Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive
Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive

Full Research Article

Assessment of Chemical and Biological Measures of Controlling Rice Sheath Blight: Field Trials and ML-based Yield Prediction

K
Kumar Avinash Biswal1
0009-0006-6431-1491
S
Siddhartha Das1,*
0000-0002-7482-2555
N
Nirakar Ranasingh2
0009-0007-5255-2914
R
Rajeeb Lochan Moharana3
0009-0002-2294-3977
S
Srujani Behera2
0000-0001-7764-9631
C
Chandana Behera4
0009-0000-2455-6108
S
Simly Das5
0009-0009-8565-5738
1Department of Plant Pathology, Centurion University of Technology and Management, Paralakhemundi-761 211, Odisha, India.
2Department of Plant Pathology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
3Department of Seed Science and Technology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
4Department of Plant Breeding and Genetics College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
5Department of Nematology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.

ABSTRACT

Background: Rhizoctonia solani is the cause of sheath blight of rice that is a significant limitation to Kharif productivity, requiring effective and unstable management approaches.

Methods: A field test, using eight treatments, was carried out: T1-Control, T2-Validamycin 3% L, T3-Azotrix (Azoxystrobin 16.7% + Tricyclazole 33.3%), T4-Nativo (Tebuconazole 50% + Trifloxystrobin 25% WG), T5-Tilt, T6-Contaf Plus, T7-Trichoderma and T8-Pseudomonas. The severity of the disease (DS%), PDI, yield, number of grains per panicle and weight of 1000 grains were noted. ANOVA was used to analyze data (Tukey, DMRT) and prediction of yields was carried out using machine-learning models (Random forest, LightGBM, CatBoost, XGBoost).

Result: The effects of treatment were very substantial (p<0.001) and the effect of year and the interaction of year × treatment were not significant (p = 0.608). Nativo (T4) performed better, with maximum yield (11.62 kg plot-1), grains per panicle (131.22) and 1000 gm seed weight (23.94 g) and minimum DS (18.47-20.25%) and PDI (16.50-19.06%). Azotrix and Validamycin followed. The maximum DS (40.57%) and PDI (52.38) were observed with minimum yield (7.49 kg plot-1). Random Forest got the best accuracy (R2 = 0.848; RMSE = 0.566), which validated nativo, azotric and validamycin as the most suitable management choices.

INTRODUCTION

Rhizoctonia solani AG1-1A (Kuhn) is a soil-borne fungus causing sheath blight, whose teleomorph is Thanatephorus cucumeris (Uppala and Zhou, 2018; Hussein et al., 2025). Disease yield loss about 10-30% and up to 50% given good conditions in the environment (Uppala and Zhou, 2018; Biswal and Das, 2024). It is tightly linked with the rise of semi-dwarf and nitrogen-sensitive varieties of rice in the world (Liu et al., 2025). The pathogen has right-angle branching hyphae and generates white mycellium and brown sclerotia (1-3 mm) on the infected tissues (Uppala and Zhou, 2018). The onset of disease is at late tillering stage of rice, early internode elongation showing water-soaked lesions and greenish-grey color center in the sheath. resulting in senescence, tiller lodging. (Uppala and Zhou, 2018; IRRI Rice Knowledge Bank, 2019). The mode of action used by Trichoderma spp. is the release of antifungal enzymes such as chitinase and glucanase causing the degradation of hypha and sclerotia, whereas Aspergillus has also been reported to exert mycoparasitic relationships with R. solani (Abbas et al., 2022).
       
The recent progress of deep learning has greatly enhanced the system of sheath blight detection, severity and decision support system in managing rice disease (Rezaei et al., 2025). The models are quite effective in reflecting the intricate non-linear integration between the severity of the disease, environmental conditions, crop developmental phases and farming practices (El Amraoui et al., 2024). There are recurrent architectures like LSTM and BiLSTM, which obtain temporal patterns of disease progression depending on the weather conditions, nitrogen content, canopy density and pathogen dynamics (Jeong et al., 2023; Li et al., 2025). Deep learning in combination with biological knowledge can be used to detect the disease in its early stages and can provide optimal and sustainable strategies of sheath blight management (Paquet et al., 2022; Sushmit et al., 2023; Singh et al., 2019).

MATERIALS AND METHODS

Experimental site and experimental design
 
The experiment was conducted rabi season of 2022-2024 at Ranadevi farm of Centurion University of Technology and Management, Parlakhemundi, Odisha. The experiment was conducted in Swarna (MTU 7029) variety of rice. The land was ploughed twice crosswise by tractor drawn harrow and weeds were removed thoroughly from the field. Nursery bed was raised and sowing was done (Fig 1). Main field was prepared in an area of (13.6 × 21.7) m2, along with 0.5 m irrigation channel. Thirty days aged seedlings were transplanted into the main field in three replications with a spacing (4.2 × 3.2) m2. Plant to plant and the row-to-row distance were 10 × 20 cm. 30 days old seedlings were transferred from the nursery to the main field.

Fig 1: Raising of seedlings in nursery bed of rice preparation under field conditions (A), and Experiment to manage sheath blight (Rhizoctonia solani) using rice nursery of variety MTU 7029 (B).


 
Symptomatic characterization
 
The symptoms of sheath blight were studied and systematically monitored in the field as well as microscopic study and photographed. The symptoms first appeared: small, water soaked, oval-irregular spots on the waterline on the leaf sheath which gradually increased in size with greyish centers and brown margins, ascending and descending the sheath and the leaf blades. Lesions formed large necrotic spots as the disease progressed and in extreme cases sheaths and leaves were heavily affected resulting in wilting and lodging (Fig 2). Microscopic studies were also able to confirm the morphological character of Rhizoctonia solani (Fig 3).

Fig 2: Development of symptoms of rice sheath blight under field conditions.



Fig 3: Microscopic image of Rhizoctonia solani distinct right-angle branching and septation of hyphae (A), Sclerotium of R. solani in stereoscopic microscope (B) and Rhizoctonia solani on potato dextrose agar (PDA) (C).


 
Experimental design and treatments
 
The experiment was laid out in a randomized complete block design (RCBD) with eight treatments and three replications per year. The field was artificially inoculated with R. solani (Fig 4,5). The treatments comprised, T1- Control, T2- Validamycin 3% L, T3- Azotrix (Azoxystrobin 16.7% + Tricyclazole 33.3%), T4- Nativo (Tebuconazole 50% + Trifloxystrobin 25% WG), T5- Tilt (Propiconazole 25% EC), T6- Contaf plus (Hexaconazole 5% SC), T7- Trichoderma harzianum, T8- Pseudomonas fluorescens. Foliar application was done at maximum tillering and panicle initiation stage of crop.

Fig 4: Field experiments Intercultural operations in transplanted rice under field conditions (A), Fungicidal and biological treatment of rice field made in knapsack sprayer (B) and Field observation of the experimental plots of rice at the time of maturity with treatment-wise crop response (C).



Fig 5: Inoculation of rice plants by artificial inoculation of Rhizoctonia solani (A) and Oblique view of reproduced irrigated rice plots using drones (B).


 
Data sources
 
DS and PDI were recorded at Ranadevi farm of Centurion University, data is taken in 10 days interval after inoculation up to 9 observations. The disease was scored by using 0-9 scale based on lesion character and lesion area, where 0 represented no lesion characteristics and 9 represented lesions in more than 75% leaf area. DS and PDI was calculated by formula given by Shrestha and Mishra (1994).



 
 Measurement of yield and yield attribute
 
Plant height was assessed at maturity from ground level to the tip of the tallest panicle on 10 randomly selected plants per plot using a measuring tape. Ten representative panicles were selected randomly in each plot and manual counting of grains was done. Weighed clean and sun-dried seeds with 1000-grain weight (g) of the adjustment to 14% of moisture. The harvested net plot yield was dried and adjusted to 14% moisture content and recorded as grain yield (Kg plot-1).
 
Statistical analysis
 
Over the course of three years, a mixed analysis of variance (ANOVA) was performed based on the described linear model. The means of treatment were isolated with the help of the Tukey’s honest significant difference (HSD) test and the Dunkan’s multiple range test (DMRT) at p<0.05. Statistical analysis was done in R software with the help of agricolae package (Das et al., 2019).
 
Machine learning analysis
 
The regression analysis was conducted using the pooled data of 2022-2024 cropping seasons (n = 72) to forecast the rice grain yield under sheath blight pressure. Before developing the model, the numerical variables had been normalized via Z-score transformation and the categorical ones transformed into codes and the dataset was splitting into training (80) and testing (20) sets. The analysis used a constant random state (42) so that it can be reproducible and comparable across models.
       
Four machine-learning models were tested on predicting the grain yield under different sheath blight pressure with the use of four regression-based models, i.e., RF regressor, light gradient boosting machine (LightGBM), extreme gradient boosting (XGBoost) and CatBoost regressor. The random forest model was configured with 500 trees, a maximum depth of 12, bootstrap sampling enabled and a fixed random state of 42, using RMSE as the evaluation criterion. LightGBM was implemented with a GBDT boosting type, 1,500 estimators, a learning rate of 0.02, 45 leaves, subsample frequency of 1, column subsampling of 0.8, minimum child samples of 15, regularization parameters (α and λ) set to 0.5 and a random state of 42. The XGBoost regressor utilized 900 estimators, a learning rate of 0.03, maximum depth of 6, subsampling and column subsampling rates of 0.85, with regularization parameters λ = 1 and α = 0.3. The CatBoost regressor was trained with 3,000 iterations, a learning rate of 0.015, maximum depth of 8, subsample rate of 0.75, RSME of 0.85 and RMSE as the loss function. All models were trained using a constant random seed (42) to ensure reproducibility. Model performance was evaluated using R² and RMSE on training and testing datasets, respectively. The model showing the most consistent performance was re-fitted using the complete dataset (n = 72) for final treatment-wise yield prediction.

RESULTS AND DISCUSSION

Fungicidal and biological treatments on the sheath blight, yield and yield attributes
 
The relative analysis of fungicidal and biological management of rice sheath blight in three seasons showed that there were vast variations in DS, PDI, grain yield and yield-attributing characters (Table 1-3). The highest incidence of the disease (38.65-41.56%), PDI (50.43-53.88%) and the lowest yield (6.81-8.07 kg plot-1) were observed in the untreated control. Uncontrolled sheath blight has been reported to cause similar decreases and losses in yields has been reported by Chitti et al., 2024).

Table 1: Yield and disease response fungicidal and biological control of rice sheath blight (2022).



Table 2: Yield and disease response fungicidal and biological control of rice sheath blight (Season-2023).



Table 3: Yield and disease response Fungicidal and biological control of rice sheath blight (Season-2024).


       
The best was nativo (Tebuconazole + Trifloxystrobin @ 0.4 g l -1) with minimum disease severity (17.67-19.85%) and PDI (15.87-17.15) and the highest grain yield (11.43-11.80 kg plot -1), grains per panicle (128-135), 1000-grain weight (22.70- 24.95 g) and panicle length. This was to be followed by the applications of validamycin 3% l L @ 2.5 ml l-1 and Azotrix (Azoxystrobin + Tricyclazole @ 1 ml l-1), which revealed similar disease suppressing and improvement in yield, these treatments also contributed significantly to disease reduction, as similarly reported by Yadav et al., (2023). Tilt (Propiconazole) and contaf plus (Hexaconazole) had moderate efficacies, whereas biological treatments (Trichoderma and Pseudomonas) offered moderate efficacies and gain yields, which implies their appropriateness to integrated disease management. Differences in treatment were found to be significant by statistical analysis (SOM, SEd, CD at 5% and 1%).
 
Correlation analysis
 
The correlation analysis showed that there was a negative correlation between yield and disease severity (DS; r = -0.69) and percent disease index (PDI; r = -0.67), which implied that with greater disease intensity, yield was reduced. The severity of the disease had a strong positive correlation with PDI (r = 0.86) and negative correlation with cluster number (r = -0.76). Cluster number had a negative association with PDI (r = -0.81) and a positive relationship with yield-related characteristics. On the contrary, there was a significant positive correlation between grain yield on the one hand and cluster number (r = 0.71), 1000-grain weight (r = 0.60) on the other hand and grains per panicle (r = 0.53) showing that they have a positive contribution to yield (Fig 6). In general, yield components had a positive impact, whereas disease parameters had a negative impact on yield.

Fig 6: Correlation analysis of yield, agronomic traits and disease parameters.


       
The cluster analysis was able to group treatments into three different categories by disease and yield characteristics (Fig 7A). The high-yield, low-disease observations (Cluster 0), intermediate performers (Cluster 1) and poor performers (Cluster 2) were separated by k-means. Cluster 0 showed high grain yield, a greater number of grains per panicle and higher 1000-grain weight, along with low disease severity (DS %) and low PDI. This cluster included Nativo, Azotrix and Validamycin., Cluster 1 had moderate levels of disease and yield and included Contaf Plus, Tilt and biological treatments and Cluster 2 had a high level of disease pressure, low panicle length and grain weight along with low yield similar to the untreated plots. The pairplot visualization ensured that the separation of clusters was clear in the case of DS%, PDI and yield traits. Clustering was supported by the principal component analysis (PC1 = 63.90 and PC2 = 13.66) with 77.56% variance being explained in the two principal components: PC1 exhibited positive loadings on yield attributes and PC2 negative loadings on the disease parameters, indicating a negative disease-yield relationship. PCA biplot (Fig 7B) placed high-yield, low-disease treatments of the positive PC1 axis and highly infected plots of the negative axis, which confirms that successful disease-suppressive treatments have a unique multivariate performance space that is consistent with field, correlation and machine-learning findings.

Fig 7: (A) Trait correlations and cluster analyses between the variables of yield and disease (B) PCA depicting significant factors of variation.


 
Correlation between grain yield and disease severity
 
It was found that grain yield had a significant negative association with the level of disease severity (DS), which is reflected in the scatter-plot (Fig 8) where a decline in yield was recorded with increase in disease severity. The regression equation and confidence band fitted indicated a statistically significant downward trend as the higher yields in the lower DS ranges (15-25%) of 10-13 kg plot-1 and that an increase in the value of the DS beyond 35% led to a sharp decrease in the yields to below 8 kg plot-1. This downward trend has been maintained consistently since the increasing sheath blight severity is very sensitive to yield and thus effective management of the disease is necessary to maintain productivity.

Fig 8: Negative correlate between yield and disease severity.


 
Machine-learning model performance
 
Among the implemented machine-learning models, RF was the most accurate in prediction (R2 = 0.848; RMSE = 0.566), then LightGBM (R2 = 0.7914; RMSE = 0.6358), XGBoost (R2 = 0.6790; RMSE = 0.7886) and CatBoost (R2 = 0.6456; RMSE = 0.8286) (Table 4, Fig 9).

Table 4: Comparison of machine learning models in predicting the yield of rice in relation to R2 and RMSE.



Fig 9: Random forest (RF) yield prediction results.


       
The preferred treatment based on the Random Forest model was very similar to the results of the fields, with the highest ranking going to Nativo, then subsequently Azotrix and Validamycin followed by the control as the lowest. This contract confirms the strength of machine-learning models to predict yields in different sheath blight pressure. Which is also explain by Xin et al. (2024).
       
Analysis of variance for disease and yield traits: ANOVA on three years found a significant difference in treatment (DS%) to the point of being highly significant (F = 88.528, p<2 × 10-16), year and year x treatment interaction were not significant.
       
The highest DS% (40.57) was observed in the untreated control, which formed a separate Tukey group (a). The lowest DS% was reported with nativo (18.47%), azotrix (19.89%) and validamycin (20.25%) and they were combined into a single group (Tukey group c) due to the lack of seasonal differences regarding the suppression of the disease (Fig 10A). The effect of treatment on PDI was very significant (F = 225.009, p<2 × 10-16). Control had the highest PDI (52.38%) whereas nativo (16.50%), azotrix (16.85%) and validamycin (19.07%) constituted the lowest DMRT group (e). Tilt, contaf plus, Trichoderma and Pseudomonas showed intermediate values of PDI (Fig 10B). A significant treatment was also found in grains per panicle (F = 12.798, p = 1.151 × 10-8). Nativo recorded the highest grains per panicle (131.22) followed by Validamycin (118.67) and Azotrix (117.67), while the lowest value was observed in control (94.44) (Fig 10C).

Fig 10: DMRT classification of treatments according to the average disease severity (DS%) (A), DMRT treatment grouping in terms of average percent disease index (PDI%) (B), DMRT treatment grouping in terms of average Grain per panicle (C), DMRT treatment grouping in terms of average 1000 Grain wt. (D), DMRT treatment grouping in terms of average plant height (E) and DMRT treatment grouping in terms of average yield (F).


       
Treatment had a considerable effect on 1000 grain weight (F = 14.472, p = 2.094 × 10-9). Nativo recorded highest grain weight (23.94g), then azotrix (23.07 g). The lowest grains weights were observed in the control and Trichoderma treatments (Fig 10D). A significant difference among treatments was also observed for plant height (F = 4.5927, p = 0.0006983). Nativo (61.18 cm), validamycin (60.97 cm) and azotrix (60.30 cm) produce the tallest plants. Contaf plus, Trichoderma and tilt showed moderate plant heights (54.90-58.49 cm). The lowest heights were recorded in the Pseudomonas (53.40 cm) and control (54.32 cm) (Fig 10E). There was also a considerable difference in grain yield between treatments (F = 19.976, p = 1.719 × 10-11). The highest yield was obtained with Nativo (11.62 kg plot-1) followed by Azotrix (10.16 kg plot-1) and Validamycin (9.93 kg plot-1). The fungicides based on hexaconazoles and the biological treatment gave intermediate values (8.28-8.91 kg plot-1), whereas the control had the lowest yield (7.49 kg plot-1) (Fig 10F).

CONCLUSION

The sheath blight is a serious challenge affecting the productivity of rice and proper fungicidal management averts these losses. Nativo was found to be better because of its dual systemic triazole-strobilurin effect, which was succeeded by validamycin and azotrix as statistically similar agents to rotate. Tilt and contaf plus gave moderate control and Trichoderma and Pseudomonas were to give an integrated management of the disease with moderate advantages of yield. Important correlations amid disease severity and yield components affirmed disease effect. The multivariate and machine-learn test confirmed the orders of treatments and proved the consistency of combined statistical methods to make predictions on yield with different degrees of sheath blight.

ACKNOWLEDGEMENT

Authors are solemnly thankful to ICAR, VC and Dean MSSSoA, CUTM for provide research facility and infrastructure.
 
Disclaimers
 
The views expressed in this article are solely those of the authors. The authors are responsible for the accuracy of the information presented.
 
Informed consent
 
This study did not include human participants, animal subjects, or any materials that would necessitate ethical approval.

CONFLICT OF INTEREST

 In this investigation there is no any potential bias or conflict of interest in terms of financial or any other means.

REFERENCES


  1. Abbas, A., Mubeen, M., Zheng, H., Sohail, M.A., Shakeel, Q., Solanki, M.K., Iftikhar, Y., Sharma, S., Kashyap, B.K., Hussain, S. and del Carmen, Z.R.M. (2022). Trichoderma spp. genes involved in the biocontrol activity against Rhizoctonia solani. Frontiers in Microbiology. 13: 884469.

  2. Biswal, K.A and Das, S. (2024). Unveiling the intricacies of the rice-Rhizoctonia pathosystem: a comprehensive review of host-pathogen interactions, molecular mechanisms and strategies for sustainable management. Journal of Plant Protection Research. 64(3): 209-233. doi: https:// doi.org/10.24425/jppr.2024.151260.

  3. Chitti, B.G., Amara, J.P., Venkatarao, P. and Chinnam, N.D. (2024). Management of rice sheath blight with new chemicals in Srikakulam District of Andhra Pradesh, India. International Journal of Plant and Soil Science. 36(9): 188-193. https:// doi.org/10.9734/ijpss/2024/v36i94966.

  4. Das, T.K., Ghosh, S. and Nath, C.P.  (2019). Brown manuring optimization in maize: Impacts on weeds, crop productivity and profitability. The Journal of Agricultural Science.  157: 7-8.

  5. El Amraoui, K., Pu, Z., Koutti, L., Masmoudi, L. and Valente de, O.J.A. (2024). Super resolution method based on generative adversarial networks with quantum feature enhancement: Application to aerial agricultural images. Neurocomputing. 577: 127346.

  6. Hussein, N.S., Safaie, N., Shamsbakhsh M. and Al-Juboory, H.H. (2025). Eco-friendly management of Rhizoctonia solani, the cause of tomato root rot disease. Agricultural Science Digest. 45(4): 657-664. doi: 10.18805/ag.DF-709.

  7. IRRI Rice Knowledge Bank: Sheath Blight. International Rice Research Institute (2019). http:// www.knowledgebank.irri.org. 

  8. Jeong, S.G., Do, Q.V. and Hwang, W.J (2023). Short-term photovoltaic power forecasting based on hybrid quantum gated recurrent unit. ICT Express. 10: 608-613.

  9. Li, Z., Pan, G., Li, Y., Wang, T., Zhang, Y. and Wang, X. (2025). Application of attention-BiLSTM within a deep learning framework for forecasting synthesis conditions of MOF materials. Materials Today Communications. pp 113838.

  10. Liu, Y., Li, W., Wang, X., Li, Y., Zhu, W., Chen, Z., Liu, X., Liu, X., Wang, W., Xue, C. and Cao, S (2025). Fine-tuning OsTCP19 expression offers broad adaptation scenarios for nitrogen-use efficiency improvement in rice. Plant Physiology. 199(2): kiaf412.

  11. Paquet, E., Soleymani, F. and Quantum, L. (2022). Hybrid quantum neural network for financial predictions. Expert Syst. Appl. 195: 116583. 

  12. Rezaei, M., Diepeveen, D., Laga, H., Gupta, S., Jones, M.G. and Sohel, F (2025). A transformer-based few-shot learning pipeline for barley disease detection from field-collected imagery. Computers and Electronics in Agriculture. 229: 109751. 

  13. Shrestha, S.M. and Misra, N.K. (1994). Evaluation of common cultivars of rice against leaf and neck blast in Nepal. J. Inst. Agric. Anim. Sci. 15: 101-103.

  14. Singh, P., Mazumdar, P., Harikrishna, J.A. and Babu, S. (2019). Sheath blight of rice: A review and identification of priorities for future research. Planta. 250(5): 1387-1407.

  15. Sushmit, M.M and Mahbubul, I.M. (2023). Forecasting solar irradiance with hybrid classical-quantum models: A comprehensive evaluation of deep learning and quantum-enhanced techniques. Energy Convers. Manag. 294: 117555.

  16. Uppala, S and Zhou, X.G. (2018). Rice Sheath Blight. The Plant Health Instructor. https://doi.org/10.1094/PHI-I-2018- 0403-01. 

  17. Xin, L., Xie, P., Wen, T., Niu, G., Yuan, J. (2024). Transfer learning enables predictions in soil-borne diseases. Soil Ecology Letters. 6(4): 240258.

  18. Yadav, L., Kumar, A., Singh, H., Punia, R. and Singh, N. (2023). Field evaluation of fungicides for the management of banded leaf and sheath blight of maize and their effect on yield parameters. Agricultural Science Digest. 43(1): 92-96. doi: 10.18805/ag.D-5521.
Published In
Agricultural Science Digest
Article Metrics
Metrics Icon
38
Views
0
Citations
Reviewed By
Share Article
Download Article
In this Article
    Contact us
    Follow us

    Full Research Article

    Assessment of Chemical and Biological Measures of Controlling Rice Sheath Blight: Field Trials and ML-based Yield Prediction

    K
    Kumar Avinash Biswal1
    0009-0006-6431-1491
    S
    Siddhartha Das1,*
    0000-0002-7482-2555
    N
    Nirakar Ranasingh2
    0009-0007-5255-2914
    R
    Rajeeb Lochan Moharana3
    0009-0002-2294-3977
    S
    Srujani Behera2
    0000-0001-7764-9631
    C
    Chandana Behera4
    0009-0000-2455-6108
    S
    Simly Das5
    0009-0009-8565-5738
    1Department of Plant Pathology, Centurion University of Technology and Management, Paralakhemundi-761 211, Odisha, India.
    2Department of Plant Pathology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
    3Department of Seed Science and Technology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
    4Department of Plant Breeding and Genetics College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.
    5Department of Nematology, College of Agriculture, Odisha University of Agriculture and Technology, Bhawanipatna-766 001, Odisha, India.

    ABSTRACT

    Background: Rhizoctonia solani is the cause of sheath blight of rice that is a significant limitation to Kharif productivity, requiring effective and unstable management approaches.

    Methods: A field test, using eight treatments, was carried out: T1-Control, T2-Validamycin 3% L, T3-Azotrix (Azoxystrobin 16.7% + Tricyclazole 33.3%), T4-Nativo (Tebuconazole 50% + Trifloxystrobin 25% WG), T5-Tilt, T6-Contaf Plus, T7-Trichoderma and T8-Pseudomonas. The severity of the disease (DS%), PDI, yield, number of grains per panicle and weight of 1000 grains were noted. ANOVA was used to analyze data (Tukey, DMRT) and prediction of yields was carried out using machine-learning models (Random forest, LightGBM, CatBoost, XGBoost).

    Result: The effects of treatment were very substantial (p<0.001) and the effect of year and the interaction of year × treatment were not significant (p = 0.608). Nativo (T4) performed better, with maximum yield (11.62 kg plot-1), grains per panicle (131.22) and 1000 gm seed weight (23.94 g) and minimum DS (18.47-20.25%) and PDI (16.50-19.06%). Azotrix and Validamycin followed. The maximum DS (40.57%) and PDI (52.38) were observed with minimum yield (7.49 kg plot-1). Random Forest got the best accuracy (R2 = 0.848; RMSE = 0.566), which validated nativo, azotric and validamycin as the most suitable management choices.

    INTRODUCTION

    Rhizoctonia solani AG1-1A (Kuhn) is a soil-borne fungus causing sheath blight, whose teleomorph is Thanatephorus cucumeris (Uppala and Zhou, 2018; Hussein et al., 2025). Disease yield loss about 10-30% and up to 50% given good conditions in the environment (Uppala and Zhou, 2018; Biswal and Das, 2024). It is tightly linked with the rise of semi-dwarf and nitrogen-sensitive varieties of rice in the world (Liu et al., 2025). The pathogen has right-angle branching hyphae and generates white mycellium and brown sclerotia (1-3 mm) on the infected tissues (Uppala and Zhou, 2018). The onset of disease is at late tillering stage of rice, early internode elongation showing water-soaked lesions and greenish-grey color center in the sheath. resulting in senescence, tiller lodging. (Uppala and Zhou, 2018; IRRI Rice Knowledge Bank, 2019). The mode of action used by Trichoderma spp. is the release of antifungal enzymes such as chitinase and glucanase causing the degradation of hypha and sclerotia, whereas Aspergillus has also been reported to exert mycoparasitic relationships with R. solani (Abbas et al., 2022).
           
    The recent progress of deep learning has greatly enhanced the system of sheath blight detection, severity and decision support system in managing rice disease (Rezaei et al., 2025). The models are quite effective in reflecting the intricate non-linear integration between the severity of the disease, environmental conditions, crop developmental phases and farming practices (El Amraoui et al., 2024). There are recurrent architectures like LSTM and BiLSTM, which obtain temporal patterns of disease progression depending on the weather conditions, nitrogen content, canopy density and pathogen dynamics (Jeong et al., 2023; Li et al., 2025). Deep learning in combination with biological knowledge can be used to detect the disease in its early stages and can provide optimal and sustainable strategies of sheath blight management (Paquet et al., 2022; Sushmit et al., 2023; Singh et al., 2019).

    MATERIALS AND METHODS

    Experimental site and experimental design
     
    The experiment was conducted rabi season of 2022-2024 at Ranadevi farm of Centurion University of Technology and Management, Parlakhemundi, Odisha. The experiment was conducted in Swarna (MTU 7029) variety of rice. The land was ploughed twice crosswise by tractor drawn harrow and weeds were removed thoroughly from the field. Nursery bed was raised and sowing was done (Fig 1). Main field was prepared in an area of (13.6 × 21.7) m2, along with 0.5 m irrigation channel. Thirty days aged seedlings were transplanted into the main field in three replications with a spacing (4.2 × 3.2) m2. Plant to plant and the row-to-row distance were 10 × 20 cm. 30 days old seedlings were transferred from the nursery to the main field.

    Fig 1: Raising of seedlings in nursery bed of rice preparation under field conditions (A), and Experiment to manage sheath blight (Rhizoctonia solani) using rice nursery of variety MTU 7029 (B).


     
    Symptomatic characterization
     
    The symptoms of sheath blight were studied and systematically monitored in the field as well as microscopic study and photographed. The symptoms first appeared: small, water soaked, oval-irregular spots on the waterline on the leaf sheath which gradually increased in size with greyish centers and brown margins, ascending and descending the sheath and the leaf blades. Lesions formed large necrotic spots as the disease progressed and in extreme cases sheaths and leaves were heavily affected resulting in wilting and lodging (Fig 2). Microscopic studies were also able to confirm the morphological character of Rhizoctonia solani (Fig 3).

    Fig 2: Development of symptoms of rice sheath blight under field conditions.



    Fig 3: Microscopic image of Rhizoctonia solani distinct right-angle branching and septation of hyphae (A), Sclerotium of R. solani in stereoscopic microscope (B) and Rhizoctonia solani on potato dextrose agar (PDA) (C).


     
    Experimental design and treatments
     
    The experiment was laid out in a randomized complete block design (RCBD) with eight treatments and three replications per year. The field was artificially inoculated with R. solani (Fig 4,5). The treatments comprised, T1- Control, T2- Validamycin 3% L, T3- Azotrix (Azoxystrobin 16.7% + Tricyclazole 33.3%), T4- Nativo (Tebuconazole 50% + Trifloxystrobin 25% WG), T5- Tilt (Propiconazole 25% EC), T6- Contaf plus (Hexaconazole 5% SC), T7- Trichoderma harzianum, T8- Pseudomonas fluorescens. Foliar application was done at maximum tillering and panicle initiation stage of crop.

    Fig 4: Field experiments Intercultural operations in transplanted rice under field conditions (A), Fungicidal and biological treatment of rice field made in knapsack sprayer (B) and Field observation of the experimental plots of rice at the time of maturity with treatment-wise crop response (C).



    Fig 5: Inoculation of rice plants by artificial inoculation of Rhizoctonia solani (A) and Oblique view of reproduced irrigated rice plots using drones (B).


     
    Data sources
     
    DS and PDI were recorded at Ranadevi farm of Centurion University, data is taken in 10 days interval after inoculation up to 9 observations. The disease was scored by using 0-9 scale based on lesion character and lesion area, where 0 represented no lesion characteristics and 9 represented lesions in more than 75% leaf area. DS and PDI was calculated by formula given by Shrestha and Mishra (1994).



     
     Measurement of yield and yield attribute
     
    Plant height was assessed at maturity from ground level to the tip of the tallest panicle on 10 randomly selected plants per plot using a measuring tape. Ten representative panicles were selected randomly in each plot and manual counting of grains was done. Weighed clean and sun-dried seeds with 1000-grain weight (g) of the adjustment to 14% of moisture. The harvested net plot yield was dried and adjusted to 14% moisture content and recorded as grain yield (Kg plot-1).
     
    Statistical analysis
     
    Over the course of three years, a mixed analysis of variance (ANOVA) was performed based on the described linear model. The means of treatment were isolated with the help of the Tukey’s honest significant difference (HSD) test and the Dunkan’s multiple range test (DMRT) at p<0.05. Statistical analysis was done in R software with the help of agricolae package (Das et al., 2019).
     
    Machine learning analysis
     
    The regression analysis was conducted using the pooled data of 2022-2024 cropping seasons (n = 72) to forecast the rice grain yield under sheath blight pressure. Before developing the model, the numerical variables had been normalized via Z-score transformation and the categorical ones transformed into codes and the dataset was splitting into training (80) and testing (20) sets. The analysis used a constant random state (42) so that it can be reproducible and comparable across models.
           
    Four machine-learning models were tested on predicting the grain yield under different sheath blight pressure with the use of four regression-based models, i.e., RF regressor, light gradient boosting machine (LightGBM), extreme gradient boosting (XGBoost) and CatBoost regressor. The random forest model was configured with 500 trees, a maximum depth of 12, bootstrap sampling enabled and a fixed random state of 42, using RMSE as the evaluation criterion. LightGBM was implemented with a GBDT boosting type, 1,500 estimators, a learning rate of 0.02, 45 leaves, subsample frequency of 1, column subsampling of 0.8, minimum child samples of 15, regularization parameters (α and λ) set to 0.5 and a random state of 42. The XGBoost regressor utilized 900 estimators, a learning rate of 0.03, maximum depth of 6, subsampling and column subsampling rates of 0.85, with regularization parameters λ = 1 and α = 0.3. The CatBoost regressor was trained with 3,000 iterations, a learning rate of 0.015, maximum depth of 8, subsample rate of 0.75, RSME of 0.85 and RMSE as the loss function. All models were trained using a constant random seed (42) to ensure reproducibility. Model performance was evaluated using R² and RMSE on training and testing datasets, respectively. The model showing the most consistent performance was re-fitted using the complete dataset (n = 72) for final treatment-wise yield prediction.

    RESULTS AND DISCUSSION

    Fungicidal and biological treatments on the sheath blight, yield and yield attributes
     
    The relative analysis of fungicidal and biological management of rice sheath blight in three seasons showed that there were vast variations in DS, PDI, grain yield and yield-attributing characters (Table 1-3). The highest incidence of the disease (38.65-41.56%), PDI (50.43-53.88%) and the lowest yield (6.81-8.07 kg plot-1) were observed in the untreated control. Uncontrolled sheath blight has been reported to cause similar decreases and losses in yields has been reported by Chitti et al., 2024).

    Table 1: Yield and disease response fungicidal and biological control of rice sheath blight (2022).



    Table 2: Yield and disease response fungicidal and biological control of rice sheath blight (Season-2023).



    Table 3: Yield and disease response Fungicidal and biological control of rice sheath blight (Season-2024).


           
    The best was nativo (Tebuconazole + Trifloxystrobin @ 0.4 g l -1) with minimum disease severity (17.67-19.85%) and PDI (15.87-17.15) and the highest grain yield (11.43-11.80 kg plot -1), grains per panicle (128-135), 1000-grain weight (22.70- 24.95 g) and panicle length. This was to be followed by the applications of validamycin 3% l L @ 2.5 ml l-1 and Azotrix (Azoxystrobin + Tricyclazole @ 1 ml l-1), which revealed similar disease suppressing and improvement in yield, these treatments also contributed significantly to disease reduction, as similarly reported by Yadav et al., (2023). Tilt (Propiconazole) and contaf plus (Hexaconazole) had moderate efficacies, whereas biological treatments (Trichoderma and Pseudomonas) offered moderate efficacies and gain yields, which implies their appropriateness to integrated disease management. Differences in treatment were found to be significant by statistical analysis (SOM, SEd, CD at 5% and 1%).
     
    Correlation analysis
     
    The correlation analysis showed that there was a negative correlation between yield and disease severity (DS; r = -0.69) and percent disease index (PDI; r = -0.67), which implied that with greater disease intensity, yield was reduced. The severity of the disease had a strong positive correlation with PDI (r = 0.86) and negative correlation with cluster number (r = -0.76). Cluster number had a negative association with PDI (r = -0.81) and a positive relationship with yield-related characteristics. On the contrary, there was a significant positive correlation between grain yield on the one hand and cluster number (r = 0.71), 1000-grain weight (r = 0.60) on the other hand and grains per panicle (r = 0.53) showing that they have a positive contribution to yield (Fig 6). In general, yield components had a positive impact, whereas disease parameters had a negative impact on yield.

    Fig 6: Correlation analysis of yield, agronomic traits and disease parameters.


           
    The cluster analysis was able to group treatments into three different categories by disease and yield characteristics (Fig 7A). The high-yield, low-disease observations (Cluster 0), intermediate performers (Cluster 1) and poor performers (Cluster 2) were separated by k-means. Cluster 0 showed high grain yield, a greater number of grains per panicle and higher 1000-grain weight, along with low disease severity (DS %) and low PDI. This cluster included Nativo, Azotrix and Validamycin., Cluster 1 had moderate levels of disease and yield and included Contaf Plus, Tilt and biological treatments and Cluster 2 had a high level of disease pressure, low panicle length and grain weight along with low yield similar to the untreated plots. The pairplot visualization ensured that the separation of clusters was clear in the case of DS%, PDI and yield traits. Clustering was supported by the principal component analysis (PC1 = 63.90 and PC2 = 13.66) with 77.56% variance being explained in the two principal components: PC1 exhibited positive loadings on yield attributes and PC2 negative loadings on the disease parameters, indicating a negative disease-yield relationship. PCA biplot (Fig 7B) placed high-yield, low-disease treatments of the positive PC1 axis and highly infected plots of the negative axis, which confirms that successful disease-suppressive treatments have a unique multivariate performance space that is consistent with field, correlation and machine-learning findings.

    Fig 7: (A) Trait correlations and cluster analyses between the variables of yield and disease (B) PCA depicting significant factors of variation.


     
    Correlation between grain yield and disease severity
     
    It was found that grain yield had a significant negative association with the level of disease severity (DS), which is reflected in the scatter-plot (Fig 8) where a decline in yield was recorded with increase in disease severity. The regression equation and confidence band fitted indicated a statistically significant downward trend as the higher yields in the lower DS ranges (15-25%) of 10-13 kg plot-1 and that an increase in the value of the DS beyond 35% led to a sharp decrease in the yields to below 8 kg plot-1. This downward trend has been maintained consistently since the increasing sheath blight severity is very sensitive to yield and thus effective management of the disease is necessary to maintain productivity.

    Fig 8: Negative correlate between yield and disease severity.


     
    Machine-learning model performance
     
    Among the implemented machine-learning models, RF was the most accurate in prediction (R2 = 0.848; RMSE = 0.566), then LightGBM (R2 = 0.7914; RMSE = 0.6358), XGBoost (R2 = 0.6790; RMSE = 0.7886) and CatBoost (R2 = 0.6456; RMSE = 0.8286) (Table 4, Fig 9).

    Table 4: Comparison of machine learning models in predicting the yield of rice in relation to R2 and RMSE.



    Fig 9: Random forest (RF) yield prediction results.


           
    The preferred treatment based on the Random Forest model was very similar to the results of the fields, with the highest ranking going to Nativo, then subsequently Azotrix and Validamycin followed by the control as the lowest. This contract confirms the strength of machine-learning models to predict yields in different sheath blight pressure. Which is also explain by Xin et al. (2024).
           
    Analysis of variance for disease and yield traits: ANOVA on three years found a significant difference in treatment (DS%) to the point of being highly significant (F = 88.528, p<2 × 10-16), year and year x treatment interaction were not significant.
           
    The highest DS% (40.57) was observed in the untreated control, which formed a separate Tukey group (a). The lowest DS% was reported with nativo (18.47%), azotrix (19.89%) and validamycin (20.25%) and they were combined into a single group (Tukey group c) due to the lack of seasonal differences regarding the suppression of the disease (Fig 10A). The effect of treatment on PDI was very significant (F = 225.009, p<2 × 10-16). Control had the highest PDI (52.38%) whereas nativo (16.50%), azotrix (16.85%) and validamycin (19.07%) constituted the lowest DMRT group (e). Tilt, contaf plus, Trichoderma and Pseudomonas showed intermediate values of PDI (Fig 10B). A significant treatment was also found in grains per panicle (F = 12.798, p = 1.151 × 10-8). Nativo recorded the highest grains per panicle (131.22) followed by Validamycin (118.67) and Azotrix (117.67), while the lowest value was observed in control (94.44) (Fig 10C).

    Fig 10: DMRT classification of treatments according to the average disease severity (DS%) (A), DMRT treatment grouping in terms of average percent disease index (PDI%) (B), DMRT treatment grouping in terms of average Grain per panicle (C), DMRT treatment grouping in terms of average 1000 Grain wt. (D), DMRT treatment grouping in terms of average plant height (E) and DMRT treatment grouping in terms of average yield (F).


           
    Treatment had a considerable effect on 1000 grain weight (F = 14.472, p = 2.094 × 10-9). Nativo recorded highest grain weight (23.94g), then azotrix (23.07 g). The lowest grains weights were observed in the control and Trichoderma treatments (Fig 10D). A significant difference among treatments was also observed for plant height (F = 4.5927, p = 0.0006983). Nativo (61.18 cm), validamycin (60.97 cm) and azotrix (60.30 cm) produce the tallest plants. Contaf plus, Trichoderma and tilt showed moderate plant heights (54.90-58.49 cm). The lowest heights were recorded in the Pseudomonas (53.40 cm) and control (54.32 cm) (Fig 10E). There was also a considerable difference in grain yield between treatments (F = 19.976, p = 1.719 × 10-11). The highest yield was obtained with Nativo (11.62 kg plot-1) followed by Azotrix (10.16 kg plot-1) and Validamycin (9.93 kg plot-1). The fungicides based on hexaconazoles and the biological treatment gave intermediate values (8.28-8.91 kg plot-1), whereas the control had the lowest yield (7.49 kg plot-1) (Fig 10F).

    CONCLUSION

    The sheath blight is a serious challenge affecting the productivity of rice and proper fungicidal management averts these losses. Nativo was found to be better because of its dual systemic triazole-strobilurin effect, which was succeeded by validamycin and azotrix as statistically similar agents to rotate. Tilt and contaf plus gave moderate control and Trichoderma and Pseudomonas were to give an integrated management of the disease with moderate advantages of yield. Important correlations amid disease severity and yield components affirmed disease effect. The multivariate and machine-learn test confirmed the orders of treatments and proved the consistency of combined statistical methods to make predictions on yield with different degrees of sheath blight.

    ACKNOWLEDGEMENT

    Authors are solemnly thankful to ICAR, VC and Dean MSSSoA, CUTM for provide research facility and infrastructure.
     
    Disclaimers
     
    The views expressed in this article are solely those of the authors. The authors are responsible for the accuracy of the information presented.
     
    Informed consent
     
    This study did not include human participants, animal subjects, or any materials that would necessitate ethical approval.

    CONFLICT OF INTEREST

     In this investigation there is no any potential bias or conflict of interest in terms of financial or any other means.

    REFERENCES


    1. Abbas, A., Mubeen, M., Zheng, H., Sohail, M.A., Shakeel, Q., Solanki, M.K., Iftikhar, Y., Sharma, S., Kashyap, B.K., Hussain, S. and del Carmen, Z.R.M. (2022). Trichoderma spp. genes involved in the biocontrol activity against Rhizoctonia solani. Frontiers in Microbiology. 13: 884469.

    2. Biswal, K.A and Das, S. (2024). Unveiling the intricacies of the rice-Rhizoctonia pathosystem: a comprehensive review of host-pathogen interactions, molecular mechanisms and strategies for sustainable management. Journal of Plant Protection Research. 64(3): 209-233. doi: https:// doi.org/10.24425/jppr.2024.151260.

    3. Chitti, B.G., Amara, J.P., Venkatarao, P. and Chinnam, N.D. (2024). Management of rice sheath blight with new chemicals in Srikakulam District of Andhra Pradesh, India. International Journal of Plant and Soil Science. 36(9): 188-193. https:// doi.org/10.9734/ijpss/2024/v36i94966.

    4. Das, T.K., Ghosh, S. and Nath, C.P.  (2019). Brown manuring optimization in maize: Impacts on weeds, crop productivity and profitability. The Journal of Agricultural Science.  157: 7-8.

    5. El Amraoui, K., Pu, Z., Koutti, L., Masmoudi, L. and Valente de, O.J.A. (2024). Super resolution method based on generative adversarial networks with quantum feature enhancement: Application to aerial agricultural images. Neurocomputing. 577: 127346.

    6. Hussein, N.S., Safaie, N., Shamsbakhsh M. and Al-Juboory, H.H. (2025). Eco-friendly management of Rhizoctonia solani, the cause of tomato root rot disease. Agricultural Science Digest. 45(4): 657-664. doi: 10.18805/ag.DF-709.

    7. IRRI Rice Knowledge Bank: Sheath Blight. International Rice Research Institute (2019). http:// www.knowledgebank.irri.org. 

    8. Jeong, S.G., Do, Q.V. and Hwang, W.J (2023). Short-term photovoltaic power forecasting based on hybrid quantum gated recurrent unit. ICT Express. 10: 608-613.

    9. Li, Z., Pan, G., Li, Y., Wang, T., Zhang, Y. and Wang, X. (2025). Application of attention-BiLSTM within a deep learning framework for forecasting synthesis conditions of MOF materials. Materials Today Communications. pp 113838.

    10. Liu, Y., Li, W., Wang, X., Li, Y., Zhu, W., Chen, Z., Liu, X., Liu, X., Wang, W., Xue, C. and Cao, S (2025). Fine-tuning OsTCP19 expression offers broad adaptation scenarios for nitrogen-use efficiency improvement in rice. Plant Physiology. 199(2): kiaf412.

    11. Paquet, E., Soleymani, F. and Quantum, L. (2022). Hybrid quantum neural network for financial predictions. Expert Syst. Appl. 195: 116583. 

    12. Rezaei, M., Diepeveen, D., Laga, H., Gupta, S., Jones, M.G. and Sohel, F (2025). A transformer-based few-shot learning pipeline for barley disease detection from field-collected imagery. Computers and Electronics in Agriculture. 229: 109751. 

    13. Shrestha, S.M. and Misra, N.K. (1994). Evaluation of common cultivars of rice against leaf and neck blast in Nepal. J. Inst. Agric. Anim. Sci. 15: 101-103.

    14. Singh, P., Mazumdar, P., Harikrishna, J.A. and Babu, S. (2019). Sheath blight of rice: A review and identification of priorities for future research. Planta. 250(5): 1387-1407.

    15. Sushmit, M.M and Mahbubul, I.M. (2023). Forecasting solar irradiance with hybrid classical-quantum models: A comprehensive evaluation of deep learning and quantum-enhanced techniques. Energy Convers. Manag. 294: 117555.

    16. Uppala, S and Zhou, X.G. (2018). Rice Sheath Blight. The Plant Health Instructor. https://doi.org/10.1094/PHI-I-2018- 0403-01. 

    17. Xin, L., Xie, P., Wen, T., Niu, G., Yuan, J. (2024). Transfer learning enables predictions in soil-borne diseases. Soil Ecology Letters. 6(4): 240258.

    18. Yadav, L., Kumar, A., Singh, H., Punia, R. and Singh, N. (2023). Field evaluation of fungicides for the management of banded leaf and sheath blight of maize and their effect on yield parameters. Agricultural Science Digest. 43(1): 92-96. doi: 10.18805/ag.D-5521.
    In this Article
      Contact us
      Follow us
      Published In
      Agricultural Science Digest

      Editorial Board

      View all (0)