003) have investigated the relationship between indices and nitrogen between different different indices and biomass, biomass, nitrogen Schmidhalter et al. spectral (2003) have investigated the relationship between different spectral indices and biomass, nitrogen uptake and yield of wheat (Triticum aestivum). They found uptake and yield of wheat (Triticum aestivum). They found between different spectral indices and biomass, nitrogen uptake andof of 94% wheatwhen (Triticum aestivum). They found -values ofyield up to to 94% when applying quadratic fits. All R22-values up applying quadratic fits. All R uptake and yield of wheat (Triticum aestivum). They found 2 -values of up to 94% when applying quadratic fits. All R indices under investigation investigation had had aa coefficient coefficient of of determination determination 2 indices under R -values of up to 94% when applying quadratic fits. All indices under investigation had a coefficient of determination above 0.7 when related to grain yield. Bach (1998) was able above when related to grain Bach of (1998) was able indices0.7 under investigation had ayield. coefficient determination above 0.7 when related yield. Bach (1998) was able to predict corn in the Rhine Valley from satellite-born to predict corn yield yield in to thegrain Rhine Valley from satellite-born above 0.7 when related to grain yield. Bach (1998) was able to predict corn yield in the Rhine Valley from satellite-born spectral data with “a high degree of accuracy”. Carter spectral with “a in high accuracy”. Carter et et al. al. to predictdata corn yield thedegree Rhine of Valley from satellite-born spectral data with “a high degree of accuracy”. Carter et al. (2001) found that reflectance of light in the far-red and green(2001) found of light the far-redCarter and greenspectral data that withreflectance “a high degree of in accuracy”. et al. (2001) that is reflectance of lightrelated in the far-red and greenyellow spectra consistently to stress. yellow found spectra consistently to plant plant stress. (2001) found that is reflectance of lightrelated in the far-red and greenyellow spectra is consistently related to plant stress. Lilienthal (2011) has presented a tractor-mounted Lilienthal (2011) has presented yellow spectra is consistently relateda totractor-mounted plant stress. Lilienthal (2011) a crop tractor-mounted hyperspectral sensing has devicepresented for estimating estimating crop parameters hyperspectral sensing device for parameters Lilienthal (2011) has presented a tractor-mounted hyperspectral sensing device for estimating crop parameters in parcel parcel plot plot trials. trials. in hyperspectral sensing device for estimating crop parameters in parcel plot trials. in parcel plot trials.
The The reflectance reflectance of of light light from from plants plants in in parcel parcel plot plot trials trials can can be be The reflectance of light from plants in parcel plot trials canthe be captured with different different approaches one aspect aspect being captured with approaches one being The reflectance of light from plants in parcel plot trials canthe be captured with different approaches one aspect being the carrier platform. platform. Lately, Lately, unmanned unmanned aerial vehicles vehicles (UAV) (UAV) carrier captured with different approaches aerial one aspect being the carrier platform. Lately,cameras unmanned (UAV) carrying multispectral have been proposed and carrying multispectral haveaerial beenvehicles proposed and carrier platform. Lately,cameras unmanned aerial vehicles (UAV) carrying multispectral cameras have been proposed and utilized for capturing data in the field. Tractor-based sensor utilized for capturing data in the field. carrying multispectral cameras have Tractor-based been proposedsensor and utilized for capturing data in the field. Tractor-based sensor systems like the Trimble Greenseeker system or the YARA systems likecapturing the Trimble or the YARA utilized for dataGreenseeker in the field. system Tractor-based sensor systems system or are the already YARA N-Sensor for site specific nitrogen N-Sensorlike forthe siteTrimble specificGreenseeker nitrogen fertilization fertilization systems like the Trimble Greenseeker system or are the already YARA N-Sensor for site specific nitrogen fertilization are already applied in arable farming and may be suitable for assessing applied in for arable may be suitable for N-Sensor site farming specificand nitrogen fertilization areassessing already applied in arable farming andplot maytrials. be suitable assessing plant characteristics in Sensors may also plant characteristics in parcel parcel Sensorsfor may also be be applied in arable farming andplot maytrials. be suitable for assessing plant parcel plot Sensors may also be carriedcharacteristics by humans humans in triggering thetrials. measurements manually carried by triggering the measurements manually plant characteristics in parcel plot trials. Sensors may also be carried byspectrometers). humans triggering the measurements manually (handheld (handheld carried byspectrometers). humans triggering the measurements manually (handheld spectrometers). (handheld spectrometers). Another aspect aspect of of measuring measuring the the reflectance reflectance of of light light are are Another Another aspectand of resolution. measuring Basic the reflectance of light are spectral range sensor systems directly spectral range and resolution. Basic sensor systems directly Another aspect of measuring the reflectance of light are spectral range and resolution. Basic sensor systems directly compute spectral indices (e.g. NDVI) without revealing the compute spectral (e.g. Basic NDVI)sensor without revealing the spectral range andindices resolution. systems directly compute spectral indices (e.g. NDVI) without revealing the underlying reflectances. Multispectral sensors or cameras underlyingspectral reflectances. sensorsrevealing or cameras compute indices Multispectral (e.g. NDVI) without the underlying reflectances. Multispectral sensors or cameras acquire the reflection of several distinct, nonadjacent spectra acquire the reflection of several distinct, sensors nonadjacent spectra underlying reflectances. Multispectral or cameras acquire reflection of sensor several systems distinct, nonadjacent spectra whereas hyperspectral seamlessly whereasthe hyperspectral seamlessly acquire acquire acquire the reflection of sensor several systems distinct, nonadjacent spectra whereas systems seamlessly acquire reflectionshyperspectral over aa given givensensor spectrum. Passive sensor systems systems reflections over spectrum. Passive sensor whereas hyperspectral sensor systems seamlessly acquire reflections overthe only measure measure thea given energyspectrum. of light light Passive reflectedsensor from systems body only energy of reflected from aa body reflections over a given spectrum. Passive sensor systems only measure the energy of light reflected from a body whereas active sensors emit light with a defined energy level whereas active sensors emit light with a defined energy level only measure the energy of light reflected from a body whereas active sensors emit light with a defined energy enabling them to directly calculate the reflectance being the enabling them to directly calculate theareflectance beinglevel the whereas active sensors emit light with defined energy level enabling them to directly calculate the reflectance being the ratio of reflected light and the total amount of radiation. ratio of reflected and the total amount of radiation. enabling them tolight directly calculate the reflectance being the ratio of reflected light and the total amount of radiation. ratio of reflected light and the total is amount compare of radiation. The The objective objective of of this this paper paper is to to compare spectral spectral The objective from of aathis paper handheld is to compare spectral measurements low-cost device data measurements low-cost device with with data The objective from of this paper handheld is to compare spectral measurements from a low-cost handheld device with from a high-end multispectral camera carried by an from a high-end camera device carriedwith by data an measurements from multispectral a low-cost handheld data from a high-end multispectral camera an unmanned aerial vehicle vehicle (UAS) with with respectcarried to their their by fitness unmanned aerial (UAS) respect to fitness from a high-end multispectral camera carried by an unmanned aerial vehicle (UAS) with respect to their fitness for anticipating anticipating the biomass biomass of of small small grains. grains. for unmanned aerialthevehicle (UAS) with respect to their fitness for anticipating the biomass of small grains. for anticipating the biomass of small grains. 2. MATERIALS MATERIALS AND AND METHODS METHODS 2. 2. MATERIALS AND METHODS 2. MATERIALS AND METHODS After After the the introduction introduction of of the the Renewable Renewable Energy Energy Act Act (EEG) (EEG) in in After the introduction of the Renewable Energy Act (EEG) the year 2000 the number and the size of biogas plants the year the number the sizeEnergy of biogas plants in After the 2000 introduction of the and Renewable Act (EEG) in the year 2000 the number and the size of biogas plants in Germany has been Biogas plants Germany has steadily steadily been growing. growing. Biogas plants primarily primarily the year 2000 the number and the size of biogas plants in Germany steadily of been growing. BiogasAt plants primarily serve production electrical energy. the time serve the the has production electrical energy. the same same time Germany has steadily of been growing. BiogasAtplants primarily serve the production of electrical energy. At the same time they produce substantial amounts of biogas slurry serving as they produce substantial amounts of biogasAt slurry serving as serve the production of electrical energy. the same time they produce substantial of biogas slurry serving as organic fertilizer in plant plant amounts production. organic fertilizer in production. they produce substantial amounts of biogas slurry serving as organic fertilizer in plant production. organic fertilizer in plant production.
Copyright © 2016 409 2405-8963 © IFAC (International Federation of Automatic Control) Copyright © 2016, 2016 IFAC IFAC 409Hosting by Elsevier Ltd. All rights reserved. Copyright 2016 responsibility IFAC 409Control. Peer review©under of International Federation of Automatic Copyright © 2016 IFAC 409 10.1016/j.ifacol.2016.10.074
IFAC AGRICONTROL 2016 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
2.1 Trial Description
triggered. The resulting NDVI is calculated from the reflectances (reflection/radiation) according to the formula below:
The fertilizing effect of animal slurry has been well investigated whereas the effect of biogas slurry is not yet fully understood. The Agricultural Academy in Triesdorf (Landwirtschaftliche Lehranstalten, LLA) has established a long-term trial (2011-2014) with four different crop rotations and fertilizing regimes.
NDVI =
Corn-Rye
TriticaleSorghum
Corn
CornWheatCanola
Fertilization
100% biogas slurry
50% biogas slurry/ mineral fertilizer
100% mineral fertilizer
No fertilization
R770 nm − R650 nm R770 nm + R650 nm
In addition, a Tetracam Mini MCA6 with ILS (Incident Light Sensor) carried by a geo-XR6 hexacopter UAV was operated over the trial on May 5, 2014, May 23, 2014 and June 30,2014. The Tetracam multispectral camera was configured to measure the reflection of light at 530 nm, 670 nm, 700 nm, 730 nm and 780 nm. Due to the light sensor compensating the spectral reflections for solar radiation the output is again represented as reflectances. The pictures were processed by geo-konzept GmbH, Adelschlag, Germany, resulting in multispectral digital orthophotos (DOP) and digital terrain models (DTM) extracted from pictures taken with a RGB camera (Sony a5000 alpha).
Table 1. Crop rotations and fertilizing regimes Rotation
405
The crop rotations under investigation are typical for farms operating biogas plants in the region.
Plant heights were derived from the digital terrain model by following an approach suggested by Bendig et al. (2015). A terrain model of the soil surface was created by neglecting the areas covered by plants. After subtracting the original DTM from the soil surface model the resulting dataset contains the plant heights (Crop Surface Model, CSM).
Triesdorf is located in Bavaria, Germany (49.210 N, 10.590 E) in the county of Ansbach which has the highest density of biogas plants in Germany. The average annual temperature is 7,7 0C. Average annual precipitation between 1960 and 1990 was 630 mm, in 2014 the annual precipitation was 661 mm.
After the parcel boundaries had been surveyed with a RTKGNSS receiver (Trimble Integrated FmX Display applying Ntrip corrections), median plant heights and median reflectances for all wavelengths captured with the multispectral camera have been assigned to the parcels using Quantum GIS.
This paper focusses on Triticale which was sown on October 1, 2013. Four parcel blocks with randomly arranged variations with respect to fertilization were established. Each parcel plot had a length of 8.20 m and was three parcels or 4.50 m wide. Yield was determined by harvesting only the center parcel (1.50 m wide).
The parcels were harvested on July 1, 2014. The dry matter yield was determined by drying and related to the area harvested resulting in dry matter yield in dt/ha (decitonnes per hectare).
The nitrogen fertilization was carried out on February 23, 2014 and March 24/25, 2014. Parcels treated solely with biogas slurry received 98 kg N /ha and 42 kg N /ha respectively. Parcels receiving both organic and mineral fertilizer received 70 kg N/ha in the form of biogas slurry as a starter dressing and a subsequent fertilizer dose of 70 kg N/ha Hydrosulfan. Parcels solely treated with mineral fertilizer received two fertilizer dressings with a dosage of 68 kg N/ha Hydrosulfan.
3. RESULTS 3.1 Dry Matter Yield As expected dry matter yield in parcels with no fertilization clearly underperformed (Ø 72 dt/ha) when compared to the parcels which received N-fertilization (Ø 152 dt/ha). The relative ranking within the repetitions revealed that parcels treated with mineral fertilizer performed best, followed by the parcels treated with biogas slurry only (Table 2).
All parcels have been treated with the herbicide Broadway on March 31, 2014. Triticale was harvested on July 1, 2014. 2.2 Measurement Devices and Data Sampling
Table 2. Relative Treatment Ranking The NDVI (Normalized Difference Vegetation Index) in the parcel plots under investigation has been measured with a Trimble GreenSeeker handheld device in the period between April 07, 2014 and June 30, 2014 on a weekly basis with four repetitions per parcel plot (576 samples). According to Wilkinson (2011) the sensor measures the reflection from the plant canopy at 650 nm (red) and 770 nm (near-infrared). The device features LEDs (Light emitting diodes) radiating light at the same wavelengths whenever a measurement is
Treatment
Fertilization
Rank 1 100% biogas slurry 2.0 2 50% biogas slurry/mineral fertilizer 2.8 3 100% mineral fertilizer 1.3 4 No fertilization 4.0 A comparison of the repetitions revealed that parcels which received fertilization in repetition 1 significantly produced 410
IFAC AGRICONTROL 2016 406 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
less biomass (130 – 151 dt/ha, 95% CI) than the parcels in repetition 4 (157 – 173 dt/ha, 95% CI). Repetitions 2 and 3 did not significantly differ from the other repetitions.
Table 3. Correlation between yield and NDVI measurements with GreenSeeker handheld 04/07
04/14
04/23
04/28
05/04
05/19
05/26
06/18
06/30
As expected, there is a close correlation between the NDVI calculated from the data collected with the Tetracam sensor and the NDVI measured with the GreenSeeker handheld (R2>0.9, Figure 1). This is true for the measurements taken with the GreenSeeker on May 5 and the pictures acquired with the Tetracam sensor on May 4 as well as for the data acquired on May 23 and May 26. In opposition, the NDVI from both sensor systems measured on June 30 does not correlate at all (R2>0.01).
Treatment
3.2 NDVI GreenSeeker vs. Tetracam Mini MCA6
1 2 3 4 All
1.00 0.69 0.97 -0.25 0.98
0.96 0.31 0.98 0.17 0.97
0.67 0.77 0.89 0.53 0.98
0.56 0.75 0.74 0.42 0.97
0.84 0.65 1.00 0.26 0.96
0.75 0.48 0.78 0.67 0.97
0.91 0.68 0.65 0.44 0.98
0.82 0.55 0.84 0.68 0.87
-0.20 0.28 -0.08 -0.31 0.01
Figure 2 shows that even early measurements of the NDVI (April 7) closely correlate with dry matter yield.
Fig. 1. Correlation between NDVI from Tetracam and NDVI from GreenSeeker handheld Fig. 2. Correlation between Triticale Dry Matter Yield and NDVI from measurements with a GreenSeeker handheld on April 7, 2014
Figure 1 shows that the NDVI acquired with the two sensor systems in the beginning of May significantly differ in spectral behavior. The slope of the regression term is approximately 2 where a value of 1 was expected. Also, the intercept has a very high value in the range of 1. The relation of data collected on May 23/May 26 is much closer to the expected range (slope 0.99, intercept 0.18).
However, it also shows that the high correlation coefficients are mainly due to the skew distribution of the data. The correlation between yield in fertilized parcels only and NDVI measured with the GreenSeeker handheld is much lower (R2= 0.70).
3.3 NDVI GreenSeeker vs. Dry Matter Yield The NDVI measurements with the GreenSeeker handheld device were sampled with four repetitions per parcel plot. The values obtained from single parcels were very consistent showing a low variation (mean standard deviation 0.02). The measurements taken in April and May were closely correlated to dry matter yield (R > 0.95, Table 2). In mid June (June 18) the correlation between dry matter yield and NDVI started decreasing with progressing senescence. The NDVI measurements taken on June 30 do not correlate with the dry matter yield harvested on the following day at all (R=0.01). Fig. 3. Correlation between Triticale Dry Matter Yield and NDVI from measurements with a GreenSeeker handheld on April 7, 2014, fertilized parcels only
It is noticeable that in the beginning of April the relation on treatment level is close for plots either dressed with mineral fertilizer or biogas slurry whereas the plots with mixed fertilizer treatment show a significantly lower correlation
Figure 3 shows that a very high correlation between dry matter yield and NDVI measurements with the GreenSeeker handheld can be established when grouping the data. Group1 represents data from parcels fertilized with biogas slurry only 411
IFAC AGRICONTROL 2016 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
and repetitions 1 and 2 from parcels fertilized with both fertilizers. Group2 contains data from parcels dressed with mineral fertilizer only and repetitions 3 and 4 from parcels fertilized with both fertilizers. In both groups yield is very closely correlated (R2= 0.98) to the NDVI measured on April 7.
407
yield is very close, despite one outlier in a parcel fertilized with both fertilizers.
3.4 NDVI Tetracam vs. Dry Matter Yield The median reflectances at 670 nm und 780 nm obtained from the Tetracam camera on May 5 and May 23 were used to calculate the NDVI and related to the dry matter yield (Figure 4). The close correlation (R2> 0.9) is again mainly due to the skew distribution of data. The NDVI measured in fertilized parcels with higher yields shows a very low response to yield.
Fig. 5. Triticale Dry Matter Yield versus reflectance at 780 nm measured with Tetracam on May 23, 2014 3.5 Crop Height Model vs. Dry Matter Yield The mean crop heights resulting from the digital terrain model derived from RGB pictures taken on May 5 show an extremely high correlation to dry matter yield (R2= 0.98).
Fig. 4. Triticale Dry Matter Yield versus NDVI from Tetracam on May 5 and May 23, 2014 The Tetracam camera collects reflectances at different wavelength, which can individually be related to yield without prior calculation of indices. Table 4 summarizes the correlation coefficients grouped by sampling dates and wavelengths. Both wavelengths related to NDVI show a close correlation to dry matter yield. The red channel (670 nm) is negatively correlated to yield on both dates (R< -0.9) while the infrared channel show a consistent positive correlation with yield (R = 0.98).
Fig. 6. Mean Crop Plant Height versus Triticale Dry Matter Yield May 5, 2014 The close relation is not only based on the uneven distribution of data. The value for Spearmen’s rank correlation between crop height and dry matter yield is also well above 0.9 (RSpearman = 0.93).
Table 4. Correlation between yield and spectral reflectance at different wavelength captured with Tetracam Mini MCA6
Date
Wavelength
7. DISCUSSION
530
670
700
730
780
May 5
-0.96
-0.95
-0.96
0.10
0.98
May 23
-0.88
-0.95
-0.91
0.87
0.98
In the field trial under investigation dry matter yield of Triticale varied substantially depending on fertilization and location. Parcels dressed with mineral fertilizer performed best when considering all repetitions. On the other hand, the yield in all fertilized parcels in repetition 1 was lower than the yield of corresponding plots in repetition 4. This indicates that the effect of soil variation may overrule the effect of fertilization even on a small scale (< 100 m).
Also the green channel (530 nm) and the second red channel (730 nm) are closely related to dry matter yield. On May5, the relation is even a bit closer than that of the red channel regarded for the NDVI. In Figure 4 dry matter yield is plotted against the reflectance at 780 nm. The coefficient of determination (R2) is 0.97. The relation between the reflectances measured and dry matter
The relation between the NDVI obtained from the GreenSeeker handheld and the Tetracam camera was looser than expected. Despite high coefficients of determination, 412
IFAC AGRICONTROL 2016 408 August 14-17, 2016. Seattle, Washington, USAP.O. Noack / IFAC-PapersOnLine 49-16 (2016) 404–408
the slope of regression and the intercept indicated substantial differences between the NDVI measurements. The time lag of 1 resp. 3 days between the measurements may have had an effect, but cannot appropriately explain the large deviations. The difference may partly be accounted to the different spectra covered for red (650 nm vs. 670 nm) and infrared (770 nm vs. 780 nm). It is also likely that the spatial resolution and the number of samples taken by the handheld sensor and the camera have affected the results.
better understand the spectral and the time domain of plant feature responses. 7. ACKNOWLEDGEMENTS The author would like to thank Markus Heinz from LLA Triesdorf for providing access to his trial fields and data collected by his staff. geo-konzept GmbH has supported this study with performing two UAS missions and processing the resulting data. Florian Baus, Dominik Weizmann, and Christian Schnitzlein deserve appreciation for very reliably collecting data with a handheld sensor on a weekly basis.
The relation between dry matter yield and the NDVI measurements taken with the GreenSeeker handheld were extremely close. This was even true for data collected at an early stage (April 7). An even closer relation could be established by arranging the data in groups. The results indicate that plots fertilized with mineral fertilizer generally produced higher NDVI values at the same level of dry matter yield when compared to plots fertilized with biogas slurry. Repetition 1 and 2 of those plots with mixed fertilisation align with the plots with biogas slurry treatment whereas repetition 3 and 4 were in line with the plots with mineral fertilization. This would again indicate that differences in soil property affect the behaviour of nitrogen uptake from mineral and organic sources.
REFERENCES Bach, H. (1998). Yield estimation of corn based on multitemporal LANDSAT-TM data as input for an agrometeorological model. Pure Appl. Opt. 7 (1998), pp. 809–825 Bendig,J., Yu, K., Aasen, H., Bolten, A., Bennertz, S., , Broscheit, J., Gnyp, M.L., Bareth, G. (2015). Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley. International Journal of Applied Earth Observation and Geoinformation 39 (2015), pp 79–87 Carter, G.A., and Knapp, A.K. (2001). Leaf optical properties in higher plants: linking spectral characteristics to stress and chlorophyll concentration. American Journal of Botany, April 2001, Vol. 88, No. 4, pp. 677-684 Gates, D.M., Keegan, H.J., Schleter, J.C. and Weidner, Victor R. (1965). Spectral Properties of Plants. Applied Optics, Vol. 4, No. 1, January 1965 Carter, G.A., and Knapp, A.K. (2001). Leaf optical properties in higher plants: linking spectral characteristics to stress and chlorophyll concentration. American Journal of Botany, April 2001, Vol. 88, No. 4, pp. 677-684 Lilienthal, H. (2011). Estimating Crop Parameters by processing hyperspectral canopy reflectance data. International Symposium on Sensing in Agriculture. February 21 - 24, 2011. Haifa Santin-Janin, H., Garel, M., Chapuis, J.-L, Pontier, D. (2009) Assessing the performance of NDVI as a proxy for plant biomass using non-linear models: a case study on the Kerguelen archipelago, Polar Biol (2009) 32:pp 861–871 Schmidhalter, U., Jungert, S., Bredemeier, C. , Gutser, R., Manhart, R., Mistele, B., and Gerl, G. (2003). Field-scale validation of a tractor based multispectral crop scanner to determine biomass and nitrogen uptake of winter wheat. Stafford, J.,Werner, A.(eds): Precision Agriculture. Wageningen Academic Publishers, pp 615-619 Wilkerson, J. (2011): Real-Time Sensor-Based Nutrient Management,http://www.utcrops.com/Presentations/Sens ors%20and%20Nutrients%20-%20Wilkerson.pdf, last visited February 29, 2016
The relation between the NDVI calculated from the reflectances captured with the Tetracam camera and dry matter yield was looser than expected. It is mainly the uneven distribution of data that is accountable for the high coefficients of determination. Especially in the fertilized plots the NDVI response on increasing yield was very low and inconsistent. This may be due to the saturation of NDVI when it comes to high biomass values as suggested by Santin-Janin et al. (2009). On the contrary, the NDVI measurements obtained with the GreenSeeker handheld were closely correlated despite high biomasses. Reflectances measured with the Tetracam camera at single wavelengths were partly much closer related to dry matter yield than the NDVI. The reflectances measured at 780 nm on May 25 were very much in line with dry matter yield, except from one outlier. This result supports the approach of Lilienthal (2011) who suggests applying complex modelling techniques (e.g. SVM) to spectral reflectance measurements at different wavelength rather than calculating indices. The relation between mean crop height of Triticale derived from a digital terrain model on the one hand and dry matter yield on the other hand was much closer (R2=0.98) than the relation between yield and the spectral data from the Tetracam camera. In case of Triticale, plant height seems to be the best proxy for dry matter yield when compared to spectral measurements. This study is limited to the measurement of biomass of Triticale. The results do not allow for the prediction of dry matter yield of other plants or even grain yield. However, the results show that there is high potential in spectral measurements with low-cost sensors (GreenSeeker) and high end technology systems like UAV for predicting plant properties such as biomass. Further research is necessary to
413