- Email: [email protected]
Field-based crop phenotyping: Multispectral aerial imaging for evaluation of winter wheat emergence and spring stand
Computers and Electronics in Agriculture 118 (2015) 372–379
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Original papers
Field-based crop phenotyping: Multispectral aerial imaging for evaluation of winter wheat emergence and spring stand Sindhuja Sankaran a,⇑, Lav R. Khot a,b, Arron H. Carter c a
Department of Biological Systems Engineering, Washington State University, PO Box 646120, Pullman, WA 99164, United States Center for Precision and Automated Agricultural Systems, Irrigated Agricultural Research and Extension Center, Washington State University, 24106 North Bunn Road, Prosser, WA 99350, United States c Department of Crop and Soil Sciences, Washington State University, PO Box 646420, Pullman, WA 99164, United States b
a r t i c l e
i n f o
Article history: Received 16 September 2014 Received in revised form 28 August 2015 Accepted 5 September 2015
Keywords: Plant breeding Unmanned aerial vehicles Remote sensing Crop growth
a b s t r a c t The physical growing environment of winter wheat can critically be affected by micro-climatic and seasonal changes in a given agroclimatic zone. Therefore, winter wheat breeding efforts across the globe focus heavily on emergence and winter survival, as these traits must first be accomplished before yield potential can be evaluated. In this study, multispectral imaging using unmanned aerial vehicle was investigated for evaluation of seedling emergence and spring stand (an estimate of winter survival) of three winter wheat market classes in Washington State. The studied market classes were soft white club, hard red, and soft white winter wheat varieties. Strong correlation between the ground-truth and aerial image-based emergence (Pearson correlation coefficient, r = 0.87) and spring stand (r = 0.86) estimates was established. Overall, aerial sensing technique can be a useful tool to evaluate emergence and spring stand phenotypic traits. Also, the image database can serve as a virtual record during winter wheat variety development and may be used to evaluate the variety performance over the study years. Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction The United States is one of the largest wheat producing country in the world (Carver, 2009). Within the U.S., Washington State produces both winter and spring wheat in various market classes such as hard red and white, soft white, and club wheat, with a production value of up to $1.18 billion annually. About 80–90% of the wheat produced in Washington is exported to countries such as Japan, Philippines, Taiwan, South Korea and Yemen (Washington Wheat Commission, 2008). Although growers produce both spring and winter wheat varieties in different market classes, approximately 75% of the wheat is planted as winter wheat, and 60% is planted as soft white winter wheat (Washington Wheat Commission, 2008). Winter wheat is the preferred wheat market class because they are high yielding (3.8 MT/ha in comparison to 2.8 MT/ha for spring wheat) and require low farm inputs. Moreover, winter wheat plants establish a soil cover to reduce wind and water erosion during the winter months (Carver, 2009). The specific physiological changes unique to winter wheat are acclimation and vernalization that occur during cold winter months. Cold acclimation allows the plants to survive the winter ⇑ Corresponding author. Tel.: +1 509 335 8828; fax: +1 509 335 2722. E-mail address: [email protected] (S. Sankaran). http://dx.doi.org/10.1016/j.compag.2015.09.001 0168-1699/Ó 2015 Elsevier B.V. All rights reserved.
and vernalization induces reproductive growth, which triggers flowering the following summer (Skinner, 2009; Snape et al., 2001). Vernalization requires growth periods in temperatures less than 4 °C, whereas freeze–thaw cycles leading up to cold winter months enhances cold tolerance in wheat (Skinner and Bellinger, 2010). If the winter hardiness of a plant variety is not sufficient to withstand extremely low temperatures (16 °C or lower), the plants will die. Similarly, if the vernalization is not met completely, it may delay plant maturity due to delayed reproductive growth (Yan, 2009). Different winter wheat varieties will differ in the period needed for vernalization and winter hardiness. Therefore, these factors are critical phenotypic traits that need to be evaluated during new winter wheat variety development. In Washington State, various agroclimatic zones provide variation in the physical growing environment of winter wheat, and thus the traits needed to maintain high yield potential. In the east-central part of Washington, wheat is grown under annual rainfall, which is typically limited to less than 30.5 cm annually (Hasslen and McCall, 1995; Higginbotham et al., 2011, 2013; Martínez et al., 2013). As such, a fallow year is incorporated into the rotation to try and store moisture to maximize crop production (Juergens et al., 2004; Higginbotham et al., 2013). After considerably dry years, wheat must be planted 15–20 cm deep in order to reach moisture and emerge (Schillinger et al., 1998; Schillinger
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
and Paulitz, 2014; Young et al., 2014). During the germination stage, the seedling root and coleoptile (leaf-like structure that protects the first leaf from the soil) develop. Several factors including soil and air temperature, water availability, oxygen, radiation intensity, substrate, maturity of the seeds, and physiological age of the seed affects the germination of winter wheat (Lindstrom et al., 1976; Winter et al., 1988; Giri and Schillinger, 2003). A limitation to emergence can be rain events. If rain occurs before plant emergence, a solid crust forms on the soil surface which further impedes the emergence of wheat seedlings (Schillinger et al., 1998; Young et al., 2014). After successful emergence, plants must then survive the cold winter months. This typically occurs as plants acclimate to the weather and go into dormancy, with a layer of snow cover protecting them from the coldest of months (Skinner, 2009; Young et al., 2014). In years of extremely low temperatures or no snow cover, wheat seedlings become very susceptible to winter kill. Thus, two very important traits for wheat cultivars grown in the east-central regions of Washington are the ability to emerge from deep planting through crust events, and winter survival. Researchers have reported proximal and remote sensing applications using ground and aerial platforms for vegetation dynamics such as leaf area index in different scales (Gitelson, 2004; Zhang et al., 2003; Lelong et al., 2008; Hunt et al., 2010, 2013; Fiorani et al., 2012; Sharabian et al., 2013; Deery et al., 2014). The platforms can range from field tractors, unmanned aerial vehicles (UAVs), or satellites. Although the satellite images can cover a large area, frequency of data collection, legal spatial resolution, and environmental factors including cloud cover can limit image-based crop evaluation (Zhang and Kovacs, 2012). Similarly, ground platforms can be limited by field conditions (especially after irrigation or rainfall) and coverage area at a given time. Such limitations can potentially be addressed by rapidly evolving UAV technology. Low cost UAVs may allow rapid need-based high resolution imaging critical for field phenotyping as number of trials and size of field plots are smaller in such applications. Some of the technological limitations of the UAV platforms for agricultural imaging include limited flight time (associated with battery life), and payload carrying capabilities (Hardin and Hardin, 2010; Zhang and Kovacs, 2012). Digital images can be used for detecting early plant vigor. Recently, 50 winter wheat cultivars were monitored for two years to establish the relationship between early plant vigor index (EPVI was defined as the ratio between the difference of reflectance at 750 and 670 nm, with that of reflectance at 862 nm) and relative amount of green pixels, and a relationship with a regression coefficient of 0.98 was found (Kipp et al., 2014). Although strong relationships have been found to estimate plant vigor, the utilization for field evaluation during plant breeding is relatively new, and not used as a standard method. Therefore, the objective of this study was to evaluate the potential of low altitude remote sensing technology as a high-throughput phenotyping tool for assessment of winter wheat emergence (vigor) and winter survival (winter hardiness) under field conditions.
2. Materials and methods 2.1. Field plots Winter wheat plots were planted into a summer fallow field in Kahlotus, WA on 23rd August, 2013. Plots planted were soft white and hard red winter wheat varieties from the Washington State University (WSU) Wheat Breeding Program, and soft white club wheat varieties from the U.S. Department of AgricultureAgricultural Research Service (USDA-ARS) Wheat Breeding Program. Plots were planted in an alpha-lattice design using a custom
373
built deep-furrow small plot planter consisting of 4 rows. Alphalattice design method divides replicates into incomplete blocks that contain a part of the total number of entries (Patterson and Williams, 1976). Alpha-lattice designs have been shown to be more efficient at analyzing data in large entry plant breeding trials when compared to randomized complete block designs (Yau, 1997). In this study, genotypes were distributed among the blocks in such a way so that all pairs occur in the same block at equal frequencies, thereby permitting removal of incomplete block effects from the plot residuals (Patterson et al., 1978). The seeds were placed at 9 cm soil depth, and plot size was 1.5 m 4.6 m. Each variety was replicated three times, except the club wheat trial that had one to two replicates (Fig. 1). The soil type in Kahlotus is a Ritzville silt loam. Currently grown cultivars in each market class (soft white, hard red, and soft white club) were planted as check cultivars (reference plots), and the remaining plots were experimental breeding lines from the two wheat breeding programs. Base fertilizer in the field was 27.2 kg of N, and all other standard field practices were performed by the cooperating grower. On 25th August, 5.6 mm of rain was recorded followed by high daytime temperatures, which created a solid crust on the soil surface approximately 5 mm thick. The first plants were observed to emerge from the soil on 30th August. Emergence continued for another ten days. On 25th September, plots were rated visually for emergence. Each plot (1.5 m 4.6 m area) was scored on a scale of 0–100% emergence using 10% increments. Plots were again rated visually on 9th April, 2014 to determine winter survival (spring stand). 2.2. Aerial data collection An unmanned aerial vehicle was used to acquire high resolution multispectral images to evaluate the winter wheat emergence (2nd October, 2013) and winter survival (16th April, 2013). The UAV (HiSystems GmbH, Moormerland, Germany) system weighs about 2 kg without the imaging system and a 6000 mA h Lithium Ion Polymer battery that provides the required power for UAV flight. The flight time can range from 10 to 20 min depending on the payload and wind conditions. The maximum recommended payload of this system is 2.5 kg. The UAV has eight brushless motors and individual motors can handle 20 A power with a maximum thrust of 2200 g. Based on preliminary flights, beechwood propellers (Xoar International, CA, USA) were found to improve the stability of the UAV platform during imaging and were used in this study. The UAV comprises of an array of onboard sensors for flight stability and waypoint navigation such as gyroscope, accelerometer, compass, global positioning system receiver, and pressure sensor. A radio transmitter (MX20 Hott, Graupner, Stuttgart, Germany) with range of up to 4 km was used to remotely control the UAV. A modified multispectral digital camera, XNiteCanon SX230 NDVI (LDC LLC, Carlstadt, NJ) with near infrared (670–750 nm) (NIR), green (G), and blue (B) bands was used for aerial imaging. The camera was mounted on a mount underneath the UAV that is capable of automatically adjusting to the nick and roll shifts during flight. Alternately, the nick and roll can be adjusted using the radio control transmitter. A firmware was used to enable georeferenced interval shooting to acquire images every 5 s during UAV flight. The captured 8-bit JPG images with resolution of 12.1 megapixels (4000 3000) were stored on-board the camera card. During preliminary evaluation, flying altitude was adjusted to acquire a multispectral camera image covering the study plots. At this time, a simple digital camera (Sony NEX-5N, Sony Electronics Inc.) was also used to estimate the spatial resolution of the two cameras at different altitudes. Sony NEX-5N is a 16.1 megapixel camera with a 4912 3264 image resolution. The images
374
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
CS-25
CS-26
CS-27
CS-28
CS-29
CS-30
CS-31
CS-32
CS-33
CS-34
CS-35
CS-36
CS-37
CS-38
CS-39
CS-40
CS-41
CS-42
CS-7
CS-8
CS-9
CS-10
CS-11
CS-12
CS-1
CS-2
CS-3
CS-4
CS-5
CS-6
CS-13 WC-28
CS-14 WC-19
CS-15 WC-15
CS-16 WC-4
CS-17 WC-12
CS-18 WC-29
CS-19 WC-3
CS-20 WC-8
CS-21 WC-1
CS-22 WC-30
CS-23 WC-20
CS-24 WC-11
WC-10
WC-2
WC-24
WC-22
WC-9
WC-17
WC-18
WC-26
WC-21
WC-25
WC-7
WC-13
WC-6
WC-23
WC-16
WC-14
WC-27
WC-5
WC-22 DC-8
WC-7 DC-24
WC-21 DC-18
WC-28 DC-29
WC-29 DC-6
WC-24 DC-13
WC-16
WC-11
WC-26
WC-3
WC-19
WC-8
WC-23
WC-17
WC-14
WC-30
WC-18
WC-9
DC-19
DC-21
DC-7
DC-30
DC-2
DC-12
WC-6 DC-15
WC-1 DC-22
WC-5 DC-28
WC-2 DC-17
WC-15 DC-14
WC-27 DC-26
WC-12 DC-1
WC-13 DC-11
WC-25 DC-5
WC-10 DC-25
WC-4 DC-23
WC-20 DC-4
DC-5
DC-25
DC-23
DC-22
DC-1
DC-4
DC-9
DC-29
DC-18
DC-12
DC-8
DC-6
DC-20
DC-27
DC-10
DC-9
DC-3
DC-16
DC-16
DC-17
DC-30
DC-13
DC-28
DC-20
DC-7
DC-2
DC-3
DC-24
DC-14
DC-21
DC-10
DC-19
DC-26
DC-27
DC-11
DC-15
B2 HR-5
B2 HR-26
B2 HR-21
B2 HR-17
B2 HR-35
B2 HR-24
B2 HR-26
B2 HR-36
B2 HR-27
B2 HR-33
B2 HR-34
B2 HR-23
B2 HR-3
B2 HR-19
B2 HR-27
B2 HR-20
B2 HR-29
B2 HR-18
HR-16
HR-15
HR-6
HR-29
HR-34
HR-31
HR-4
HR-25
HR-24
HR-8
HR-29
HR-10
HR-24
HR-16
HR-9
HR-35
HR-33
HR-17
HR-36
HR-11
HR-27
HR-20
HR-12
HR-13
HR-18
HR-11
HR-7
HR-20
HR-14
HR-28
HR-21
HR-13
HR-36
HR-14
HR-28
HR-32
HR-23
HR-22
HR-28
HR-33
HR-19
HR-7
HR-30
HR-17
HR-16
HR-3
HR-21
HR-6
HR-4
HR-26
HR-8
HR-34
HR-7
HR-5
HR-30
HR-18
HR-4
HR-2
HR-9
HR-14
HR-19
HR-22
HR-1
HR-12
HR-9
HR-13
HR-15
HR-1
HR-25
HR-6
HR-2
HR-23
HR-25
HR-10
HR-8
HR-32
HR-1
HR-3
HR-15
HR-2
HR-31
HR-32
HR-5
HR-35
HR-22
HR-30
HR-10
HR-31
HR-12
HR-11
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
SW-4
SW-2
SW-19
SW-9
SW-16
SW-8
SW-32
SW-7
SW-31
SW-30
SW-17
SW-20
SW-4
SW-32
SW-30
SW-33
SW-8
SW-17
SW-23
SW-15
SW-28
SW-32
SW-17
SW-21
SW-27
SW-12
SW-21
SW-10
SW-23
SW-24
SW-10
SW-35
SW-23
SW-36
SW-5
SW-27
SW-20
SW-6
SW-22
SW-1
SW-24
SW-11
SW-36
SW-26
SW-28
SW-33
SW-22
SW-18
SW-22
SW-25
SW-12
SW-24
SW-31
SW-20 SW-13
SW-26
SW-29
SW-12
SW-27
SW-14
SW-25
SW-16
SW-13
SW-35
SW-11
SW-5
SW-15
SW-11
SW-9
SW-18
SW-3
SW-26
SW-3
SW-31
SW-10
SW-33
SW-7
SW-34
SW-25
SW-9
SW-29
SW-4
SW-6
SW-34
SW-28
SW-29
SW-16
SW-21
SW-19
SW-2
SW-5
SW-13
SW-36
SW-35
SW-30
SW-18
SW-2
SW-8
SW-1
SW-19
SW-14
SW-3
SW-7
SW-15
SW-1
SW-6
SW-14
SW-34
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
Fig. 1. Field map showing distribution of replicate plots in the experiment. The B1, B2 and B3 plots were soft white club wheat Bruehl, B4 plots were soft white common wheat Otto, SW1–SW36 refers to 36 soft wheat winter wheat varieties, HR1–HR36 refers to 36 hard red winter wheat varieties, DC1–DC20 refers 30 ‘dry’ soft white club, WC1–WC20 refers to 30 ‘wet’ soft white club, and CS refers to single entries of different soft white club varieties.
(Red–Green–Blue [RGB] bands) were acquired from 100 m, 75 m, and 55 m altitude. Fig. 2a displays the false color multispectral image acquired from 100 m altitude showing representative plots. The RGB images had spatial resolution of about 2.3 cm/pixel, 2.0 cm/pixel, and 1.4 cm/pixel at 100 m, 75 m and 55 m, respectively. Similarly, for respective altitudes, the corresponding spatial resolution was 3.0 cm/pixel, 2.3 cm/pixel, and 1.6 cm/pixel in multispectral images. Using the above information and methodology, a total of 378 plots (450 plots with border rows) were imaged using the multispectral camera from 100 m altitude for emergence and spring stand evaluation. The image recorded three wheat market classes, soft white club (162 plots), hard red (108 plots), and soft white (108 plots) planted in-between the border rows. The border rows were soft white club wheat Bruehl and soft white common wheat Otto (Fig. 1). 2.3. Image processing and analysis Data analysis was performed to estimate the emergence or spring stand after winter survival from individual trial plots in the imaged study area. Feature extraction and analysis algorithms were developed using ImageJ (Schneider et al., 2012) and MatlabÒ (v. R2009b, The MathWorks, Inc. MA). The feature extraction steps involved: (a) separating an image into three bands, (b) extracting green normalized difference vegetation index (GNDVI) image, (c) thresholding of the GNDVI image to remove background, and (d) convert the GNDVI into binary image. The GNDVI (Sankaran et al., 2013) was extracted using Eq. (1), where NIR is the reflectance at near infrared band and G is the reflectance at green band:
GNDVI ¼
NIR G NIR þ G
ð1Þ
The GNDVI images were further processed to remove the soil and background noise using threshold. Resulting images were converted to binary image and the average pixel values for individual
plots were calculated as the crop emergence or spring stand (Eq. (2)), where N represents number of pixels in individual plot, and GNDVIi refers to GNDVI pixel values of individual pixel.
X¼
N 1X GNDVIi N i¼1
ð2Þ
The binary image pixel value was 0 for soil and 255 for plant region. The same equation was used to calculate emergence and spring stand, each representing two different plant traits. The difference in crop trait estimation is based on different time periods that the images were acquired. Winter wheat emergence was estimated from images acquired in the Fall (about one month after planting), and spring stand was estimated from images acquired in the Spring (about 1 month after plants have broken dormancy requirements and begin to resume normal growth, assuming they survived the winter and were not killed). The image results were correlated with ground-truth measurements (as described in Section 2.1) taken from the same plots. 2.4. Seasonal micro-climatic variation Preliminary analysis of air temperature data from a nearby weather station (AgWeatherNet, Hatton, WA) for the study period (interval between two aerial data collections) indicated that for 112 days, the average air temperature, recorded at 1.5 m above ground surface, was <0 °C. Although most winter wheat varieties can survive temperatures below 0 °C, temperatures below 10 °C can start to differentiate varieties and their ability to survive winter temperatures. On 4th December 2013, minimum daytime temperatures decreased rapidly, and the following eight day period ranged from 13 to 20 °C daytime minimum temperature, and averaged 15 °C. Similarly, the soil temperature at 20 cm depth was <0 °C for 105 days during the same time interval, and averaged 3 °C during this eight day long cold period. Another cold period during the first week of February had minimum air temperatures averaging 11 °C for seven days, with maximum temperatures only averaging 8 °C.
375
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
Fig. 2. (a) UAV-based false color (NIR, G, B as RGB) images of representative winter wheat field plots showing emergence, and (b) corresponding pseudocolor GNDVI image extracted from multispectral images. The colors represents different GNDVI scales, commonly used for better visualization of GNDVI image. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
During the time between aerial imaging, no snow cover was present on the wheat seedlings, and only 7.6 cm of total precipitation was reported. Considering the two weeks of low temperatures, the low precipitation and drought conditions, along with no snow cover, a low winter survival rate could be anticipated. 2.5. Statistical analysis Statistical analysis of percent emergence and percent spring stand was analyzed using SAS version 9.3 (SAS Institute, Raleigh, NC). The ‘Proc GLM’ procedure was used to test the replication, genotype, and interaction effect for each of the three market class trials separately. Pearson’s correlations analysis was conducted using ‘Proc CORR’ procedure in SAS. 3. Results and discussion 3.1. Winter wheat emergence and stand Germination and emergence rates depend on the wheat variety. Under favorable conditions, seedling emergence can occur within
seven days. However, deep planting or rain events can cause a delay and reduction in seedling emergence. Standard methods of evaluating this phenotype involve visual rating of individual plots. Aerial image-based quantification technique developed in this study can be an alternative method to evaluate the wheat plant emergence phenotype under field conditions. Binary image extracted from processed GNDVI image (Fig. 2b) after thresholding. The white area corresponds to emergence or spring stand of wheat plants and the black area corresponds to non-emergence. The correlation of plant emergence and the spring stand with ground-truth data for individual wheat varieties (plotto-plot comparison in each variety class) are summarized in Table 1 Results of correlation (P < 0.0001) between UAV image-based plant emergence (or spring stand) and ground-truth data for winter wheat varieties planted in Kahlotus, WA. Wheat variety
Average emergence (%)
r
Average spring stand (%)
r
Soft white club Hard red Soft white
45.9 59.1 64.7
0.82 0.93 0.86
29.5 51.7 47.1
0.84 0.88 0.83
376
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
Table 1. ANOVA indicated that there were significant differences (P < 0.0001) between genotypes (breeding lines) in all three market class trials for both percent emergence and percent spring stand. There was no replication effect (significant differences between the replicate plots) for percent spring stand in any of the market class trials or for the percent emergence in the club wheat trial. There was a significant effect (P < 0.0001) of replication for percent emergence in both the soft white and hard red market class trials. This significant effect was found to be a scalar interaction, with replicate one (the front of the trial), two, and three having an average emergence of 53%, 65%, and 68%, respectively. Relative rankings of each genotype within each replicate did not change and there was no significant genotype replicate interaction found. A good correlation between the aerial image-based winter emergence (%) and ground-truth was observed with Pearson correlation coefficient (r) ranging from 0.82 to 0.93 (Table 1). Plot-toplot comparisons combining all three market classes indicated the r for emergence was 0.87 (P < 0.0001). Similarly, for plant spring stand after the winter season, the overall r between the aerial image data and ground-truth was 0.86 (P < 0.0001). Based on the results presented in Table 1, a high correlation between ground-truth and UAV multispectral image data was observed. Comparison of average emergence and spring stand rates indicated that the wheat varieties in all three market classes were affected by winter weather. It should again be noted here that the varieties studied in this experiment were unreleased breeding lines, and variation in both emergence and winter survival were expected to be present between the lines. Amongst the varieties tested, a drastic reduction in spring stand rates was observed with the soft white club varieties, whereas the hard red wheat varieties were found to be more winter hardy as a whole. Furthermore, for determining the plant growth relationship, the average emergence/stand rates for each cultivar in the three
market classes were computed by averaging the ground-truth data from replicate plots. Similarly, the emergence and spring stand data estimated using aerial images for each variety were averaged, and compared with ground-truth ratings (Fig. 3). A high correlation (>0.92) was observed between the estimates and ground-truth data for the soft white and hard red varieties. For the soft white club varieties, the correlation coefficient was between 0.86 and 0.91 for the ‘dry’ and ‘wet’ soft white club varieties. Comparing all the average ground-truth ratings (averaging the replicates) of all varieties from three market classes (hard red, soft white, and soft club white) with that of corresponding image data, the r was 0.91 for emergence and 0.92 during spring stand evaluation (P < 0.0001). 3.2. Differences between emergence and winter survival The winter wheat varieties were further analyzed to determine the relationship between the ground-truth emergence/stand rates of individual varieties (36 varieties 3 replicates for soft white and hard red each; 30 varieties 2 replicates for ‘dry’ and ‘wet’ soft white club each) with that of image data. After analysis, four distinct types of breeding lines were observed in the trials. These were the lines with (1) good emergence and spring stands, (2) good emergence and poor spring stands, (3) poor emergence and good spring stands, and (4) poor emergence and poor spring stands. For example, the three check (reference) cultivars (Bruehl, Xerpha, and Otto) used in the breeding program for soft white wheat emerged very well during the fall, with average emergence of 87%, 90%, and 100%, respectively. Spring stands of these varieties were 47%, 53%, and 87%, respectively, indicating the ability of the cultivar Otto to survive the winter better than Bruehl or Xerpha. A similar trend was observed in the hard wheat trial, with the check (reference) cultivars (Bauermeister, Finley, and Farnum)
160
120 100 80 60 40 20 0
Emergence, r = 0.97 Spring Stand, r = 0.93
140 120 100 80 60 40 20 0
(a) Soft white
0 160 Image Data (pixel value)
Image Data (pixel value)
Emergence, r = 0.93 Spring Stand, r = 0.92
140
20
40 60 Visual Rating (%)
80
(b) Hard Red
100 160
Emergence, r = 0.91 Spring Stand, r = 0.89
140
0
Image Data (pixel value)
Image Data (pixel value)
160
120 100 80 60 40 20 0
20
20
40 60 Visual Rating (%)
80
100
80
100
Emergence, r = 0.86 Spring Stand, r = 0.91
140 120 100 80 60 40 20 0
(d) 'Wet' soft white club
(c) 'Dry' soft white club
0
40 60 Visual Rating (%)
0
20
40 60 Visual Rating (%)
80
100
Fig. 3. Correlations (P < 0.0001) between average ground-truth emergence/spring stand and image data for hard red, soft white, and soft white club (dry and wet) winter wheat varieties (averaging the replicate plots for each cultivar within a market class).
377
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
having good fall emergence ratings (80%, 97% and 100%, respectively), yet spring stands (50%, 83%, and 90%, respectively) indicate Finley and Farnum have better winter tolerance than Bauermeister. Overall, there were 17% of the plots which showed good emergence and good winter survival (as indicated by a >70% stand), and these will be useful to advance in the breeding programs for further evaluation for cultivar release. On average, 23% of lines had good emergence in the fall, but poor spring stands, indicating these lines are not tolerant to cold winter temperatures. There were 45% of the lines which had poor emergence and poor spring stands, although no observations can be made about their ability to withstand winter conditions as few plants were present in the fall. About 14% of the lines showed poor emergence characteristics, but had an increase in the stand percentage in the spring, indicating these lines survived the winter well, but were also able to take advantage of the available space and moisture to produce more tillers per plant, giving the indication of a greater spring stand then was present at emergence. The varieties with higher than 25% change in loss of vigor after winter and ones showing increased vigor (>10% difference) after winter are summarized in Table 2. Lines with a positive number indicate lines which experienced some winter kill, whereas those with negative numbers are the lines which demonstrated an increase in the percent stand observed from fall to spring. Lines with negative numbers will be useful in breeding programs to identify lines with increased tillering and overall competitiveness, but which will need increased
emergence characteristics. The overall performance of these varieties can clearly be observed in Fig. 4. Visual emergence and spring stand ranking, which can overlay plot images have been shown in Fig. 4. Continuous blue rows represent the border rows, with the first and last border rows not shown. As seen from Fig. 4a and b, the soft white wheat breeding lines had better emergence potential than the hard red before winter (>55%), whereas the hard red wheat breeding lines tended to withstand winter better than the soft white between the tested varieties. The WSU winter wheat breeding program has a long history of breeding soft white winter wheat cultivars for excellent emergence potential. Thus, it is expected that the germplasm in this market class have better emergence than the hard red or club wheat market classes. Observing the individual breeding lines within each market class, expected trends can be detected. In the soft wheat market class, there were more lines which had above average emergence, but only a few listed in the 80–100% range and only a few in the 0–20% range. Contrast to that, in the hard red wheat market class, there were many more lines in the 80– 100% emergence range, and many in the 0–20% range, with few in between. Thus, there was a trend to a lower overall emergence average in the hard red market class. The club wheat program had the most lines with poor emergence, which is to be expected as this program has a smaller germplasm pool for good emerging lines and has not been screening breeding material for emergence as long as the soft white and hard red programs have been.
Table 2 Emergence (October 2013) and spring stand (April 2014) percentages of selected winter wheat breeding lines grown in Kahlotus, WA. Plot number*
Type
Variety (Trail No.y)
352, 289, 299, 273, 278, 276, 280, 291, 334, 359, 286, 347,
Soft club winter wheat program
27, 64, 77, 57, 52, 25, 51, 46, 49, 75, 58, 28, 24,
391 321 325 310 311 301 326 323 362 367 302 390
33, 76 67, 78 88, 102 87, 98 96, 99 54, 110 61, 113 109, 121 81, 93 83, 85 97, 108 53, 111 69, 84
162, 149, 178, 164, 147, 155, 196, 169, 158, 153, 194, 157, 199,
200, 206 171, 176 192, 219 205, 252 195, 226 198, 235 221,245 185, 248 163, 187 160, 222 236, 241 170, 182 213, 242
Emergence rating (%)
Spring stand rating (%)
Difference = emergence stand (%)
BRUEHL (WC-3) XERPHA (DC-5) X010746-4C (DC-8) HS980158-1L (DC-30) ARS-Selbu (DC-2) X000175-1-1 (DC-20) X070070-0-0-16L (DC-24) X070051-0-0-47C (DC-23) ARS-Chrystal (WC-2) HS00293-2C-1 (WC-18) X06135-9C (DC-27) X20060126-0-3C (WC-29)
85.0 70.0 65.0 80.0 70.0 95.0 95.0 65.0 70.0 80.0 5.0 20.0
60.0 45.0 40.0 50.0 40.0 50.0 30.0 0.0 45.0 25.0 20.0 35.0
25 25 25 30 30 45 65 65 25 55 15 15
Soft white winter wheat program
Eltan (SW-1) WA 8199 (SW-11) 3J080408-0-15 (SW-24) 5J062268-1 (SW-12) 3J080209-0-1 (SW-21) Xerpha (SW-2) WA 8201 (SW-16) Bruehl (SW-4) SSDRES193-3-SW (SW-28) 3J080221-0-3 (SW-22) 3J080565-0-1 (SW-27) WA 8202 (SW-19) SSD760097-5-R (SW-18)
100.0 90.0 76.7 70.0 70.0 90.0 76.7 86.7 83.3 70.0 83.3 50.0 36.7
76.7 66.7 50.0 36.7 33.3 53.3 40.0 46.7 40.0 20.0 16.7 63.3 60.0
23 23 27 33 37 37 37 40 43 50 67 13 23
Hard red winter wheat program
WA 8180 (HR-11) Bauermeister (HR-1) Sprinter (HR-6) WA 8208 (HR-18) WA008156 (HR-8) NORWEST 553 (HR-5) FarEd277T (HR-34) 3J080150-0-1 (HR-19) 3J080125-0-T1 (HR-30) 3J080147-0-T1 (HR-31) FarEd176 (HR-26) FarEd32 (HR-22) W7YSSD100001 (HR-36)
96.7 80.0 93.3 80.0 80.0 76.7 53.3 23.3 20.0 40.0 30.0 30.0 30.0
66.7 50.0 60.0 43.3 16.7 3.3 66.7 40.0 36.7 56.7 46.7 53.3 53.3
30 30 33 37 63 73 13 17 17 17 17 23 23
* The plots were replicated three times and numbered from left to right and bottom to top. The plots 1–18, 127–144, 253–270, 433–450 were border rows. The plots 19– 126, 145–252, 271–330, and 331–396 were soft white, hard red, ‘dry’ soft white club, and ‘wet’ soft white club breeding lines, respectively. y Trail No. is referenced to Fig. 1.
378
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
100 2
90
4 80
6
90 2 80 4 6
70
8
60
70
8 10
60
12
50
12
14
40
14
30
16
30
18
20
16 18
20 20
10
50 40
20 10
22
10 22
0 5
10
0
15
5
(a) Emergence
10
15
(b) Spring Stand
2
80
4 60
6 8
40
10 12
20 14 16
0
18 -20
20 22
-40 5
10
15
(c) Emergence -Spring Stand Fig. 4. Visual rating of (a) emergence, (b) winter wheat spring stand, and (c) spring stand rates subtracted from emergence rates. Each box represents individual plots. 18 plots from left to right, and 23 plots rows from top to bottom.
The spring stand image (Fig. 4b) shows similar results. The club wheat program had the most lines with poor spring stands, in part due to winter kill, but in part due to an initial poor emergence. Although the soft white program had overall the best emergence potential of the lines tested, the increase in the number of plots with a poor spring stand indicates that more of these plants died over the winter, and thus had less winter survival as a market class than did the hard red varieties. The hard red varieties showed an overall improvement in plant stand over the winter months. Although some lines which emerged well experienced some winter kill, there were other lines which had an average emergence of 40– 60% which had an increase in spring stand percentage. The spring stand was calculated based on the percent greenness observed, not the individual plants present. So, although the number of plants in the hard red plots may have stayed the same, they were covering more space. This indicates that the tillering capacity of the hard red cultivars was above average, and the plants which did emerge and survive the winter were able to produce more tillers, effectively taking advantage of the available space and increasing the number of wheat heads per square foot. Comparing the winter wheat emergence and spring stand percentages (Fig. 4c), the ones seen in red and orange are varieties that had good emergence rates, but did not survive the winter.
Similarly, the ones marked in blue are the varieties that did not have high emergence rates, but had a good spring stand after winter. Again, the lines in blue indicate lines which, although few plants emerged, were able to take advantage of the space around them and produce more tillers per square foot. This is an important trait for wheat breeders to select on, as tillering and competitiveness are beneficial traits in wheat production where annual rainfall is limited. The ability of breeders to visualize this phenotype is enhanced through the aerial images provided by this study. Overall, the study successfully evaluated emergence and spring stand of winter wheat using UAV-based images. Both are critical phenotypes during winter wheat variety development. Existing standard methods involves visual rating from 0% to 100%, with each plot given a rating in increments of 10%. Such evaluation may be biased and subject to human error. The results from this study confirm that aerial sensing techniques are reliable (with significant correlations), and at the same time comparable to visual evaluation. Given the high correlation achieved in this study, such sensing technique tools can be useful to evaluate the phenotypes of emergence and spring stand. Moreover, the images can also serve as a visual record during winter wheat variety development and can be compared from one season to another.
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
4. Conclusion The UAV-based sensing technique showed great potential for high-throughput phenomics in wheat breeding field trials. This study used multispectral imaging to assess the emergence and spring stand of soft white, hard red, and soft white club winter wheat varieties. Visual rating and the aerial image-based emergence and stand estimates for individual plots showed strong correlation (r = 0.87). When the replicate plots for each variety was averaged (hard red and soft white), the correlation coefficient was 0.92 and higher. Overall, UAV-based sensing can be a suitable alternative to traditional methods for high-throughput emergence and spring stand quantification for field-based crop phenotyping. Acknowledgements The authors would like to thank Dr. Kimberly A GarlandCampbell from Wheat Genetics, Quality Physiology and Disease Research Unit, USDA-ARS, Pullman for providing her plots (Soft White Club) for evaluating the remote sensing system for winter wheat phenotype characterization in this study. We are also grateful to Ryan Higginbotham and Gary Shelton for maintenance of the trial field location and technical assistance. Funding for this project was provided in part by a Washington State University Emerging Research Issues grant (#3632) and Washington Grain Alliance. In addition, the project was supported by the USDA National Institute of Food and Agriculture National Research Initiative Competitive Grants CAP project 2011-68002-30029, and Hatch Projects 1002864 (WNP00821) and 1005756 (WNP0745). References Carver, B., 2009. Wheat: Science and Trade. Wiley-Blackwell, Ames, Iowa. Deery, D., Jimenez-Berni, J., Jones, H., Sirault, X., Furbank, R., 2014. Proximal remote sensing buggies and potential applications for field-based phenotyping. Agronomy 4 (3), 349–379. Fiorani, F., Rascher, U., Jahnke, S., Schurr, U., 2012. Imaging plants dynamics in heterogenic environments. Curr. Opin. Biotechnol. 23 (2), 227–235. Giri, G.S., Schillinger, W.F., 2003. Seed priming winter wheat for germination, emergence, and yield. Crop Sci. 43 (6), 2135–2141. Gitelson, A.A., 2004. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161 (2), 165–173. Hardin, P.J., Hardin, T.J., 2010. Small-scale remotely piloted vehicles in environmental research. Geogr. Compass 4 (9), 1297–1311. Hasslen, D.A., McCall, J., 1995. Washington Agricultural Statistics 1994–1995. Wash. Agric. Stat. Serv., Olympia. Higginbotham, R.W., Jones, S.S., Carter, A.H., 2011. Adaptability of wheat cultivars to a late-planted no-till fallow production system. Sustainability 3 (8), 1224–1233. Higginbotham, R.W., Jones, S.S., Carter, A.H., 2013. Wheat cultivar performance and stability between no-till and conventional tillage systems in the Pacific Northwest of the United States. Sustainability 5 (3), 882–895.
379
Hunt, E.R., Doraiswamy, P.C., McMurtrey, J.E., Daughtry, C.S., Perry, E.M., Akhmedov, B., 2013. A visible band index for remote sensing leaf chlorophyll content at the canopy scale. Int. J. Appl. Earth Obs. Geoinf. 21, 103–112. Hunt, E.R., Hively, W.D., Fujikawa, S.J., Linden, D.S., Daughtry, C.S., McCarty, G.W., 2010. Acquisition of NIR-green-blue digital photographs from unmanned aircraft for crop monitoring. Remote Sens. 2 (1), 290–305. Juergens, L.A., Young, D.L., Schillinger, W.F., Hinman, H.R., 2004. Economics of alternative no-till spring crop rotations in Washington’s wheat–fallow region. Agron. J. 96 (1), 154–158. Kipp, S., Mistele, B., Baresel, P., Schmidhalter, U., 2014. High-throughput phenotyping early plant vigour of winter wheat. Eur. J. Agron. 52, 271–278. Lelong, C.C., Burger, P., Jubelin, G., Roux, B., Labbé, S., Baret, F., 2008. Assessment of unmanned aerial vehicles imagery for quantitative monitoring of wheat crop in small plots. Sensors 8 (5), 3557–3585. Lindstrom, M.J., Papendick, R.I., Koehler, F.E., 1976. A model to predict winter wheat emergence as affected by soil temperature, water potential, and depth of planting. Agron. J. 68 (1), 137–141. Martínez, E., Fuentes, J.-P., Pino, V., Silva, P., Acevedo, E., 2013. Chemical and biological properties as affected by no-tillage and conventional tillage systems in an irrigated Haploxeroll of Central Chile. Soil Till. Res. 126, 238–245. Patterson, H.D., Williams, E.R., 1976. A new class of resolvable incomplete block designs. Biometrika 63, 83–92. Patterson, H.D., Williams, E.R., Hunter, E.A., 1978. Block designs for variety trials. J. Agric. Sci. 90, 395–400. Sankaran, S., Khot, L.R., Maja, J.M., Ehsani, R., 2013. Comparison of two multiband cameras for use on small UAVs in Agriculture. In: 5th IEEE Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), 25–28 June, 2013, Gainesville, FL. Schillinger, W.F., Paulitz, T., 2014. Natural suppression of Rhizoctonia bare patch in a long-term no-till cropping systems experiment. Plant Dis. 98 (3), 389–394. Schillinger, W.F., Donaldson, E., Allan, R.E., Jones, S.S., 1998. Winter wheat seedling emergence from deep sowing depths. Agron. J. 90 (5), 582–586. Schneider, C.A., Rasband, W.S., Eliceiri, K.W., 2012. NIH Image to ImageJ: 25 years of image analysis. Nat. Methods 9 (7), 671–675. Sharabian, V.R., Noguchi, N., Han-Ya, I., Ishi, K., 2013. Evaluation of an active remote sensor for monitoring winter wheat growth status. Eng. Agric. Environ. Food 6 (3), 118–127. Skinner, D.Z., 2009. Post-acclimation transcriptome adjustment is a major factor in freezing tolerance of winter wheat. Funct. Integr. Genomics 9 (4), 513–523. Skinner, D.Z., Bellinger, B.S., 2010. Exposure to subfreezing temperature and a freeze-thaw cycle affect freezing tolerance of winter wheat in saturated soil. Plant Soil 332 (1–2), 289–297. Snape, J.W., Butterworth, K., Whitechurch, E., Worland, A.J., 2001. Waiting for fine times: genetics of flowering time in wheat. Euphytica 119 (1–2), 185–190. Washington Wheat Commission. 2008. Washington wheat facts. Available at: (accessed 8.08.2014). Winter, S.R., Musick, J.T., Porter, K.B., 1988. Evaluation of screening techniques for breeding drought-resistant winter wheat. Crop Sci. 28 (3), 512–516. Yan, L., 2009. The flowering pathway in wheat. In: Carver, B.F. (Ed.), Wheat Science and Trade. Wiley-Blackwell Ames, Iowa, pp. 57–72. Yau, S.K., 1997. Efficiency of alpha-lattice designs in international yield trials of barley and wheat. J. Agric. Sci. 128, 5–9. Young, F.L., Whaley, D.K., Pan, W.L., Roe, R.D., Alldredge, J., 2014. Introducing winter canola to the winter wheat-fallow region of the Pacific Northwest. Crop Manage. 13 (1). Zhang, C., Kovacs, J.M., 2012. The application of small unmanned aerial systems for precision agriculture: a review. Precision Agric. 13 (6), 693–712. Zhang, X., Friedl, M.A., Schaaf, C.B., Strahler, A.H., Hodges, J.C., Gao, F., Reed, B.C., Huete, A., 2003. Monitoring vegetation phenology using MODIS. Remote Sens. Environ. 84 (3), 471–475.
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Original papers
Field-based crop phenotyping: Multispectral aerial imaging for evaluation of winter wheat emergence and spring stand Sindhuja Sankaran a,⇑, Lav R. Khot a,b, Arron H. Carter c a
Department of Biological Systems Engineering, Washington State University, PO Box 646120, Pullman, WA 99164, United States Center for Precision and Automated Agricultural Systems, Irrigated Agricultural Research and Extension Center, Washington State University, 24106 North Bunn Road, Prosser, WA 99350, United States c Department of Crop and Soil Sciences, Washington State University, PO Box 646420, Pullman, WA 99164, United States b
a r t i c l e
i n f o
Article history: Received 16 September 2014 Received in revised form 28 August 2015 Accepted 5 September 2015
Keywords: Plant breeding Unmanned aerial vehicles Remote sensing Crop growth
a b s t r a c t The physical growing environment of winter wheat can critically be affected by micro-climatic and seasonal changes in a given agroclimatic zone. Therefore, winter wheat breeding efforts across the globe focus heavily on emergence and winter survival, as these traits must first be accomplished before yield potential can be evaluated. In this study, multispectral imaging using unmanned aerial vehicle was investigated for evaluation of seedling emergence and spring stand (an estimate of winter survival) of three winter wheat market classes in Washington State. The studied market classes were soft white club, hard red, and soft white winter wheat varieties. Strong correlation between the ground-truth and aerial image-based emergence (Pearson correlation coefficient, r = 0.87) and spring stand (r = 0.86) estimates was established. Overall, aerial sensing technique can be a useful tool to evaluate emergence and spring stand phenotypic traits. Also, the image database can serve as a virtual record during winter wheat variety development and may be used to evaluate the variety performance over the study years. Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction The United States is one of the largest wheat producing country in the world (Carver, 2009). Within the U.S., Washington State produces both winter and spring wheat in various market classes such as hard red and white, soft white, and club wheat, with a production value of up to $1.18 billion annually. About 80–90% of the wheat produced in Washington is exported to countries such as Japan, Philippines, Taiwan, South Korea and Yemen (Washington Wheat Commission, 2008). Although growers produce both spring and winter wheat varieties in different market classes, approximately 75% of the wheat is planted as winter wheat, and 60% is planted as soft white winter wheat (Washington Wheat Commission, 2008). Winter wheat is the preferred wheat market class because they are high yielding (3.8 MT/ha in comparison to 2.8 MT/ha for spring wheat) and require low farm inputs. Moreover, winter wheat plants establish a soil cover to reduce wind and water erosion during the winter months (Carver, 2009). The specific physiological changes unique to winter wheat are acclimation and vernalization that occur during cold winter months. Cold acclimation allows the plants to survive the winter ⇑ Corresponding author. Tel.: +1 509 335 8828; fax: +1 509 335 2722. E-mail address: [email protected] (S. Sankaran). http://dx.doi.org/10.1016/j.compag.2015.09.001 0168-1699/Ó 2015 Elsevier B.V. All rights reserved.
and vernalization induces reproductive growth, which triggers flowering the following summer (Skinner, 2009; Snape et al., 2001). Vernalization requires growth periods in temperatures less than 4 °C, whereas freeze–thaw cycles leading up to cold winter months enhances cold tolerance in wheat (Skinner and Bellinger, 2010). If the winter hardiness of a plant variety is not sufficient to withstand extremely low temperatures (16 °C or lower), the plants will die. Similarly, if the vernalization is not met completely, it may delay plant maturity due to delayed reproductive growth (Yan, 2009). Different winter wheat varieties will differ in the period needed for vernalization and winter hardiness. Therefore, these factors are critical phenotypic traits that need to be evaluated during new winter wheat variety development. In Washington State, various agroclimatic zones provide variation in the physical growing environment of winter wheat, and thus the traits needed to maintain high yield potential. In the east-central part of Washington, wheat is grown under annual rainfall, which is typically limited to less than 30.5 cm annually (Hasslen and McCall, 1995; Higginbotham et al., 2011, 2013; Martínez et al., 2013). As such, a fallow year is incorporated into the rotation to try and store moisture to maximize crop production (Juergens et al., 2004; Higginbotham et al., 2013). After considerably dry years, wheat must be planted 15–20 cm deep in order to reach moisture and emerge (Schillinger et al., 1998; Schillinger
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
and Paulitz, 2014; Young et al., 2014). During the germination stage, the seedling root and coleoptile (leaf-like structure that protects the first leaf from the soil) develop. Several factors including soil and air temperature, water availability, oxygen, radiation intensity, substrate, maturity of the seeds, and physiological age of the seed affects the germination of winter wheat (Lindstrom et al., 1976; Winter et al., 1988; Giri and Schillinger, 2003). A limitation to emergence can be rain events. If rain occurs before plant emergence, a solid crust forms on the soil surface which further impedes the emergence of wheat seedlings (Schillinger et al., 1998; Young et al., 2014). After successful emergence, plants must then survive the cold winter months. This typically occurs as plants acclimate to the weather and go into dormancy, with a layer of snow cover protecting them from the coldest of months (Skinner, 2009; Young et al., 2014). In years of extremely low temperatures or no snow cover, wheat seedlings become very susceptible to winter kill. Thus, two very important traits for wheat cultivars grown in the east-central regions of Washington are the ability to emerge from deep planting through crust events, and winter survival. Researchers have reported proximal and remote sensing applications using ground and aerial platforms for vegetation dynamics such as leaf area index in different scales (Gitelson, 2004; Zhang et al., 2003; Lelong et al., 2008; Hunt et al., 2010, 2013; Fiorani et al., 2012; Sharabian et al., 2013; Deery et al., 2014). The platforms can range from field tractors, unmanned aerial vehicles (UAVs), or satellites. Although the satellite images can cover a large area, frequency of data collection, legal spatial resolution, and environmental factors including cloud cover can limit image-based crop evaluation (Zhang and Kovacs, 2012). Similarly, ground platforms can be limited by field conditions (especially after irrigation or rainfall) and coverage area at a given time. Such limitations can potentially be addressed by rapidly evolving UAV technology. Low cost UAVs may allow rapid need-based high resolution imaging critical for field phenotyping as number of trials and size of field plots are smaller in such applications. Some of the technological limitations of the UAV platforms for agricultural imaging include limited flight time (associated with battery life), and payload carrying capabilities (Hardin and Hardin, 2010; Zhang and Kovacs, 2012). Digital images can be used for detecting early plant vigor. Recently, 50 winter wheat cultivars were monitored for two years to establish the relationship between early plant vigor index (EPVI was defined as the ratio between the difference of reflectance at 750 and 670 nm, with that of reflectance at 862 nm) and relative amount of green pixels, and a relationship with a regression coefficient of 0.98 was found (Kipp et al., 2014). Although strong relationships have been found to estimate plant vigor, the utilization for field evaluation during plant breeding is relatively new, and not used as a standard method. Therefore, the objective of this study was to evaluate the potential of low altitude remote sensing technology as a high-throughput phenotyping tool for assessment of winter wheat emergence (vigor) and winter survival (winter hardiness) under field conditions.
2. Materials and methods 2.1. Field plots Winter wheat plots were planted into a summer fallow field in Kahlotus, WA on 23rd August, 2013. Plots planted were soft white and hard red winter wheat varieties from the Washington State University (WSU) Wheat Breeding Program, and soft white club wheat varieties from the U.S. Department of AgricultureAgricultural Research Service (USDA-ARS) Wheat Breeding Program. Plots were planted in an alpha-lattice design using a custom
373
built deep-furrow small plot planter consisting of 4 rows. Alphalattice design method divides replicates into incomplete blocks that contain a part of the total number of entries (Patterson and Williams, 1976). Alpha-lattice designs have been shown to be more efficient at analyzing data in large entry plant breeding trials when compared to randomized complete block designs (Yau, 1997). In this study, genotypes were distributed among the blocks in such a way so that all pairs occur in the same block at equal frequencies, thereby permitting removal of incomplete block effects from the plot residuals (Patterson et al., 1978). The seeds were placed at 9 cm soil depth, and plot size was 1.5 m 4.6 m. Each variety was replicated three times, except the club wheat trial that had one to two replicates (Fig. 1). The soil type in Kahlotus is a Ritzville silt loam. Currently grown cultivars in each market class (soft white, hard red, and soft white club) were planted as check cultivars (reference plots), and the remaining plots were experimental breeding lines from the two wheat breeding programs. Base fertilizer in the field was 27.2 kg of N, and all other standard field practices were performed by the cooperating grower. On 25th August, 5.6 mm of rain was recorded followed by high daytime temperatures, which created a solid crust on the soil surface approximately 5 mm thick. The first plants were observed to emerge from the soil on 30th August. Emergence continued for another ten days. On 25th September, plots were rated visually for emergence. Each plot (1.5 m 4.6 m area) was scored on a scale of 0–100% emergence using 10% increments. Plots were again rated visually on 9th April, 2014 to determine winter survival (spring stand). 2.2. Aerial data collection An unmanned aerial vehicle was used to acquire high resolution multispectral images to evaluate the winter wheat emergence (2nd October, 2013) and winter survival (16th April, 2013). The UAV (HiSystems GmbH, Moormerland, Germany) system weighs about 2 kg without the imaging system and a 6000 mA h Lithium Ion Polymer battery that provides the required power for UAV flight. The flight time can range from 10 to 20 min depending on the payload and wind conditions. The maximum recommended payload of this system is 2.5 kg. The UAV has eight brushless motors and individual motors can handle 20 A power with a maximum thrust of 2200 g. Based on preliminary flights, beechwood propellers (Xoar International, CA, USA) were found to improve the stability of the UAV platform during imaging and were used in this study. The UAV comprises of an array of onboard sensors for flight stability and waypoint navigation such as gyroscope, accelerometer, compass, global positioning system receiver, and pressure sensor. A radio transmitter (MX20 Hott, Graupner, Stuttgart, Germany) with range of up to 4 km was used to remotely control the UAV. A modified multispectral digital camera, XNiteCanon SX230 NDVI (LDC LLC, Carlstadt, NJ) with near infrared (670–750 nm) (NIR), green (G), and blue (B) bands was used for aerial imaging. The camera was mounted on a mount underneath the UAV that is capable of automatically adjusting to the nick and roll shifts during flight. Alternately, the nick and roll can be adjusted using the radio control transmitter. A firmware was used to enable georeferenced interval shooting to acquire images every 5 s during UAV flight. The captured 8-bit JPG images with resolution of 12.1 megapixels (4000 3000) were stored on-board the camera card. During preliminary evaluation, flying altitude was adjusted to acquire a multispectral camera image covering the study plots. At this time, a simple digital camera (Sony NEX-5N, Sony Electronics Inc.) was also used to estimate the spatial resolution of the two cameras at different altitudes. Sony NEX-5N is a 16.1 megapixel camera with a 4912 3264 image resolution. The images
374
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
B1
CS-25
CS-26
CS-27
CS-28
CS-29
CS-30
CS-31
CS-32
CS-33
CS-34
CS-35
CS-36
CS-37
CS-38
CS-39
CS-40
CS-41
CS-42
CS-7
CS-8
CS-9
CS-10
CS-11
CS-12
CS-1
CS-2
CS-3
CS-4
CS-5
CS-6
CS-13 WC-28
CS-14 WC-19
CS-15 WC-15
CS-16 WC-4
CS-17 WC-12
CS-18 WC-29
CS-19 WC-3
CS-20 WC-8
CS-21 WC-1
CS-22 WC-30
CS-23 WC-20
CS-24 WC-11
WC-10
WC-2
WC-24
WC-22
WC-9
WC-17
WC-18
WC-26
WC-21
WC-25
WC-7
WC-13
WC-6
WC-23
WC-16
WC-14
WC-27
WC-5
WC-22 DC-8
WC-7 DC-24
WC-21 DC-18
WC-28 DC-29
WC-29 DC-6
WC-24 DC-13
WC-16
WC-11
WC-26
WC-3
WC-19
WC-8
WC-23
WC-17
WC-14
WC-30
WC-18
WC-9
DC-19
DC-21
DC-7
DC-30
DC-2
DC-12
WC-6 DC-15
WC-1 DC-22
WC-5 DC-28
WC-2 DC-17
WC-15 DC-14
WC-27 DC-26
WC-12 DC-1
WC-13 DC-11
WC-25 DC-5
WC-10 DC-25
WC-4 DC-23
WC-20 DC-4
DC-5
DC-25
DC-23
DC-22
DC-1
DC-4
DC-9
DC-29
DC-18
DC-12
DC-8
DC-6
DC-20
DC-27
DC-10
DC-9
DC-3
DC-16
DC-16
DC-17
DC-30
DC-13
DC-28
DC-20
DC-7
DC-2
DC-3
DC-24
DC-14
DC-21
DC-10
DC-19
DC-26
DC-27
DC-11
DC-15
B2 HR-5
B2 HR-26
B2 HR-21
B2 HR-17
B2 HR-35
B2 HR-24
B2 HR-26
B2 HR-36
B2 HR-27
B2 HR-33
B2 HR-34
B2 HR-23
B2 HR-3
B2 HR-19
B2 HR-27
B2 HR-20
B2 HR-29
B2 HR-18
HR-16
HR-15
HR-6
HR-29
HR-34
HR-31
HR-4
HR-25
HR-24
HR-8
HR-29
HR-10
HR-24
HR-16
HR-9
HR-35
HR-33
HR-17
HR-36
HR-11
HR-27
HR-20
HR-12
HR-13
HR-18
HR-11
HR-7
HR-20
HR-14
HR-28
HR-21
HR-13
HR-36
HR-14
HR-28
HR-32
HR-23
HR-22
HR-28
HR-33
HR-19
HR-7
HR-30
HR-17
HR-16
HR-3
HR-21
HR-6
HR-4
HR-26
HR-8
HR-34
HR-7
HR-5
HR-30
HR-18
HR-4
HR-2
HR-9
HR-14
HR-19
HR-22
HR-1
HR-12
HR-9
HR-13
HR-15
HR-1
HR-25
HR-6
HR-2
HR-23
HR-25
HR-10
HR-8
HR-32
HR-1
HR-3
HR-15
HR-2
HR-31
HR-32
HR-5
HR-35
HR-22
HR-30
HR-10
HR-31
HR-12
HR-11
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
SW-4
SW-2
SW-19
SW-9
SW-16
SW-8
SW-32
SW-7
SW-31
SW-30
SW-17
SW-20
SW-4
SW-32
SW-30
SW-33
SW-8
SW-17
SW-23
SW-15
SW-28
SW-32
SW-17
SW-21
SW-27
SW-12
SW-21
SW-10
SW-23
SW-24
SW-10
SW-35
SW-23
SW-36
SW-5
SW-27
SW-20
SW-6
SW-22
SW-1
SW-24
SW-11
SW-36
SW-26
SW-28
SW-33
SW-22
SW-18
SW-22
SW-25
SW-12
SW-24
SW-31
SW-20 SW-13
SW-26
SW-29
SW-12
SW-27
SW-14
SW-25
SW-16
SW-13
SW-35
SW-11
SW-5
SW-15
SW-11
SW-9
SW-18
SW-3
SW-26
SW-3
SW-31
SW-10
SW-33
SW-7
SW-34
SW-25
SW-9
SW-29
SW-4
SW-6
SW-34
SW-28
SW-29
SW-16
SW-21
SW-19
SW-2
SW-5
SW-13
SW-36
SW-35
SW-30
SW-18
SW-2
SW-8
SW-1
SW-19
SW-14
SW-3
SW-7
SW-15
SW-1
SW-6
SW-14
SW-34
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
B4
Fig. 1. Field map showing distribution of replicate plots in the experiment. The B1, B2 and B3 plots were soft white club wheat Bruehl, B4 plots were soft white common wheat Otto, SW1–SW36 refers to 36 soft wheat winter wheat varieties, HR1–HR36 refers to 36 hard red winter wheat varieties, DC1–DC20 refers 30 ‘dry’ soft white club, WC1–WC20 refers to 30 ‘wet’ soft white club, and CS refers to single entries of different soft white club varieties.
(Red–Green–Blue [RGB] bands) were acquired from 100 m, 75 m, and 55 m altitude. Fig. 2a displays the false color multispectral image acquired from 100 m altitude showing representative plots. The RGB images had spatial resolution of about 2.3 cm/pixel, 2.0 cm/pixel, and 1.4 cm/pixel at 100 m, 75 m and 55 m, respectively. Similarly, for respective altitudes, the corresponding spatial resolution was 3.0 cm/pixel, 2.3 cm/pixel, and 1.6 cm/pixel in multispectral images. Using the above information and methodology, a total of 378 plots (450 plots with border rows) were imaged using the multispectral camera from 100 m altitude for emergence and spring stand evaluation. The image recorded three wheat market classes, soft white club (162 plots), hard red (108 plots), and soft white (108 plots) planted in-between the border rows. The border rows were soft white club wheat Bruehl and soft white common wheat Otto (Fig. 1). 2.3. Image processing and analysis Data analysis was performed to estimate the emergence or spring stand after winter survival from individual trial plots in the imaged study area. Feature extraction and analysis algorithms were developed using ImageJ (Schneider et al., 2012) and MatlabÒ (v. R2009b, The MathWorks, Inc. MA). The feature extraction steps involved: (a) separating an image into three bands, (b) extracting green normalized difference vegetation index (GNDVI) image, (c) thresholding of the GNDVI image to remove background, and (d) convert the GNDVI into binary image. The GNDVI (Sankaran et al., 2013) was extracted using Eq. (1), where NIR is the reflectance at near infrared band and G is the reflectance at green band:
GNDVI ¼
NIR G NIR þ G
ð1Þ
The GNDVI images were further processed to remove the soil and background noise using threshold. Resulting images were converted to binary image and the average pixel values for individual
plots were calculated as the crop emergence or spring stand (Eq. (2)), where N represents number of pixels in individual plot, and GNDVIi refers to GNDVI pixel values of individual pixel.
X¼
N 1X GNDVIi N i¼1
ð2Þ
The binary image pixel value was 0 for soil and 255 for plant region. The same equation was used to calculate emergence and spring stand, each representing two different plant traits. The difference in crop trait estimation is based on different time periods that the images were acquired. Winter wheat emergence was estimated from images acquired in the Fall (about one month after planting), and spring stand was estimated from images acquired in the Spring (about 1 month after plants have broken dormancy requirements and begin to resume normal growth, assuming they survived the winter and were not killed). The image results were correlated with ground-truth measurements (as described in Section 2.1) taken from the same plots. 2.4. Seasonal micro-climatic variation Preliminary analysis of air temperature data from a nearby weather station (AgWeatherNet, Hatton, WA) for the study period (interval between two aerial data collections) indicated that for 112 days, the average air temperature, recorded at 1.5 m above ground surface, was <0 °C. Although most winter wheat varieties can survive temperatures below 0 °C, temperatures below 10 °C can start to differentiate varieties and their ability to survive winter temperatures. On 4th December 2013, minimum daytime temperatures decreased rapidly, and the following eight day period ranged from 13 to 20 °C daytime minimum temperature, and averaged 15 °C. Similarly, the soil temperature at 20 cm depth was <0 °C for 105 days during the same time interval, and averaged 3 °C during this eight day long cold period. Another cold period during the first week of February had minimum air temperatures averaging 11 °C for seven days, with maximum temperatures only averaging 8 °C.
375
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
Fig. 2. (a) UAV-based false color (NIR, G, B as RGB) images of representative winter wheat field plots showing emergence, and (b) corresponding pseudocolor GNDVI image extracted from multispectral images. The colors represents different GNDVI scales, commonly used for better visualization of GNDVI image. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
During the time between aerial imaging, no snow cover was present on the wheat seedlings, and only 7.6 cm of total precipitation was reported. Considering the two weeks of low temperatures, the low precipitation and drought conditions, along with no snow cover, a low winter survival rate could be anticipated. 2.5. Statistical analysis Statistical analysis of percent emergence and percent spring stand was analyzed using SAS version 9.3 (SAS Institute, Raleigh, NC). The ‘Proc GLM’ procedure was used to test the replication, genotype, and interaction effect for each of the three market class trials separately. Pearson’s correlations analysis was conducted using ‘Proc CORR’ procedure in SAS. 3. Results and discussion 3.1. Winter wheat emergence and stand Germination and emergence rates depend on the wheat variety. Under favorable conditions, seedling emergence can occur within
seven days. However, deep planting or rain events can cause a delay and reduction in seedling emergence. Standard methods of evaluating this phenotype involve visual rating of individual plots. Aerial image-based quantification technique developed in this study can be an alternative method to evaluate the wheat plant emergence phenotype under field conditions. Binary image extracted from processed GNDVI image (Fig. 2b) after thresholding. The white area corresponds to emergence or spring stand of wheat plants and the black area corresponds to non-emergence. The correlation of plant emergence and the spring stand with ground-truth data for individual wheat varieties (plotto-plot comparison in each variety class) are summarized in Table 1 Results of correlation (P < 0.0001) between UAV image-based plant emergence (or spring stand) and ground-truth data for winter wheat varieties planted in Kahlotus, WA. Wheat variety
Average emergence (%)
r
Average spring stand (%)
r
Soft white club Hard red Soft white
45.9 59.1 64.7
0.82 0.93 0.86
29.5 51.7 47.1
0.84 0.88 0.83
376
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
Table 1. ANOVA indicated that there were significant differences (P < 0.0001) between genotypes (breeding lines) in all three market class trials for both percent emergence and percent spring stand. There was no replication effect (significant differences between the replicate plots) for percent spring stand in any of the market class trials or for the percent emergence in the club wheat trial. There was a significant effect (P < 0.0001) of replication for percent emergence in both the soft white and hard red market class trials. This significant effect was found to be a scalar interaction, with replicate one (the front of the trial), two, and three having an average emergence of 53%, 65%, and 68%, respectively. Relative rankings of each genotype within each replicate did not change and there was no significant genotype replicate interaction found. A good correlation between the aerial image-based winter emergence (%) and ground-truth was observed with Pearson correlation coefficient (r) ranging from 0.82 to 0.93 (Table 1). Plot-toplot comparisons combining all three market classes indicated the r for emergence was 0.87 (P < 0.0001). Similarly, for plant spring stand after the winter season, the overall r between the aerial image data and ground-truth was 0.86 (P < 0.0001). Based on the results presented in Table 1, a high correlation between ground-truth and UAV multispectral image data was observed. Comparison of average emergence and spring stand rates indicated that the wheat varieties in all three market classes were affected by winter weather. It should again be noted here that the varieties studied in this experiment were unreleased breeding lines, and variation in both emergence and winter survival were expected to be present between the lines. Amongst the varieties tested, a drastic reduction in spring stand rates was observed with the soft white club varieties, whereas the hard red wheat varieties were found to be more winter hardy as a whole. Furthermore, for determining the plant growth relationship, the average emergence/stand rates for each cultivar in the three
market classes were computed by averaging the ground-truth data from replicate plots. Similarly, the emergence and spring stand data estimated using aerial images for each variety were averaged, and compared with ground-truth ratings (Fig. 3). A high correlation (>0.92) was observed between the estimates and ground-truth data for the soft white and hard red varieties. For the soft white club varieties, the correlation coefficient was between 0.86 and 0.91 for the ‘dry’ and ‘wet’ soft white club varieties. Comparing all the average ground-truth ratings (averaging the replicates) of all varieties from three market classes (hard red, soft white, and soft club white) with that of corresponding image data, the r was 0.91 for emergence and 0.92 during spring stand evaluation (P < 0.0001). 3.2. Differences between emergence and winter survival The winter wheat varieties were further analyzed to determine the relationship between the ground-truth emergence/stand rates of individual varieties (36 varieties 3 replicates for soft white and hard red each; 30 varieties 2 replicates for ‘dry’ and ‘wet’ soft white club each) with that of image data. After analysis, four distinct types of breeding lines were observed in the trials. These were the lines with (1) good emergence and spring stands, (2) good emergence and poor spring stands, (3) poor emergence and good spring stands, and (4) poor emergence and poor spring stands. For example, the three check (reference) cultivars (Bruehl, Xerpha, and Otto) used in the breeding program for soft white wheat emerged very well during the fall, with average emergence of 87%, 90%, and 100%, respectively. Spring stands of these varieties were 47%, 53%, and 87%, respectively, indicating the ability of the cultivar Otto to survive the winter better than Bruehl or Xerpha. A similar trend was observed in the hard wheat trial, with the check (reference) cultivars (Bauermeister, Finley, and Farnum)
160
120 100 80 60 40 20 0
Emergence, r = 0.97 Spring Stand, r = 0.93
140 120 100 80 60 40 20 0
(a) Soft white
0 160 Image Data (pixel value)
Image Data (pixel value)
Emergence, r = 0.93 Spring Stand, r = 0.92
140
20
40 60 Visual Rating (%)
80
(b) Hard Red
100 160
Emergence, r = 0.91 Spring Stand, r = 0.89
140
0
Image Data (pixel value)
Image Data (pixel value)
160
120 100 80 60 40 20 0
20
20
40 60 Visual Rating (%)
80
100
80
100
Emergence, r = 0.86 Spring Stand, r = 0.91
140 120 100 80 60 40 20 0
(d) 'Wet' soft white club
(c) 'Dry' soft white club
0
40 60 Visual Rating (%)
0
20
40 60 Visual Rating (%)
80
100
Fig. 3. Correlations (P < 0.0001) between average ground-truth emergence/spring stand and image data for hard red, soft white, and soft white club (dry and wet) winter wheat varieties (averaging the replicate plots for each cultivar within a market class).
377
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
having good fall emergence ratings (80%, 97% and 100%, respectively), yet spring stands (50%, 83%, and 90%, respectively) indicate Finley and Farnum have better winter tolerance than Bauermeister. Overall, there were 17% of the plots which showed good emergence and good winter survival (as indicated by a >70% stand), and these will be useful to advance in the breeding programs for further evaluation for cultivar release. On average, 23% of lines had good emergence in the fall, but poor spring stands, indicating these lines are not tolerant to cold winter temperatures. There were 45% of the lines which had poor emergence and poor spring stands, although no observations can be made about their ability to withstand winter conditions as few plants were present in the fall. About 14% of the lines showed poor emergence characteristics, but had an increase in the stand percentage in the spring, indicating these lines survived the winter well, but were also able to take advantage of the available space and moisture to produce more tillers per plant, giving the indication of a greater spring stand then was present at emergence. The varieties with higher than 25% change in loss of vigor after winter and ones showing increased vigor (>10% difference) after winter are summarized in Table 2. Lines with a positive number indicate lines which experienced some winter kill, whereas those with negative numbers are the lines which demonstrated an increase in the percent stand observed from fall to spring. Lines with negative numbers will be useful in breeding programs to identify lines with increased tillering and overall competitiveness, but which will need increased
emergence characteristics. The overall performance of these varieties can clearly be observed in Fig. 4. Visual emergence and spring stand ranking, which can overlay plot images have been shown in Fig. 4. Continuous blue rows represent the border rows, with the first and last border rows not shown. As seen from Fig. 4a and b, the soft white wheat breeding lines had better emergence potential than the hard red before winter (>55%), whereas the hard red wheat breeding lines tended to withstand winter better than the soft white between the tested varieties. The WSU winter wheat breeding program has a long history of breeding soft white winter wheat cultivars for excellent emergence potential. Thus, it is expected that the germplasm in this market class have better emergence than the hard red or club wheat market classes. Observing the individual breeding lines within each market class, expected trends can be detected. In the soft wheat market class, there were more lines which had above average emergence, but only a few listed in the 80–100% range and only a few in the 0–20% range. Contrast to that, in the hard red wheat market class, there were many more lines in the 80– 100% emergence range, and many in the 0–20% range, with few in between. Thus, there was a trend to a lower overall emergence average in the hard red market class. The club wheat program had the most lines with poor emergence, which is to be expected as this program has a smaller germplasm pool for good emerging lines and has not been screening breeding material for emergence as long as the soft white and hard red programs have been.
Table 2 Emergence (October 2013) and spring stand (April 2014) percentages of selected winter wheat breeding lines grown in Kahlotus, WA. Plot number*
Type
Variety (Trail No.y)
352, 289, 299, 273, 278, 276, 280, 291, 334, 359, 286, 347,
Soft club winter wheat program
27, 64, 77, 57, 52, 25, 51, 46, 49, 75, 58, 28, 24,
391 321 325 310 311 301 326 323 362 367 302 390
33, 76 67, 78 88, 102 87, 98 96, 99 54, 110 61, 113 109, 121 81, 93 83, 85 97, 108 53, 111 69, 84
162, 149, 178, 164, 147, 155, 196, 169, 158, 153, 194, 157, 199,
200, 206 171, 176 192, 219 205, 252 195, 226 198, 235 221,245 185, 248 163, 187 160, 222 236, 241 170, 182 213, 242
Emergence rating (%)
Spring stand rating (%)
Difference = emergence stand (%)
BRUEHL (WC-3) XERPHA (DC-5) X010746-4C (DC-8) HS980158-1L (DC-30) ARS-Selbu (DC-2) X000175-1-1 (DC-20) X070070-0-0-16L (DC-24) X070051-0-0-47C (DC-23) ARS-Chrystal (WC-2) HS00293-2C-1 (WC-18) X06135-9C (DC-27) X20060126-0-3C (WC-29)
85.0 70.0 65.0 80.0 70.0 95.0 95.0 65.0 70.0 80.0 5.0 20.0
60.0 45.0 40.0 50.0 40.0 50.0 30.0 0.0 45.0 25.0 20.0 35.0
25 25 25 30 30 45 65 65 25 55 15 15
Soft white winter wheat program
Eltan (SW-1) WA 8199 (SW-11) 3J080408-0-15 (SW-24) 5J062268-1 (SW-12) 3J080209-0-1 (SW-21) Xerpha (SW-2) WA 8201 (SW-16) Bruehl (SW-4) SSDRES193-3-SW (SW-28) 3J080221-0-3 (SW-22) 3J080565-0-1 (SW-27) WA 8202 (SW-19) SSD760097-5-R (SW-18)
100.0 90.0 76.7 70.0 70.0 90.0 76.7 86.7 83.3 70.0 83.3 50.0 36.7
76.7 66.7 50.0 36.7 33.3 53.3 40.0 46.7 40.0 20.0 16.7 63.3 60.0
23 23 27 33 37 37 37 40 43 50 67 13 23
Hard red winter wheat program
WA 8180 (HR-11) Bauermeister (HR-1) Sprinter (HR-6) WA 8208 (HR-18) WA008156 (HR-8) NORWEST 553 (HR-5) FarEd277T (HR-34) 3J080150-0-1 (HR-19) 3J080125-0-T1 (HR-30) 3J080147-0-T1 (HR-31) FarEd176 (HR-26) FarEd32 (HR-22) W7YSSD100001 (HR-36)
96.7 80.0 93.3 80.0 80.0 76.7 53.3 23.3 20.0 40.0 30.0 30.0 30.0
66.7 50.0 60.0 43.3 16.7 3.3 66.7 40.0 36.7 56.7 46.7 53.3 53.3
30 30 33 37 63 73 13 17 17 17 17 23 23
* The plots were replicated three times and numbered from left to right and bottom to top. The plots 1–18, 127–144, 253–270, 433–450 were border rows. The plots 19– 126, 145–252, 271–330, and 331–396 were soft white, hard red, ‘dry’ soft white club, and ‘wet’ soft white club breeding lines, respectively. y Trail No. is referenced to Fig. 1.
378
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
100 2
90
4 80
6
90 2 80 4 6
70
8
60
70
8 10
60
12
50
12
14
40
14
30
16
30
18
20
16 18
20 20
10
50 40
20 10
22
10 22
0 5
10
0
15
5
(a) Emergence
10
15
(b) Spring Stand
2
80
4 60
6 8
40
10 12
20 14 16
0
18 -20
20 22
-40 5
10
15
(c) Emergence -Spring Stand Fig. 4. Visual rating of (a) emergence, (b) winter wheat spring stand, and (c) spring stand rates subtracted from emergence rates. Each box represents individual plots. 18 plots from left to right, and 23 plots rows from top to bottom.
The spring stand image (Fig. 4b) shows similar results. The club wheat program had the most lines with poor spring stands, in part due to winter kill, but in part due to an initial poor emergence. Although the soft white program had overall the best emergence potential of the lines tested, the increase in the number of plots with a poor spring stand indicates that more of these plants died over the winter, and thus had less winter survival as a market class than did the hard red varieties. The hard red varieties showed an overall improvement in plant stand over the winter months. Although some lines which emerged well experienced some winter kill, there were other lines which had an average emergence of 40– 60% which had an increase in spring stand percentage. The spring stand was calculated based on the percent greenness observed, not the individual plants present. So, although the number of plants in the hard red plots may have stayed the same, they were covering more space. This indicates that the tillering capacity of the hard red cultivars was above average, and the plants which did emerge and survive the winter were able to produce more tillers, effectively taking advantage of the available space and increasing the number of wheat heads per square foot. Comparing the winter wheat emergence and spring stand percentages (Fig. 4c), the ones seen in red and orange are varieties that had good emergence rates, but did not survive the winter.
Similarly, the ones marked in blue are the varieties that did not have high emergence rates, but had a good spring stand after winter. Again, the lines in blue indicate lines which, although few plants emerged, were able to take advantage of the space around them and produce more tillers per square foot. This is an important trait for wheat breeders to select on, as tillering and competitiveness are beneficial traits in wheat production where annual rainfall is limited. The ability of breeders to visualize this phenotype is enhanced through the aerial images provided by this study. Overall, the study successfully evaluated emergence and spring stand of winter wheat using UAV-based images. Both are critical phenotypes during winter wheat variety development. Existing standard methods involves visual rating from 0% to 100%, with each plot given a rating in increments of 10%. Such evaluation may be biased and subject to human error. The results from this study confirm that aerial sensing techniques are reliable (with significant correlations), and at the same time comparable to visual evaluation. Given the high correlation achieved in this study, such sensing technique tools can be useful to evaluate the phenotypes of emergence and spring stand. Moreover, the images can also serve as a visual record during winter wheat variety development and can be compared from one season to another.
S. Sankaran et al. / Computers and Electronics in Agriculture 118 (2015) 372–379
4. Conclusion The UAV-based sensing technique showed great potential for high-throughput phenomics in wheat breeding field trials. This study used multispectral imaging to assess the emergence and spring stand of soft white, hard red, and soft white club winter wheat varieties. Visual rating and the aerial image-based emergence and stand estimates for individual plots showed strong correlation (r = 0.87). When the replicate plots for each variety was averaged (hard red and soft white), the correlation coefficient was 0.92 and higher. Overall, UAV-based sensing can be a suitable alternative to traditional methods for high-throughput emergence and spring stand quantification for field-based crop phenotyping. Acknowledgements The authors would like to thank Dr. Kimberly A GarlandCampbell from Wheat Genetics, Quality Physiology and Disease Research Unit, USDA-ARS, Pullman for providing her plots (Soft White Club) for evaluating the remote sensing system for winter wheat phenotype characterization in this study. We are also grateful to Ryan Higginbotham and Gary Shelton for maintenance of the trial field location and technical assistance. Funding for this project was provided in part by a Washington State University Emerging Research Issues grant (#3632) and Washington Grain Alliance. In addition, the project was supported by the USDA National Institute of Food and Agriculture National Research Initiative Competitive Grants CAP project 2011-68002-30029, and Hatch Projects 1002864 (WNP00821) and 1005756 (WNP0745). References Carver, B., 2009. Wheat: Science and Trade. Wiley-Blackwell, Ames, Iowa. Deery, D., Jimenez-Berni, J., Jones, H., Sirault, X., Furbank, R., 2014. Proximal remote sensing buggies and potential applications for field-based phenotyping. Agronomy 4 (3), 349–379. Fiorani, F., Rascher, U., Jahnke, S., Schurr, U., 2012. Imaging plants dynamics in heterogenic environments. Curr. Opin. Biotechnol. 23 (2), 227–235. Giri, G.S., Schillinger, W.F., 2003. Seed priming winter wheat for germination, emergence, and yield. Crop Sci. 43 (6), 2135–2141. Gitelson, A.A., 2004. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161 (2), 165–173. Hardin, P.J., Hardin, T.J., 2010. Small-scale remotely piloted vehicles in environmental research. Geogr. Compass 4 (9), 1297–1311. Hasslen, D.A., McCall, J., 1995. Washington Agricultural Statistics 1994–1995. Wash. Agric. Stat. Serv., Olympia. Higginbotham, R.W., Jones, S.S., Carter, A.H., 2011. Adaptability of wheat cultivars to a late-planted no-till fallow production system. Sustainability 3 (8), 1224–1233. Higginbotham, R.W., Jones, S.S., Carter, A.H., 2013. Wheat cultivar performance and stability between no-till and conventional tillage systems in the Pacific Northwest of the United States. Sustainability 5 (3), 882–895.
379
Hunt, E.R., Doraiswamy, P.C., McMurtrey, J.E., Daughtry, C.S., Perry, E.M., Akhmedov, B., 2013. A visible band index for remote sensing leaf chlorophyll content at the canopy scale. Int. J. Appl. Earth Obs. Geoinf. 21, 103–112. Hunt, E.R., Hively, W.D., Fujikawa, S.J., Linden, D.S., Daughtry, C.S., McCarty, G.W., 2010. Acquisition of NIR-green-blue digital photographs from unmanned aircraft for crop monitoring. Remote Sens. 2 (1), 290–305. Juergens, L.A., Young, D.L., Schillinger, W.F., Hinman, H.R., 2004. Economics of alternative no-till spring crop rotations in Washington’s wheat–fallow region. Agron. J. 96 (1), 154–158. Kipp, S., Mistele, B., Baresel, P., Schmidhalter, U., 2014. High-throughput phenotyping early plant vigour of winter wheat. Eur. J. Agron. 52, 271–278. Lelong, C.C., Burger, P., Jubelin, G., Roux, B., Labbé, S., Baret, F., 2008. Assessment of unmanned aerial vehicles imagery for quantitative monitoring of wheat crop in small plots. Sensors 8 (5), 3557–3585. Lindstrom, M.J., Papendick, R.I., Koehler, F.E., 1976. A model to predict winter wheat emergence as affected by soil temperature, water potential, and depth of planting. Agron. J. 68 (1), 137–141. Martínez, E., Fuentes, J.-P., Pino, V., Silva, P., Acevedo, E., 2013. Chemical and biological properties as affected by no-tillage and conventional tillage systems in an irrigated Haploxeroll of Central Chile. Soil Till. Res. 126, 238–245. Patterson, H.D., Williams, E.R., 1976. A new class of resolvable incomplete block designs. Biometrika 63, 83–92. Patterson, H.D., Williams, E.R., Hunter, E.A., 1978. Block designs for variety trials. J. Agric. Sci. 90, 395–400. Sankaran, S., Khot, L.R., Maja, J.M., Ehsani, R., 2013. Comparison of two multiband cameras for use on small UAVs in Agriculture. In: 5th IEEE Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), 25–28 June, 2013, Gainesville, FL. Schillinger, W.F., Paulitz, T., 2014. Natural suppression of Rhizoctonia bare patch in a long-term no-till cropping systems experiment. Plant Dis. 98 (3), 389–394. Schillinger, W.F., Donaldson, E., Allan, R.E., Jones, S.S., 1998. Winter wheat seedling emergence from deep sowing depths. Agron. J. 90 (5), 582–586. Schneider, C.A., Rasband, W.S., Eliceiri, K.W., 2012. NIH Image to ImageJ: 25 years of image analysis. Nat. Methods 9 (7), 671–675. Sharabian, V.R., Noguchi, N., Han-Ya, I., Ishi, K., 2013. Evaluation of an active remote sensor for monitoring winter wheat growth status. Eng. Agric. Environ. Food 6 (3), 118–127. Skinner, D.Z., 2009. Post-acclimation transcriptome adjustment is a major factor in freezing tolerance of winter wheat. Funct. Integr. Genomics 9 (4), 513–523. Skinner, D.Z., Bellinger, B.S., 2010. Exposure to subfreezing temperature and a freeze-thaw cycle affect freezing tolerance of winter wheat in saturated soil. Plant Soil 332 (1–2), 289–297. Snape, J.W., Butterworth, K., Whitechurch, E., Worland, A.J., 2001. Waiting for fine times: genetics of flowering time in wheat. Euphytica 119 (1–2), 185–190. Washington Wheat Commission. 2008. Washington wheat facts. Available at: