Traitement en cours

Veuillez attendre...

Paramétrages

Paramétrages

Aller à Demande

1. WO2020112013 - PROCÉDÉS, MODÈLES ET SYSTÈMES DE PRÉDICTION DE LA ROUILLE JAUNE DANS DES CULTURES DE BLÉ

Note: Texte fondé sur des processus automatiques de reconnaissance optique de caractères. Seule la version PDF a une valeur juridique

[ EN ]

METHODS, MODELS AND SYSTEMS FOR PREDICTING YELLOW RUST IN WHEAT CROPS

FIELD OF THE INVENTION

The present invention relates to the field of methods, models and systems for predicting yellow rust in wheat crops. These methods, models and systems rely on preferably Random Forest modelling of hyperspectral reflectance data and related disease presence data from wheat crops and provide robust new tools for efficiently and accurately predicting the presence of Yellow Rust .

BACKGROUND OF THE INVENTION

Yellow Rust is a foliar disease which affects bread wheat

( Triticum aestivum L.) crops, thus significantly reducing grain yields. Yellow Rust is caused by Puccinia striiformis f. sp tritici (Pst), and is one of the most devastating fungal diseases of wheat in major parts of the world. In Sweden, yellow rust has become increasing common in the southern and central parts of the country, and has therefore become a major concern.

Yellow Rust can be detected manually, by the farmer in the field; however this method of detection is time consuming and requires that that the farmer is suitable trained. It is therefore also a task which cannot be left to untrained

personnel .

Some wheat plant varieties, and other accessions, may, due to their genetic make-up, have varying degrees of immunity to Yellow Rust infection. A major purpose of wheat plant breeding is thus to screen and select for wheat plant varieties and individuals that are immune to plant diseases such as Yellow Rust .

In order to automate or otherwise make the detection of Yellow Rust more efficient, hyperspectral phenotyping has been tried. This involves photographing or otherwise obtaining hyperspectral images or spectra from light reflected off the canopies of wheat using cameras and spectroradiometers , sometimes mounted on vehicles such as unmanned aerial vehicles (UAVs) to obtain data on wheat crops growing in the field.

It is also known to calculate vegetation indices (Vis) from the reflectance at individual wavelengths in obtained hyperspectral data. One example of such a vegetation index is the NDVI

(Normalized Difference Vegetation Index) index defined by (Rsoo - R68O) / (Rsoo + R68O) ) wherein Rx is the reflectance at wavelength x nm. At least 140 such vegetation indices have been proposed with the aim to being able to determine properties of vegetation remotely based on hyperspectral images or spectra.

Despite these methods there is still a need for methods, models and systems in which this data is associated with, and capable of, predicting the presence of Yellow Rust with sufficient accuracy, robustness and efficiency to render them useful for deployment in the field. There is in particular a need for methods, models and systems capable of predicting the presence of Yellow Rust using a minimum of sensor and computing

resources .

Accordingly, objects of the present invention include the provision of methods, models and systems capable of more accurately predicting the presence of Yellow Rust in wheat crops .

Further objects of the present invention include the provision of methods, models and systems capable of robustly predicting the presence of Yellow Rust in wheat crops .

Still further objects of the present invention include the provision of methods, models and systems capable of more efficiently predicting the presence of Yellow Rust in wheat crops .

SUMMARY OF THE INVENTION

At least one of the above mentioned objects are, according to the first aspect of the present invention achieved by a method of constructing a model, preferably a Random Forest model, for predicting the presence of Yellow Rust in wheat crops,

comprising the steps of:

i. obtaining, for a plurality of sample wheat plots, a

plurality of values for first, second and third vegetation indices VI, V2, V3, wherein

a. the first vegetation index VI is defined by

Vl= ( R531+10nm - R570±10nm) / ( R531±10nm + R570±10nm) , b. the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 is defined by V4= ( R800+10nm - R670+10nm) / ( RsOO+lOnm + R670±10nm) and c. the third vegetation index V3 is defined by

V3= ( R734+10nm - R747±10nm) / ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

ii. obtaining, from the plurality of sample wheat plots, a plurality of Yellow Rust scores, the score for each sample wheat plot specifying the presence of Yellow Rust in that sample wheat plot, and

iii. constructing or calculating a model associating the

plurality of values for the vegetation indices VI, V2 and V3 with the plurality of Yellow Rust scores.

Thus the present invention is based on the discovery that hyperspectral reflectance data of wheat crops can accurately and robustly predict the presence of Yellow Rust when represented by a set of at least three vegetation indices VI, V2 , V3, these three vegetation indices being related to the three vegetation indices PRI, PRI norm and Vogelmann2 as also explained in the example section. As no more than the three vegetation indices VI, V2, V3 are needed the method is also efficient.

The model constructed by the method is as explained above a valuable tool for predicting Yellow Rust in wheat.

The model is preferably a Random Forest model, however it is contemplated within the context of the present invention that other models could be used, such as in particular a gradient boosting model.

When the model is a Random Forest model it may typically comprise 500-5000, such as 1500-3000, such as 2000 trees. The number of tree iteration may be 500-5000, such as 1500-3000, such as 2000.

Predicting is to be understood as encompassing providing a likelihood or probability for the presence of Yellow Rust.

Predicting may further encompass determining.

The presence of Yellow Rust may be represented qualitatively (as yes/no) or quantitatively (on a scale ranging from no or only an insignificant amount of Yellow Rust being present, such as only a few plants being infected, to a significant amount of Yellow rust being present, such as all wheat plants being infected) .

The presence of Yellow rust may alternatively or additionally be represented as a percentage based on the number of wheat plants that are infected by Yellow Rust vs . the total number of wheat plants in a wheat plot. The presence of Yellow Rust may further be represented quantitatively with regard to the severity of the infection, such as by the extent of infection or damage in a wheat plant or wheat plot. The presence of Yellow Rust may be predicted for a single wheat plant or for a plurality of wheat plants in a wheat plot.

Wheat comprises species of the genus Triticum.

The obtaining of the values for the first, second and third vegetation indices VI, V2 , V3 encompasses obtaining measurements of light reflected off the wheat canopy. The measurements may be obtained by a sensor which may be a hyperspectral sensor, i.e. capable of measuring reflected light at each wavelength over a range of wavelengths, such as 400-1000nm, producing a spectra. The measurements may alternatively be performed by a sensor that is only capable of measuring reflected light over a narrower region of the spectra, or only capable of measuring reflected light at a discrete number of individual wavelengths or a discrete number of individual wavelength ranges.

The obtaining of the values for the first, second and third vegetation indices VI, V2 , V3 further encompasses calculating the vegetation indices based on the measured reflectance at the specified wavelengths according to:

a. the first vegetation index VI being defined by

Vl= ( R531+10nm - R570±10nm) / ( R531±10nm + R570±10nm) , b. the second vegetation index V2 being defined by

V2=V1* (-1) / (V4*R700+I0nm /R67o+ionm) , wherein V4 being defined by V4= ( Rsoo+ionm - R670±10nm) /V (Rsoo+ionm + R670+10nm) and

c. the third vegetation index V3 being defined by

V3= ( R734+10nm - R747±10nm) / ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

Each reflectance value Rx is determined at a single wavelength. Thus, for example, R53i±ionm corresponds to a Reflectance value obtained at a single wavelength wherein X is an integer selected from the range 520-541 nm.

The reflectance may for example be measured using light sensor or a spectroradiometer . The light sensor or spectroradiometer may be positioned 0.5 to 2 m or higher, such as 1 meter, above the wheat canopy. The light sensor or spectroradiometer may be handheld, mounted on a vehicle, or suspended on or by a flying vehicle such as an unmanned aerial vehicle, UAV.

Preferably a white reference measurement is performed prior to each measurement. This minimizes the effect of different lighting condition (sunshine, cloudy, overcast, rain) that otherwise could affect the reflectance measurements by for measurement first measuring the reflectance of a white object or surface .

A sample wheat plot may be any area in which wheat is grown. A sample wheat plot may be of rectangular dimension with a circumference of 2 to 25 m, or larger or smaller. The dimensions of the sample wheat plot preferably correspond to the field of view of the sensor used to obtain the measurements of the reflectance, however the sample wheat plot may be larger in which case the sensor may take several readings .

The plurality of yellow rust scores may be obtained manually by visually inspecting the wheat plants in each sample wheat plot. Alternatively the plurality of yellow rust scores may be obtained automatically such as by sampling a plurality of wheat plants in the plurality of sample wheat plots and determining whether the plurality of wheat plants comprise DNA from Puccinia striiformis f. sp tritici (Pst), the fungi causing Yellow Rust.

The yellow rust scores may specify the presence qualitatively, (as yes/no) or quantitatively (on a scale ranging from no or only an insignificant amount of Yellow Rust being present, such as only a few plants being infected, to a significant amount of Yellow rust being present, such as all wheat plants being infected) . The presence of Yellow rust may alternatively or additionally be represented as a percentage based on the number of wheat plants that are infected by Yellow Rust vs. the total number of wheat plants in a wheat plot. The Yellow Rust scores may further be represented quantitatively with regard to the severity of the infection, such as by the extent of infection or damage in a wheat plant or wheat plot.

The model associates the plurality of values for the vegetation indices VI V2 , V3 with the plurality of Yellow Rust scores.

Alternatively the model may be constructed or calculated using the plurality of the values for the vegetation indices VI V2 , V3 and the plurality of Yellow Rust scores for a selection of the sample wheat plots as a training data set, whereby the

construction or calculation further comprises evaluating the model so constructed on values for the vegetation indices VI V2, V3 and the Yellow Rust scores for the non-selected sample wheat plots .

It is further contemplated within the context of the present invention that a selection of the three vegetation indices VI, V2, V3, the selection for example being any of VI and V2 , V2 and V3, or V3 and VI, can be used in the methods and systems of the present invention instead of all three vegetation indices .

In the preferred embodiment of the method according to the first aspect of the present invention the first vegetation index VI is PRI, Photochemical Reflectance Index, the second vegetation index V2 is PRI norm, Renormalized Difference Vegetation Index, and the third vegetation index V3 is Vogelmann2, Vogelmann indices 2.

Here, the vegetation indices are defined as follows:

a. PRI = (R531 - R570) / (R531 + R570) ,

b. PRI_norm = PRI* ( -1 ) / (RDVI*R7oo /R67O) , wherein RDVI = (Rsoo - R67O) /V ( Rsoo + R67O) , and

c. Vogelmann2 = (R734 - R747) / (R715 + R726)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

As shown in the examples these specific vegetation indices provide an overall accuracy of 0.54 for predicting three different scores of presence of Yellow Rust.

In the preferred embodiment of the method according to the first aspect of the present invention no other vegetation indices than the vegetation indices VI, V2 and V3 are used in step iii for constructing or calculating the model.

Preferably also no other vegetation indices than the vegetation indices VI, V2 and V3 are obtained in step i.

As surprisingly found and described in example 1 the addition of further vegetation indices did not improve the accuracy of the model. By limiting the number of indices to the vegetation indices VI, V2 and V3 the sensor and computation requirements are minimized, i.e. a simpler sensor only capable of detecting reflected light in the wavelength regions needed for calculating the vegetation indices VI, V2 , V3 can be used, and the

computations needed to construct the model are minimized.

In the preferred embodiment of the method according to the first aspect of the present invention the method further comprises the step of normalizing the plurality of values for the first, second and third vegetation index VI, V2, and V3 prior to constructing the model in step iii.

The normalizing comprises normalizing the plurality of values for the first, second and third vegetation indices VI, V2, and V3 relative to the values obtained for a selection of the plurality of the sample wheat plots .

In the preferred embodiment of the method according to the first aspect of the present invention the wheat is bread wheat

( Triticum aestivum L.) .

An alternative first aspect of the present invention pertains to the use of the three vegetation indices VI, V2, V3, wherein:

- the first vegetation index VI is defined by Vl= ( R53i±ionm

- R570+10nm ) / (R531+10nm + R570±10nm) ,

- the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 is defined by V4= ( RsoOilOnm - R670±10nm) / ( RsOO+lOnm + R670±10nm) and

- the third vegetation index V3 is defined by V3= ( R734±ionm

- R747+10nm) / ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

for constructing a model, preferably a Random forest model, for predicting the presence of Yellow Rust in wheat crops, by associating values for the vegetation indices VI, V2 and V3 for a plurality of sample wheat plots with Yellow Rust scores for the plurality of sample wheat plots .

The features of use according to the alternative first aspect of the present invention are as given for the method according to the first aspect.

At least one of the above mentioned objects are, according to a second aspect of the present invention, which second aspect corresponds to the first aspect, further achieved by a method of predicting the presence of Yellow Rust in a wheat plant or wheat plot, comprising the steps of:

i. Obtaining, for the wheat plant or the wheat plot values for first, second and third vegetation indices VI, V2, V3, wherein

a. the first vegetation index VI is defined by

Vl= ( R531+10nm - R570±10nm) / ( R531±10nm + R570±10nm,

b. the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R700+I0nm /R67o+ionm), wherein V4 is defined by

V4= ( RsoOilOnm - R670±10nm) / ( RsOO+lOnm + R670±10nm) and c. the third vegetation index V3 is defined by

V3= ( R734+10nm - R747±10nm) / ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm, and

ii. predicting the presence of Yellow Rust in the wheat plant or wheat plot by subjecting the values for the first, second and third vegetation indices VI, V2 , V3 to the model obtained by the method according the first aspect of the present invention or obtained by the use according to the alternative first aspect of the present invention.

The model used in step (ii) of predicting the presence of Yellow Rust is obtained by the method according the first aspect of the present invention or obtained by the use according to the alternative first aspect of the present invention.

Accordingly, the method according the second aspect of the present invention may, as a step, comprise performing the method according to the first aspect of the present invention.

The method according to the second aspect of the present invention may thus be formulated as a method of predicting the presence of Yellow Rust in a wheat plant or wheat plot,

comprising the steps of:

i. constructing a model, preferably a Random Forest model, for predicting the presence of Yellow Rust in wheat crops, comprising the substeps of:

a. obtaining, for a plurality of sample wheat plots, a plurality of values for first, second and third vegetation indices VI, V2 , V3m wherein

i. the first vegetation index VI is defined by

Vl= ( R531+10nm - R570±10nm)/ ( R531±10nm + R570±10nm), ii. the second vegetation index V2 is defined by V2=V1* (-1) / (V4 *R7oo+ionm /R670+I0nm) , wherein V4 is defined by V4= ( Rsoo+ionm - R67o±ionm) /V (Rsoo+ionm +

R67o+ionm) and

iii. the third vegetation index V3 is defined by

V3= ( R734+10nm - R747±10nm)/ ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

b. obtaining, from the plurality of sample wheat plots, a plurality of Yellow Rust scores, the score for each sample wheat plot specifying the presence of Yellow rust in that sample wheat plot, and

c. constructing or calculating a model associating the plurality of values for the vegetation indices VI, V2 and V3 with the plurality of Yellow rust scores, ii. obtaining, for the wheat plant or the wheat plot values for first, second and third vegetation indices VI, V2, V3, wherein

a. the first vegetation index VI is defined by

Vl= ( R531+10nm - R570±10nm)/ ( R531±10nm + R570±10nm,

b. the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 is defined by

V4= ( RsoOilOnm - R670±10nm) / ( RsOO+lOnm + R670±10nm) and c. the third vegetation index V3 is defined by

V3= ( R734+10nm - R747±10nm)/ ( R715±10nm + R726±10nm) wherein Rx is the wheat canopy reflectance at the wavelength X nm, and

iii. predicting the presence of Yellow Rust in the wheat plant or wheat plot by subjecting the values for the first, second and third vegetation indices VI, V2 , V3 to the model obtained in step i .

Preferably no other vegetation indices than the vegetation indices VI, V2 and V3 are used in substep c for constructing or calculating the model, and preferably no other vegetation indices than the vegetation indices VI, V2 and V3 are obtained in substep a.

Preferably the method further comprises the step of normalizing the plurality of values for the first, second and third

vegetation indices VI, V2 , and V3 prior to constructing the model in substep c.

The features of the method according to the second aspect of the present invention are as given for the method according to the first aspect, with the following additions.

The method provides for predicting the presence of Yellow Rust in a wheat plant or wheat plot. The wheat plot may have similar dimensions as the sample wheat plot.

In the preferred embodiment of the method according to the second aspect of the present invention the the first, second and third vegetation indices VI, V2, V3 are obtained using at least one sensor, the at least one sensor being capable of detecting light in the wavelength range of 520-810 nm, the at least one sensor preferably only being capable of detecting light in the wavelength ranges needed to determine the first, second and third vegetation indices VI, V2 and V3.

As the calculation of the vegetation indices used in the method does not require measuring reflectance below 520 nm the sensor can be made simpler and less expensive, and simpler and less expensive sensors can be used. The sensor can be made even simpler if it is only capable of detecting light, i.e. measuring the reflectance off the wheat canopy, is the wavelength ranges or even specific wave lengths, needed to determine the

vegetation indices VI, V2 , V3.

An alternative second aspect of the present invention pertains to the use of the three vegetation indices VI, V2, V3, wherein: - the first vegetation index VI is defined by Vl= ( R53i±ionm

- R570+10nm) / ( R531±10nm + R570±10nm)

- the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 is defined by V4= ( RsotVlOnm _ R670±10nm) / ( RsOO+lOnm + R670±10nm) and

- the third vegetation index V3 is defined by V3= ( R734±ionm

- R747+10nm) / ( R715±10nm + R726±10nm)

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

for predicting the presence of Yellow Rust in wheat crops, by subjecting values for the vegetation indices VI, V2 and V3 to a model associating values for the vegetation indices VI, V2 and V3 for a plurality of sample wheat plots with Yellow Rust scores for the plurality of sample wheat plots .

The features of use according to the alternative second aspect of the present invention are as given for the method according to the second aspect.

At least one of the above mentioned objects is, according to a third aspect of the present invention, further achieved by a method of breeding wheat, comprising the steps of

i. predicting the presence of Yellow Rust in a plurality of wheat plants or wheat plots using the method according the second aspect of the present invention or by the use according to the alternative second aspect to the present invention,

ii. selecting at least one wheat plant or wheat plot predicted to be free of Yellow Rust, and

iii. further breeding the at least one wheat plant or wheat

plot .

The features of the method according to the third aspect of the present invention are as given for the method according to the first aspect, with the following additions.

The method of breeding wheat may be for obtaining wheat

varieties having partial or complete immunity against Yellow Rust .

The selecting may comprise any of; moving the at least one wheat plant or the plants in the wheat plot predicted to be free of Yellow Rust to a new location; taking seeds or other genetic material from at least one wheat plant or the plants in the wheat plot predicted to be free of Yellow Rust; and removing, terminating or sterilizing the wheat plants not predicted of being free of Yellow Rust.

further breeding the at least one wheat plant or wheat plot may comprise further any of: breeding the at least one wheat plant or the plants in the wheat plot predicted to be free of Yellow Rust, planting seeds or genetically modifying a wheat plant using the other genetic material taken from at least one wheat plant or the plants in the wheat plot predicted to be free of Yellow Rust.

The method according to the third aspect of the present

invention may alternatively be expressed as a use.

At least one of the above mentioned objects are, according to a fourth aspect of the present invention corresponding to the second aspect, further achieved by a system for predicting the presence of Yellow Rust in wheat crops, the system comprising:

- at least one sensor module capable of measuring reflected light from a wheat canopy in the wavelength range of 520-810 nm,

- a computation module configured to provide first, second and third vegetation indices VI, V2, V3, wherein

the first vegetation index VI is defined by Vl= ( R53i±ionm

- R570+10nm) / ( R531±10nm + R570±10nm,

the second vegetation index V2 is defined by V2=V1* (- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 is defined by V4= ( R800+10nm _ R670±10nm) / ( RsOO+lOnm + R670±10nm), and the third vegetation index V3 is defined by V3= ( R734±ionm

- R747+10nm) / ( R715±10nm + R726±10nm) ,

wherein Rx is the wheat canopy reflectance at the wavelength X nm,

based on the reflected light measured by the at least one sensor module,

- a model module comprising a model, preferably a Random Forest model, associating a plurality of values for the vegetation indices VI, V2 and V3 for a plurality of sample wheat plots with a plurality of Yellow Rust scores for the plurality of sample wheat plots,

Wherein the model module and/or the computation module is further configured to subject the first, second and third vegetation indices VI, V2 , V3 to the model to obtain a

prediction of the presence of yellow rust.

The features of the system according to the fourth aspect of the present invention are as given for the method according to the second aspect, with the following additions.

The system may be comprised by a single apparatus that may be handheld or mounted on a vehicle, including tracked and wheeled vehicles as well as flying vehicles. Alternatively the system

may be implemented in a distributed fashion wherein at least one of the sensor module, computation module, and model module is separated from the others. In one example the sensor module and optionally the computation module may be implemented as a single apparatus that may be handheld or mounted on a vehicle as described above, while the model module, and optionally the computation module, are implemented as a separate apparatus, such as a separate computer or server, which receives the measurements of reflected light measured by the at least one sensor module, and optionally receives the vegetation indices computed by the computation module, wirelessly or by other way of electronic communication, and thereafter obtains and provides the prediction of the presence of Yellow Rust.

The at least one sensor module is preferably a sensor module capable of only measuring reflected light in the wavelengths needed to obtain the three vegetation indices VI, V2 , V3.

The computation module may be a general purpose microcomputer or a processor. The model module may comprise a memory or database for storing the model.

At least one of the above mentioned objects are, according to a fifth aspect of the present invention corresponding to the second aspect, further achieved by a computer program product for being used in the system according to the fourth aspect of the present invention, wherein the computer program product comprises program code instructions configured to, when executed by the computation module and/or model module of the system, cause the computation module and/or model module to perform the method according to the second aspect of the present invention.

BRIEF DESCRIPTION OF THE FIGURES AND DETAILED DESCRIPTION

A more complete understanding of the abovementioned and other features and advantages of the present invention will be apparent from the following detailed description of preferred embodiments in conjunction with the appended drawings, wherein:

Fig . 1 shows the different canopy reflectance spectra (400nm to 1000 nm) obtained from wheat at the different Yellow Rust scores 1 to 9,

Fig. 2 shows a histogram of the scoring for Yellow Rust for the wheat plots,

Fig . 3 shows a heatmap for data from different years for the three vegetation indices Vogelmann2, Pri and PRI norm for the wheat plots for the three different rescored Yellow Rust classes (yr=0, 1, 2), in which lighter lines indicate a high level of reflectance for that vegetation index, whereas a dark color indicates a low level of reflectance for that vegetation index,

Fig. 4A-C show boxplots for the respective Vegetation indices

Vogelmann2, PR and PRI norm showing the intensity of this vegetation index for the different rescored Yellow Rust classes (x-axis = 0, 1, 2),

Fig. 4D shows the pair-wise correlation between the three

Vegetation indices the for all wheat plot samples,

Fig. 5A-C show flow diagrams of embodiments of the methods

according to the first, second and third aspects of the present invention, and

Fig. 6 shows an embodiment of the system according to the fourth aspect of the present invention.

In the below description of the figures the same reference numerals are used to designate the same features throughout the figures. Further, a 'added to a reference numeral indicates that the feature is a variant of the feature designated with the corresponding reference numeral not carrying the '-sign.

EXAMPLE 1 - OBTAINING OF HYPERSPECTRAL REFLECTANCE DATA,

DETERMINATION OF VEGETATION INDICES, SCORING FOR YELLOW RUST,

AND CONSTRUCTION OF MODEL

Materials and methods

A light reflectance spectrum of winter wheat canopies was recorded at the milk development stage (Zadoks 71-77) over the wavelength range of 300-1100 nm using a spectrometer (Apogee Instruments, Inc., United States) . The spectrometer was held one meter above the wheat canopy. The spectrometer was calibrated using a white reference standard Apogee AS-004 (Apogee

Instruments, Inc., United States) after every 4-5 measurements. Measurements were done in a field in Skane, Sweden, on which winter wheat was grown, the field being divided into 367 wheat plots (each a square of 1 m by 1 m) , under two seasons. The measurements were performed around noon in the month of June.

A manual scoring for yellow rust disease was performed for each plot by a professional plant breeder and the result of the manual scoring was defined as a score on the scale of 0-9 (9 as susceptible by Yellow rust) during the two seasons.

Fig. 1 shows the different canopy reflectance spectra (400nm to 1000 nm) obtained from wheat at the different Yellow Rust scores 1 to 9. As can be seen in the figure Yellow Rust infection causes progressively diminishing reflectance in the higher wavelengths, although the reflectance between two adjacent

scores may in some cases increase even though the scoring for Yellow Rust increases. This phenomena, and the need for high bandwidth/high resolution sensors to obtain these canopy reflectance spectra underscore the objects of the invention and the need for more efficient methods of predicting Yellow Rust as discussed above and as presented herein.

Fig. 2 shows a histogram of the scoring for Yellow Rust for the wheat plots. As seen in the figure the majority of the scored wheat plots were found in the scores 2-4 although the sampled plots contained examples of all scores 1-9.

Data processing

Pre-processing of spectrum data was done to remove outliers and technical noise. Thereafter, areas around the edges of the spectra were cropped based on low signal to noise ratio.

Subsequently 119 previously known vegetative indices (Vis) were estimated using the software Specalyzer (Koc A, Henriksson T, Chawade A (2018) Specalyzer—an interactive online tool to analyze spectral reflectance measurements. PeerJ 6.

doi:10.7717/peerj .5031.

Outlier samples were removed by inspection of principal

component analysis plots obtained from the 119 Vis of all samples .

The 119 vegetation indices used are listed in table 1 below:

Table 1 : list of used vegetation indices

Number. Abbreviation Name

1. ARI Anthocyanin Reflectance Index

2. ARI2 Anthocyanin Reflectance index 2

3. BGI Blue Green Pigment Index

4. SB703/Boochs Single Band 703 Boochs

5. SB720/Boochs2 Single Band 720 Boochs 2

6. BRI Browning Reflectance Index

8. CARI Chlorophyll Absorption Ratio Index

9. Ctr Carter

10. Ctr2 Carter 2

11. Ctr3 Carter 3

12. Ctr4 Carter 4

13. Ctr5 Carter 5

14. Ctr6 Carter 6

15. Cl Coloration Index

16. CI2 Coloration Index 2

17. ClAInt

18. CRI Carotenoid Reflectance Index

19. CRI 1 Carotenoid Reflectance Index 1

20. CRI2 Carotenoid Reflectance Index 2

21. CRI 3 Carotenoid Reflectance Index 3

22. CRI 4 Carotenoid Reflectance Index 4

23. D1 Derivative index

24. D2 Derivative index

25. Datt Datt

26. Datt2 Datt 2

27. Datt3 Datt 3

28. Datt4 Datt 4

29. Datt5 Datt 5

30. Datt6 Datt 6

31. Datt7 Datt 7

32. Datt8 Datt 8

33. DD Double Difference Index

34. DDn New Double Difference Index

35. DPI Double Peak Index

36. DWSI1 Disease water stress index 1

37. DWSI2 Disease water stress index 2

38. DWSI3 Disease water stress index 3

39. DWSI4 Disease water stress index 4

40. DWSI5 Disease water stress index 5

41. EGFN Edge green first derivative normalized difference

42. EGFR Edge green first derivative ratio

43. EVI Enhanced Vegetation Index

44. GDVI2 Green Difference Vegetation Index 2

45. GDVI 3 Green Difference Vegetation Index 3

46. GDVI 4 Green Difference Vegetation Index 4

47. GI Greenness Index

48. - Gitelson

49. - Gitelson 2

50. GMI 1 Gitelson and Merzlyak Index 1

51. GMI2 Gitelson and Merzlyak Index 2

52. Green NDVI Green Normalized Difference Vegetation Index

53. GVI Greenness Vegetation Indx

54. LIC Lichtenthaler indices

55. LRDSI1 Leaf Rust Disease Severity Index 1

56. LRDSI2 Leaf Rust Disease Severity Index 2

57. LWVI 1 Normalized Difference 1094/983 Leaf water VI 1

58. LWVI2 Normalized Difference 1094/1205 Leaf water VI 2

59. - Maccioni

60. Modified Chlorophyll Absorption in Reflectance Index

61. Modified Chlorophyll Absorption in Reflectance Index 1

62. Modified Chlorophyll Absorption in Reflectance Index 2

63. MCARI2/OSAVI2

64. MCARI2/OSAVI

65. Modified NDVI

66. Modified NDVI 2

67. Modified Photochemical Reflectance Index

68. Modified Red-Edge Inflection Point

69. Modified Soil Adjusted Vegetation Index

70. Moisture Stress Index

71. modified Simple Ratio

72. modified Simple Ratio 2

73. modified Simple Ratio 3

74. modified Simple Ratio 705

75. MERIS Terrestrial Chlorophyll Index

76. modified Triangular Vegetation Index

77. Modified Vegetation Stress Ratio

78. Narrow-Band Normalised Difference Vegetation Index

79.
Normalized Difference Lignin Index

80. NDNI Normalized Difference Nitrogen Index

81. NDVI Normalized Difference Vegetation Index

82. NDVI2 Normalized Difference Vegetation Index 2

83. NDVI 3 Normalized Difference Vegetation Index 3

84. NPCI Normalized Pigment Chlorophyll Index

85. NPQI Normalized Difference 415/435 Normalized

Phaeophytinization Index

86. NRI Nitrogen Reflectance Index

87. OSAVI Optimized Soil Adjusted Vegetation Index

88. 0SAVI2 Optimized Soil Adjusted Vegetation Index 2

89. RARS Ratio Analysis of Reflectance Spectra

90. PhRI Physiological Reflectance Index

91. PRI Photochemical Reflectance Index

92. PRI2 Photochemical Reflectance Index 2

93. PRI*CI2 PRI*CI2

94. PRI_norm normalized PRI

95. PSND Pigment specific normalized difference

96. PSRI Plant Senescence Reflectance Index

97. PSSR Pigment specific simple ratio

98. PWI Plant Water Index

99. RDVI Renormalized Difference Vegetation Index

100. REP Red-Edge Position

101. REP_LE

102. REP_Li

103. RGI Red/Green Index

104. SAVI Soil Adjusted Vegetation Index

105. SIPI Structure Intensive Pigment Index

106. SIPI2 Structure Intensive Pigment Index 2

107. SI PI 3 Structure Intensive Pigment Index 3

108. SPVI Spectral Polygon Vegetation Index

109. SR Simple Ratio

110. SRI Simple Ratio 1

111. SR2 Simple Ratio 2

112. SR3 Simple Ratio 3

113. SR4 Simple Ratio 4

114. SR5 Simple Ratio 5

115. SR6 Simple Ratio 6

116. SR7 Simple Ratio 7

117. SR705 Simple Ratio 705

118. SR8 Simple Ratio

119. SR9 Simple Ratio 9

The mathematical definition of each of the above listed

vegetation indices is given in Supplementary file 1 to Koc A, Henriksson T, Chawade A (2018) Specalyzer—an interactive online tool to analyze spectral reflectance measurements. PeerJ 6.

doi:10.7717/peerj .5031.

Correction for the developmental stages and time over two years was done by using 10 plots as checks. For each measurement plot, VI measurements were normalized to these 10 checks for each season by taking the ratio of the VI measurement of each plot to the check. Normalized Vis were thereafter used for training and testing the prediction models for yellow rust.

Model training

The 367 wheat plots were split into a training and a test set with 215 wheat plots in the training set and 152 in the test set .

The scoring results from the manual scoring was converted to 0,1 and 2 such that all original values between 0-3 received a score of 0, from 4-6 received 1 and 7-9 received 2.

The Machine learning method RandomForest (RF) was thereafter used to identify the most predictive Vis for identifying yellow rust disease symptoms based on the converted scores of 0,1 and 2. The R package varselRF was used for model building. The parameters to build a model were "ntree (number of trees) 2000; ntreelterat (number of tree iteration) 2000; vars . drop . frac (...) =0.2" .

In total, 101 models were trained on the training set.

Thereafter, Vis were identified as highly predictive based on their importance in each of the 101 models. Three Vis namely

(PRI, PRI norm, Vogelmann2), were identified by all 101 models as highly predictive, and therefore, these three Vis were thus exclusively used to build the final model. The final model was thereafter tested on the test set and an overall accuracy of 0.54 was obtained for predicting all three classes (0,1,2) and individual class accuracies were 0.65, 0.56 and 0.66

respectively.

Including additional Vis selected in the 101 models did not further improve the accuracy of the final model.

Fig. 3 shows a heatmap for data from different years for the three vegetation indices Vogelmann2, Pri and PRI norm for the wheat plots for the three different rescored Yellow Rust classes (Yr=0 (darkest), 1, 2 (lightest)), see the bar at the top. The lighter colored lines indicate that a high level of reflectance was obtained for that vegetation index, whereas a dark colored line indicates a low level of reflectance for that vegetation index. The lines in the heat map show one of four values: -1 (darkest, i.e. black, lines), -0.5 (lighter than -1), 0.5

(lighter than -0.5, and 1 (the lightest) .

Fig. 4A-C shows boxplots for the three Vegetation indices showing the intensity of the respective vegetation index for the different rescored Yellow Rust classes (x-axis = 0, 1, 2) .

Together the three vegetation indices are capable of an overall

accuracy of 0.54 for predicting all three classes (0, 1, 2) and individual class accuracies of 0.65, 0.56 and 0.66 respectively.

Fig. 4D shows the pair-wise correlation between the three

Vegetation indices the for all wheat plot samples . As can be seen in the figure there is a higher degree of correlation between PRI and PRI norm than for the correlation of either of these with Vogelmann2.

From the three vegetation indices PRI, PRI norm, and Vogelmann2 three corresponding but generalized vegetation indices VI, V2 and V3 were defined as follows:

- the first vegetation index VI was defined by

Vl= ( R531+10nm _ R570±10nm)/ ( R531±10nm + R570±10nm),

- the second vegetation index V2 was defined by V2=V1*(- 1 ) / (V4*R7oo+ionm /R67o+ionm), wherein V4 was defined by V4= ( RsotVlOnm _ R670+10nm) / ( RsOO+lOnm + R670±10nm) and

- the third vegetation index V3 was defined by

V3= ( R734+10nm _ R747±10nm)/ ( R715±10nm + R726±10nm) ·

The generalized vegetation indices where then used in the methods and systems according to the various aspects of the present invention.

Fig. 5A-C show flow diagrams of embodiments of the methods according to the first, second and third aspects of the present invention .

Thus fig. 5A shows a flow diagram of an embodiment of the method according to the first aspect of the present invention including the first step 100 of obtaining a plurality of values for the vegetation indices VI, V2 , and V3 for a plurality of sample wheat plots, the second step 200 of obtaining a plurality of Yellow Rust scores for the plurality of sample wheat plots, and the thirds step of constructing a model which associates the plurality of values for the vegetation indices VI, V2, V3 with the plurality of Yellow Rust scores for the sample wheat plots .

Fig. 5B shows a flow diagram of an embodiment of the method according to the second aspect of the present invention

including the first step 400 of obtaining a plurality of values for the vegetation indices VI, V2 , and V3 for a wheat plant or wheat plot, and the second step 500 of predicting the presence of Yellow Rust the wheat plant or wheat plot by subjecting 502 the vegetation indices VI, V2, V3 to a model which associates a plurality of values for the vegetation indices VI, V2, V3 with the plurality of Yellow Rust scores for a plurality of sample wheat plots.

Finally fig. 5C shows a flow diagram of an embodiment of the method according to the third aspect of the present invention including the first step 600 of predicting the presence of Yellow Rust in a plurality of wheat plants or wheat plots using the method according to the second aspect of the present invention, followed by the second step 700 of selecting wheat plants that is predicted to be free from Yellow Rust, followed by the final step 800 of further breeding the selected wheat plants .

Fig. 6 shows an embodiment of the system according to the fourth aspect of the present invention. Fig. 6 thus shows a field 2 in which wheat is grown. The field is divided into a number of wheat plots, one being designated the reference numeral 4. The system 10 for predicting the presence of Yellow Rust comprises a sensor module 12 which is hand held or positioned on a vehicle.

The vertical distance of the sensor module 12 is adapted so that the field of view, indicated by the dashed lines, one of which is designated the reference numeral 14, matches the dimensions of a wheat plot. The sensor modules output is a measurement of the reflectance of the canopies of wheat growing in the wheat plot 4 that is within the field of view 14 of the sensor module 12. A computation module 16 is connected to the sensor module 12 via a connection 18, which connection may be by wire or

wireless. The computation module 16 is configured to determine and provide the three vegetation indices VI, V2 and V3 based on the measurements of reflected light provided by the sensor module 12. The computation module 16 may be separate from the sensor module 12, as shown here, or may alternatively be integrated with the sensor module 12. The vegetation indices, once determined for the wheat plot 4, are then provided to the model module 20 via connection 22. Connection 22 may be wired or wireless. Typically the model module 20 is provided separate from the sensor module 12, such as for example being provided as a central computer or server capable of processing vegetation indices provided by a plurality of sensor modules 12 and computation modules 16. In the model module 20 the vegetation indices VI, V2, V3 are subjected to a model which associated a plurality of values for the vegetation indices VI, V2 , V3 with a plurality of Yellow Rust scores from a plurality of sample wheat plots, i.e. wheat plots where the presence of Yellow Rust has been scored manually or automatically, and the result of

subjecting the vegetation indices to the model is a prediction of the presence of Yellow Rust in the sample wheat plot 4. After obtaining the prediction of the presence of Yellow Rust for wheat plot 4 the sensor module 12 may be moved to another wheat plot to obtain a prediction of the presence of Yellow Rust in that other wheat plot .

Feasible modifications of the Invention

The invention is not limited only to the embodiments described above and shown in the drawings, which primarily have an illustrative and exemplifying purpose. This patent application is intended to cover all adjustments and variants of the preferred embodiments described herein, thus the present invention is defined by the wording of the appended claims and the equivalents thereof.