search for




 

Application of Numerical Weather Prediction Data to Estimate Infection Risk of Bacterial Grain Rot of Rice in Korea
Plant Pathol. J. 2020;36:54-66
Published online February 1, 2020
© 2020 The Korean Society of Plant Pathology.

Hyo-suk Kim1 , Ki Seok Do2, Joo Hyeon Park3 , Wee Soo Kang4 , Yong Hwan Lee4, and Eun Woo Park1,2,5,6*

1Department of Agricultural Biotechnology, Seoul National University, Seoul 08826, Korea
2National Center for Agrometeorology, Seoul 08826, Korea
3EPINET Corporation, Anyang 14056, Korea
4Department of Agro-food Safety and Crop Protection, National Institute of Agricultural Sciences, Rural Development Administration, Wanju 55365, Korea
5Interdisciplinary Program in Agricultural and Forest Meteorology, Seoul National University, Seoul 08826, Korea
6Research Institute of Agriculture and Life Sciences, Seoul National University, Seoul 08826, Korea
Correspondence to: *Phone) +82-2-880-4672, FAX) +82-2-872-2317
E-mail) ewpark@snu.ac.kr
ORCID
Hyo-suk Kim
https://orcid.org/0000-0002-3031-9333
Joo Hyeon Park
https://orcid.org/0000-0001-6218-4456
Wee Soo Kang
https://orcid.org/0000-0001-9162-4086
Eun Woo Park
https://orcid.org/0000-0001-8305-5709

Handling Editor : Sook-Young Park
Received November 26, 2019; Revised November 28, 2019; Accepted November 28, 2019.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract

This study was conducted to evaluate usefulness of numerical weather prediction data generated by the Unified Model (UM) for plant disease forecast. Using the UM06- and UM18-predicted weather data, which were released at 0600 and 1800 Universal Time Coordinated (UTC), respectively, by the Korea Meteorological Administration (KMA), disease forecast on bacterial grain rot (BGR) of rice was examined as compared with the model output based on the automated weather stations (AWS)-observed weather data. We analyzed performance of BGRcast based on the UM-predicted and the AWS-observed daily minimum temperature and average relative humidity in 2014 and 2015 from 29 locations representing major rice growing areas in Korea using regression analysis and two-way contingency table analysis. Temporal changes in weather conduciveness at two locations in 2014 were also analyzed with regard to daily weather conduciveness (Ci) and the 20-day and 7-day moving averages of Ci for the inoculum build-up phase (Cinc) prior to the panicle emergence of rice plants and the infection phase (Cinf) during the heading stage of rice plants, respectively. Based on Cinc and Cinf, we were able to obtain the same disease warnings at all locations regardless of the sources of weather data. In conclusion, the numerical weather prediction data from KMA could be reliable to apply as input data for plant disease forecast models. Weather prediction data would facilitate applications of weather-driven disease models for better disease management. Crop growers would have better options for disease control including both protective and curative measures when weather prediction data are used for disease warning.

Keywords : bacterial grain rot of rice, numerical weather prediction data, plant disease forecast, Unified Model
Materials and Methods

Disease forecast model

The BGRcast uses conduciveness of weather conditions for BGR development as disease risk factor that measures likeliness of bacterial inoculum buildup (Cinc) and infection (Cinf) at the pre- and late-heading stages of rice, respectively (Lee et al., 2015). With the base temperature and relative humidity being 22°C and 80%, respectively, the disease risk thresholds adopted in BGRcast were Cinc=0.3 and Cinf=0.5. The base relative humidity and temperature used in BGRcast were determined by Lee et al. (2015) based on the field observations that BGR was often detected when daily minimum temperature was ≥22°C and daily average relative humidity was ≥80%. When Cinc and Cinf are above the thresholds, disease warnings are made to advise bactericide sprays at the pre- and late-heading stages of rice, respectively. Consequently, the fixed two-spray scheme, which Korean rice growers commonly adopt to control BGR, can be improved by eliminating unnecessary sprays depending on the BGRcast forecast.

Observed crop and weather data

The BGRcast requires heading dates of rice cultivars to estimate environmental conduciveness for bacterial inoculum build-up prior to the panicle emergence and infection during the panicle heading period. In this study, crop data on cultivars and heading dates of rice grown at 29 rice paddy fields in 2014 were collected from the National Crop Pest Management System (NCPMS) of the Rural Development Administration of Korea. Heading dates of 14 rice cultivars at 29 rice paddy fields were given with the location data in Table 1. The rice heading dates varied depending on cultivars and locations with different weather conditions. The heading date was defined to be the date that approximately 40% of rice panicles were emerged (Lee et al., 2015).

The observed daily minimum temperature and average relative humidity data in 2014 and 2015 were collected from the automated weather observation network of KMA. Based on the longitude and latitude of the 29 paddy fields, the nearest AWS from individual paddy fields were identified. The distance between the paddy fields and their nearest AWS were in the range of 0.04–17.67 km (Table 1). The site ID denoted the ascending order of the distance from the rice paddy fields to their nearest AWS. The geographical locations of rice paddy fields and nearest AWS were displayed in Fig. 1. The data collection sites for rice heading dates and weather conditions were distributed in the major rice growing areas throughout the country, and their ground elevation varied in the range of 0 to 327 m for rice paddy fields and 6 to 353 m for weather stations.

Weather prediction data

Daily weather prediction data to be used as input for BGRcast were generated by LDAPS utilizing the ‘UM release version 8.2’ (Lee and Chun, 2015). Of the four sets of weather prediction data released at different times, we used the UM-data released at 0600 (UM06) and 1800 (UM18) UTC to generate 2-day disease forecast by BGRcast on the conduciveness of weather conditions for BGR development. Because the Korea Standard Time (KST) is 9 hours ahead of UTC, UM18 and UM06 were released at 0300 and 1500 KST during a day, respectively. Consequently, hourly weather prediction data for 0000–0200 KST from UM18 of yesterday and for 0300–2300 KST from UM18 of today were used to forecast today’s disease risk. In order to forecast tomorrow’s disease risk, hourly weather prediction data for 0000–2300 KST from UM06 were used. The UM data in GRIB2 file format were converted to generic text files using the KWGRIB2 software which was developed by KMA to extract weather prediction data. Hourly weather prediction data by UM in 2014 and 2015 were retrieved from KMA to run BGRcast in this study.

Evaluation of daily weather prediction data

Accuracy of daily weather prediction data from UM for the locations of 29 paddy fields was evaluated by comparing with the observed daily weather data from corresponding locations in 2014 and 2015. Daily minimum temperature and average relative humidity during May 5 to October 31, 2014 and 2015 were used to compare the UM-predicted and the AWS-observed data using regression analysis. The absolute differences between the UM-predicted and the AWS-observed data and their root mean square error (RMSE) in daily minimum temperature were calculated from 10,411 data points ([(180 days in 2014) + (179 days in 2015)] × 29 sites) to examine the magnitude and variability of the differences. In the case of daily average relative humidity, differences between the predicted and the observed data and their RMSE were examined using 10,380 data points. There were missing relative humidity data for 31 days. The two-way contingency table analysis (Sokal and Rohlf, 1973) on the UM-predicted and the AWS-observed relative humidity was carried out to investigate the impact of input weather data on the disease warnings by BGRcast. A total of 10,380 data points were categorized into four groups with reference to 80% relative humidity, which is the threshold of relative humidity for BGR development. The appropriateness of daily weather prediction data as meteorological inputs was evaluated based on the coefficient of determination (R2) and RMSE values. The ‘SciPy’ and ‘Pandas’ packages in Python version 3.7 (Python Software Foundation, Wilmington, DE, USA) were used to perform the regression analysis and RMSE calculation.

Verification on use of the UM-predicted weather data

The UM-based BGR forecast was verified in three ways. Firstly, relationships between the UM-based and the AWS-based estimates of Cinc and Cinf were examined by regression analysis. Secondly, the two-way contingency table analysis was conducted to evaluate concurrence of the BGRcast warnings from the UM-based and the AWS-based disease forecasts for all 29 locations in Table 1. Thirdly, temporal changes in the UM-based estimates of Ci, Cinc, and Cinf were compared with the AWS-based estimates over the rice growing season for two locations in 2014. We selected two locations in consideration of the distance to their nearest AWS and the BGRcast warnings. In this analysis, we used only the UM06 data as input for BGRcast in order to avoid redundancy. For the first and second verification tests, we used weather data from 2014 and 2015, and calculated Cinc and Cinf assuming that heading dates of rice plants varied from July 15 to September 9. For the third verification tests, we used weather data only from 2014 since the heading dates of rice cultivars in Table 1 were not available for 2015. Without heading date data, it is not possible to delineate the periods of inoculum build-up and infection phases of the bacteria during the rice growing season (Lee et al., 2015). The disease risk thresholds of Cinc and Cinf for BGRcast warnings were 0.3 and 0.5, respectively, as suggested by Lee et al. (2015).

Results

Evaluation of daily weather prediction data

The regression analysis in Fig. 2 indicated that the UM-predicted and the AWS-observed daily minimum temperature appeared similar to each other in both cases of UM06 and UM18. The coefficient of determination (R2) of regression equations was approximately 0.9, and the regression coefficients were close to 1.0. The absolute mean differences between the observed and the predicted daily minimum temperatures were less than 2°C in the most locations except Site 22, which is an East coast area of South Korea (Fig. 3). The differences were a little greater for UM18 than for UM06 as was indicated by RMSE.

As for daily average relative humidity, the regression equations between the UM-predicted and the AWS-observed were statistically significant for both UM06 and UM18 (Fig. 4). However, R2’s and the scatter plots indicated that the relationship between the observed and the predicted daily average relative humidity were not as similar to each other as in the case of daily minimum temperature. The differences between the UM-predicted and the AWS-observed daily average relative humidity varied widely and RMSE for the differences was almost 10% (Fig. 5). The overall patterns of differences between the UM-predicted and the AWS-observed daily average relative humidity across the 29 locations appeared similar for UM06 and UM18. The difference in relative humidity was particularly high in the case of Site 22.

Results of two-way contingency table analysis on concurrence of the UM-predicted and the AWS-observed daily average relative humidity were graphically presented with reference to the threshold of relative humidity in Fig. 6. The threshold of daily average relative humidity was 80% as suggested by Lee et al. (2015). When the threshold was used as a criterion for categorization of the relationship between the UM-predicted and the AWS-observed data, the probability of detection (POD) and the false alarm ratio (FAR) by the UM06 were 58.4% and 29.2%, respectively. As for UM18, they were 63.5% and 33.6%, respectively. In this analysis, POD and FAR are the percent detection and false-detection, respectively, of the AWS-observed relative humidity of ≥80% by the UM-predicted.

Verification on use of the UM-predicted weather data

The relationships between the UM-based and the AWS-based estimates of Cinc and Cinf were presented in Fig. 7. The regression coefficients for all four graphs were close to 1.0 with the intercept being approximately 0.0, suggesting that use of the UM-predicted weather data as input for BGRcast would result in Cinc and Cinf that are similar to the model outputs produced by using the AWS-observed weather data as input. The R2’s were higher for both Cinc and Cinf when the UM06-predicted data were used (Fig. 7A and C) than the UM18-predicted data (Fig. 7B and D). It was also found that R2’s for Cinf (Fig. 7C and D) was higher than for Cinc (Fig. 7A and B) in both cases of using the UM06- and the UM18-predicted weather data as input for BGRcast.

Results of the two-way contingency table analysis on concurrence of BGRcast warnings from the UM-based and the AWS-based disease forecasts were presented in Table 2. A total of 3,306 cases (57 days × 29 locations × 2 years) were included in the analysis by using weather data from 2014 and 2015 with varying heading dates of rice plants from July 15 to September 9. As compared with the AWS-based disease warnings for the pre-heading spray, the UM06-based BGRcast was able to provide the same disease warnings at 98.18% POD, and issue incorrect disease warnings at 3.74% FAR. Regarding disease warnings for the post-heading spray, POD and FAR were 95.22% and 7.12%, respectively. In general, the UM06 and UM18-predicted weather data resulted in almost the same outputs on Cinc and Cinf with negligible differences. The accuracy and bias score indices indicated that both the UM-based and the AWS-based disease forecasts by BGRcast would provide almost the same outputs.

Using the heading dates and the AWS-observed weather data at 29 locations in 2014, Cinc and Cinf were estimated by BGRcast for the inoculum build-up and infection phases, respectively (Table 3). Durations of the inoculum build-up phase and the infection phase were determined with reference to the observed heading dates of rice cultivars. As was defined by Lee et al. (2015), the infection phase spans 7 days during the period between 3 days prior to and 3 days posterior to the heading date. The inoculum build-up phase extends over 20 days before the panicle emergence, which is the staring day of the infection phase. A warning for the pre-heading spray was advised on the last day of inoculum build-up phase if Cinc ≥0.3, which is the threshold for disease risk. In this study, none of 29 locations had a warning for the post-heading spray because Cinf on the last day of the infection phase at all locations was less than the threshold of Cinf, which is 0.5. Based on disease warnings that were advised by BGRcast using the AWS-observed weather data, 29 locations were categorized into two disease warning groups (Table 3). The first group includes 14 locations where both Cinc and Cinf were not reached their thresholds and no disease warnings were advised. The second group of 15 locations had a pre-heading warning with Cinc ≥0.3. None of 29 locations had only a post-heading warning or both pre- and post-heading warnings when the AWS-observed weather data were used as input data for BGRcast.

Temporal changes of Ci, Cinc and Cinf over the rice growing season in 2014 were examined for two locations, Yangju and Goseong, representing each of the two disease waring groups (Fig. 8). The two locations had AWS for weather monitoring at the closest distance among all locations in the respective disease warning groups (Table 3). The UM06-predicted and the AWS-observed weather data resulted in different daily weather conduciveness (Ci). However, there were not much differences in the moving averages of Ci for 20 days of the inoculum build-up phase (Cinc) and for 7 days of the infection phase (Cinf) between the outputs of BGRcast using the UM06-predicted and the AWS-observed weather data. The daily weather conduciveness (Ci) suggested that weather conditions in 2014 were more favorable for BGR development at Goseong than Yangju. In Yangju, the UM06-based Cinc over the inoculum build-up phase during July 24–August 12 was ≤0.12, and the UM06-based Cinf over the infection phase during August 13-August 19 was 0.00. In the case of Goseong, the UM06-based Cinc during July 30–August 18 and the UM06-based Cinf during August 19–August 25 ranged 0.42–1.04 and 0.08–0.32, respectively.

Discussion

It was found that the UM-predicted weather data would be as useful as the AWS-observed weather data for disease forecast by BGRcast. Even though the differences between the UM-predicted and the AWS-observed daily average relative humidity was obvious, the difference in relative humidity did not affect the BGRcast output significantly as was shown in Fig. 7. This was due to the fact that BGRcast adopted 80% relative humidity as the threshold for estimating conduciveness of weather conditions (Lee et al., 2015). The two-way contingency table analysis with reference to the threshold of daily average relative humidity in Fig. 6 suggested that approximately 75% of 10,380 data points in Fig. 5 should have resulted in no differences in the BGRcast output although the UM-predicted and the AWS-observed relative humidity were not exactly same. It is also shown in Fig. 6 that there might be approximately 13–15% and 9–12% chances of missing and false warnings, respectively, on disease risk by BGRcast. However, the chances of miss-forecasting by BGRcast could be reduced in reality because of the temperature threshold in the model. For example, when daily minimum temperature is lower than 22°C, BGRcast would not advise disease risk warnings even if daily average relative humidity were higher than 80%. The results from BGRcast suggested possible use of the UM-predicted weather data for plant disease forecast in general even if accuracy of disease forecast may vary depending on the sensitivity of disease forecast models to weather variables.

Numerical weather prediction models like UM produce gridded weather data covering a large area at a certain spatial resolution using various methods to calculate meteorological variables within the grids (Collins et al., 2013; Staniforth et al., 2014). Difference between the UM-predicted and the AWS-observed data was partially attributed to grid generation methods of UM and physical nature of meteorological variables (Cullen and Davies, 1991; Mesinger, 1981; Walters et al., 2019). The UM that was used in this study has 470,162 grid points at the spatial resolution of 1.5 km × 1.5 km covering the Korean Peninsula. Unlike other sites in Table 1, Site 22 (Gangneung) is a coastal area at the distance of only 332m from the East Sea. The UM-grid cell encompassing Site 22 covers the area which consists of both ocean and inland. The geographical location of Site 22 probably has caused the particularly large differences between the UM-predicted and the AWS-observed weather data, especially relative humidity, as compared with other locations. The R2’s of the regression equations in Fig. 7 indicated that the UM06-predicted weather data were generally more accurate than the UM18 when compared with the AWS-observed weather data. Since Cinc is the 20-day moving average of the daily conduciveness of weather conditions (Ci), the relationship between the UM-based and the AWS-based BGRcast showed higher R2 for Cinc than Cinf, which is the 7-day moving average of Ci (Fig. 7). Based on the results from the regression analyses in Fig. 7, the UM-based BGRcast would be as good in estimating Cinc and Cinf as the AWS-based BGRcast. Furthermore, the two-way contingency table analysis in Table 2 showed the disease warnings based on the UM-predicted weather data were highly concurrent with the warnings based on the AWS-observed weather data.

It was suggested by Lee et al. (2015) that Cinc and Cinf estimated by BGRcast could be used as a disease risk factor for determining whether or not to spray bactericides at the pre- and post-heading stages, respectively. Based on the UM06-based BGRcast warnings, there was no need for bactericide sprays at both the pre- and post-heading stages at Yangju, whereas only the pre-heading spray was necessary at Goseong. Although the AWS-based BGRcast resulted in slightly higher Cinc and Cinf at Goseong than the UM-based BGRcast, disease warnings for bactericide sprays should have been the same regardless of the sources of weather data for both Goseong and Yangju in 2014. By using the UM-predicted weather data, it was possible to forecast possible risk of BGR two days earlier than when the AWS-observed weather data were used as input data for BGRcast.

In conclusion, it was found in this study that the UM-predicted weather data released by KMA were useful for their use in plant disease forecast. The UM-predicted weather data could be applicable to various disease forecast models other than BGRcast. A major advantage of using the numerical weather prediction data is that disease forecast information should be available prior to actual infection by pathogen (Firanj Sremac et al., 2018), which allows crop growers to take better options of disease control measures including both protective and curative chemicals (Beresford and Manktelow, 1994). By incorporating protective measures in plant disease management, it is possible to suppress development of fungicide resistance in pathogen population (Darolt et al., 2016; Hollomon, 2015; Horsfield et al., 2010). Besides, the UM-predicted weather data are available free of charge (Magarey and Isard, 2017) throughout the whole country of Korea at the spatial resolution of 1.5 km × 1.5 km.

Acknowledgments

This work was funded by the Korea Meteorological Administration Research and Development Program under Grant KMI2018-04511.

Figures
Fig. 1. Geographical locations of 29 rice paddy fields and their nearest automated weather stations in Korea. The distances between rice paddy fields and their nearest automated weather stations (AWS) are listed in .
Fig. 2. The relationship between the Unified Model (UM)-predicted and the automated weather stations-observed daily minimum temperature at 29 locations of paddy fields during the period from May 5 to October 31 in 2014 and 2015. The UM-predicted weather data for one day at all 29 locations were missing in the plot. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 3. The absolute differences between the Unified Model (UM)-predicted and the automated weather stations-observed daily minimum temperature at each location of rice paddy field during the period from May 5 to October 31 in 2014 and 2015 and the root mean squared error (RMSE) of the differences. The UM-predicted weather data for one day at all 29 locations were missing in the plot. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 4. The relationship between the Unified Model (UM)-predicted and the automated weather stations (AWS)-observed daily average relative humidity at 29 locations of paddy fields during the period from May 5 to October 31 in 2014 and 2015. The AWS-observed relative humidity data for 31 days at all 29 locations were missing in the plot. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 5. The differences between the Unified Model (UM)-predicted and the automated weather stations (AWS)-observed daily average relative humidity at each location of rice paddy field during the period from May 5 to October 31 in 2014 and 2015 and the root mean squared error (RMSE) of the differences. The AWS-observed relative humidity data for 31 days at all 29 locations were missing in the plot. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 6. Graphic presentation of the two-way contingency table analysis on the Unified Model (UM)-predicted and the automated weather stations (AWS)-observed daily average relative humidity (RH) at 29 locations of paddy fields during the period from May 5 to October 31 in 2014 and 2015. Data points were categorized into four groups with reference to 80% RH, which is the threshold of relative humidity for bacterial grain rot development. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 7. The relationships between the Unified Model (UM)-based and the automated weather stations (AWS)-based estimates of Cinc and Cinf, which were calculated assuming that heading dates of rice plants varied from July 15 to September 9. Weather data for the period from May 5 to October 31 in 2014 and 2015 were used to run BGRcast. (A) UM06-based Cinc vs. AWS-based Cinc. (B) UM18-based Cinc vs. AWS-based Cinc. (C) UM06-based Cinf vs. AWS-based Cinf. (D) UM18-based Cinf vs. AWS-based Cinf. UM06, UM-data released at 0600 Universal Time Coordinated (UTC); UM18, UM-data released at 1800 UTC.
Fig. 8. Temporal changes of Cinc and Cinf over the rice growing season in 2014 for two locations, Yangju and Goseong. The automated weather stations (AWS)-observed (gray) and the UM06-predicted (red) weather data were used as input data for BGRcast to estimate Ci, Cinc and Cinf.
Tables

Heading dates of rice varieties cultivated at 29 locations in 2014, geographical locations of paddy fields and distances from the paddy fields to the nearest automated weather stations

Site IDa Location Latitude Longitude Distance (km) Cultivar Heading date
1 Goseong 34.9905 128.3309 0.04 Honong 22 Aug
2 Seocheon 36.0622 126.7043 0.06 Ilpum 15 Aug
3 Ganghwa 37.7074 126.4463 0.36 Chuchung 18 Aug
4 Yangju 37.8312 126.9905 0.79 Daean 16 Aug
5 Hampyeong 35.0602 126.5264 1.52 Ilmi 20 Aug
6 Jinan 35.7619 127.4375 1.87 Shindongjin 19 Aug
7 Jangheung 34.6888 126.9195 2.22 Hopyoung 22 Aug
8 Gimhae 35.2300 128.8910 2.5 Yonghojinmi 23 Aug
9 Yeonggwang 35.2837 126.4778 2.8 Saeilmi 21 Aug
10 Gimcheon 36.0813 128.1016 2.86 Ilpum 15 Aug
11 Jeongeup 35.5632 126.8661 3.04 Hwangeumnuri 20 Aug
12 Yeoncheon 38.0265 127.0781 3.52 Daean 19 Aug
13 Buan 35.7295 126.7166 3.54 Saenuri 23 Aug
14 Yeongam 34.7998 126.7013 3.79 Saenuri 19 Aug
15 Yeoju 37.2688 127.6396 3.82 Chuchung 15 Aug
16 Taean 36.7585 126.2964 4.29 Chuchung 13 Aug
17 Anseong 37.0038 127.2500 4.88 Chuchung 17 Aug
18 Damyang 35.3102 126.9727 4.91 Ilmi 19 Aug
19 Miryang 35.4915 128.7441 5.08 Ilmi 22 Aug
20 Gunsan 36.0053 126.7614 5.13 Hopum 14 Aug
21 Hapcheon 35.5650 128.1699 5.61 Chilbo 11 Aug
22 Gangneung 37.8046 128.8554 5.79 Odae 6 Aug
23 Goheung 34.6182 127.2757 5.96 Shindongjin 24 Aug
24 Icheon 37.2640 127.4842 7.84 Chuchung 15 Aug
25 Uiryeong 35.3226 128.2881 7.89 Ilmi 28 Aug
26 Hwaseong 37.1956 126.8201 8.9 Chuchung 20 Aug
27 Gumi 36.1300 128.3200 11.66 Ilpum 15 Aug
28 Pyeongtaek 36.9922 127.1124 11.91 Samgwang 13 Aug
29 Goyang 37.6343 126.8917 17.67 Chuchung 15 Aug

aSite ID is the number in ascending order of the distance between individual paddy fields and the nearest automated weather stations.


Two-way contingency table analysis on concurrence of BGRcast warnings based on the UM-predicted and the AWS-observed weather dataa

Indexb Warning for the pre-heading spra
(Cinc ≥ 0.3)
Warning for the post-heading spray
(Cinf ≥ 0.5)

UM 06 UM 18 UM 06 UM18
Hit 1,131 1,124 678 668
Miss 21 28 34 44
False alarm 44 52 52 95
Correct negative 2,110 2,102 2,542 2,499
POD (%) 98.18 97.57 95.22 93.82
FAR (%) 3.74 4.42 7.12 12.45
ACC (%) 98.03 97.58 97.40 95.80
Bias score 1.02 1.02 1.03 1.07

UM, Unified Model; AWS, automated weather stations.

aBGRcast warnings were determined by varying heading dates of rice plants from July 15 to September 9 for 29 locations (The 29 sites are listed in Table 1) of paddy fields in 2014 and 2015.

bHit, miss, false alarm, and correct negative are relative frequency that event occurred in both the observed and the predicted, event occurred in the observed but not in the predicted, event did not occur in the observed but occurred in the predicted, and event did not occur in both the observed and the predicted, respectively. POD, FAR, CSI, and ACC indicate the probability of detection, false alarm ratio, critical success index, and accuracy, respectively. POD = Hit/(Miss + Hit); FAR = False alarm/(False alarm + Hit); ACC = (Correct negative + Hit)/(Correct negative + Miss + False alarm + Hit); and Bias = (Hit + False alarm)/(Hit + Miss).


The BGRcast-estimated conduciveness of weather conditions during the inoculum build-up phase (Cinc) and the infection phase (Cinf), and dates of warning based on the conduciveness for bacterial grain rot development at 29 locations of rice paddy fields in 2014

Warning group Site ID Location Heading date Inoculum build-up phase Infection phase

Duration Cinc Warning date Duration Cinf Warning date
No warning 4 Yangju 16 Aug 24 Jul–12 Aug 0.09 - 13 Aug–19 Aug 0.00 -
6 Jinan 19 Aug 27 Jul–15 Aug 0.05 - 16 Aug–22 Aug 0.00 -
7 Jangheung 22 Aug 30 Jul–18 Aug 0.25 - 19 Aug–25 Aug 0.36 -
10 Gimcheon 15 Aug 23 Jul–11 Aug 0.16 - 12 Aug–18 Aug 0.00 -
12 Yeoncheon 19 Aug 27 Jul–15 Aug 0.15 - 16 Aug–22 Aug 0.00 -
15 Yeoju 15 Aug 23 Jul–11 Au 0.19 - 12 Aug–18 Aug 0.00 -
19 Miryang 22 Aug 30 Jul–18 Aug 0.12 - 19 Aug–25 Aug 0.31 -
21 Hapcheon 11 Aug 19 Jul–7 Aug 0.22 - 8 Aug–14 Aug 0.00 -
22 Gangneung 6 Aug 14 Jul–2 Aug 0.04 - 3 Aug–9 Aug 0.36 -
24 Icheon 15 Aug 23 Jul–11 Aug 0.12 - 12 Aug–18 Aug 0.03 -
25 Uiryeong 28 Aug 5 Aug–24 Aug 0.07 - 25 Aug–31 Aug 0.00 -
26 Hwaseong 20 Aug 28 Jul–16 Aug 0.28 - 17 Aug–23 Aug 0.00 -
27 Gumi 15 Aug 23 Jul–11 Aug 0.22 - 12 Aug–18 Aug 0.00 -
29 Goyang 15 Aug 23 Jul–11 Aug 0.17 - 12 Aug–18 Aug 0.00 -
Warning at the pre-heading stage (Cinc) 1 Goseong 22 Aug 30 Jul–18 Aug 0.69 18 Aug 19 Aug–25 Aug 0.43 -
2 Seocheon 15 Aug 23 Jul–11 Aug 0.83 11 Aug 12 Aug–18 Aug 0.00 -
3 Ganghwa 18 Aug 26 Jul–14 Aug 0.40 14 Aug 15 Aug–21 Aug 0.00 -
5 Hampyeong 20 Aug 28 Jul–16 Aug 0.54 16 Aug 17 Aug–23 Aug 0.13 -
8 Gimhae 23 Aug 31 Jul–19 Aug 0.45 19 Aug 20 Aug–26 Aug 0.41 -
9 Yeonggwang 21 Aug 29 Jul–17 Aug 0.49 17 Aug 18 Aug–24 Aug 0.04 -
11 Jeongeup 20 Aug 28 Jul–16 Aug 0.53 16 Aug 17 Aug–23 Aug 0.10 -
13 Buan 23 Aug 31 Jul–19 Aug 0.30 19 Aug 20 Aug–26 Aug 0.16 -
14 Yeongam 19 Aug 27 Jul–15 Aug 0.66 15 Aug 16 Aug–22 Aug 0.00 -
16 Taean 13 Aug 21 Jul–9 Aug 0.41 9 Aug 10 Aug–16 Aug 0.00 -
17 Anseong 17 Aug 25 Jul–13 Aug 0.39 13 Aug 14 Aug–20 Aug 0.00 -
18 Damyang 19 Aug 27 Jul–15 Aug 0.67 15 Aug 16 Aug–22 Aug 0.01 -
20 Gunsan 14 Aug 22 Jul–10 Aug 1.28 10 Aug 11 Aug–17 Aug 0.00 -
23 Goheung 24 Aug 1 Aug–20 Aug 0.54 20 Aug 21 Aug–27 Aug 0.20 -
28 Pyeongtaek 13 Aug 21 Jul–9 Aug 0.66 9 Aug 10 Aug–16 Aug 0.00 -
References
  1. Ashfaq M, Mubashar U, Haider MS, Ali M, Ali A, and Sajjad M. 2017. Grain discoloration: an emerging threat to rice crop in Pakistan. J Anim Plant Sci 27: 696-707.
  2. Beresford RM, and Manktelow DWL. 1994. Economics of reducing fungicide use by weather-based disease forecasts for control of Venturia inaequalis in apples. N Z J Crop Hortic Sci 22: 113-120.
    CrossRef
  3. Bourke PMA. 1970. Use of weather information in the prediction of plant disease epiphytotics. Annu Rev Phytopathol 8: 345-370.
    CrossRef
  4. Branislava L, Mihailović DT, Radovanović S, Balaž J, and Ćirišan A. 2007. Input data representativeness problem in plant disease forecasting models. Q J Hung Meteorol Serv 111: 199-208.
  5. Bregaglio S, Donatelli M, Confalonieri R, Acutis M, and Orlandini S. 2011. Multi metric evaluation of leaf wetness models for large-area application of plant disease models. Agric For Meteorol 151: 1163-1172.
    CrossRef
  6. Brown A, Milton S, Cullen M, Golding B, Mitchell J, and Shelly A. 2012. Unified modeling and prediction of weather and climate: a 25-year journey. Bull Am Meteorol Soc 93: 1865-1877.
    CrossRef
  7. Chakraborty S, Ghosh R, Ghosh M, Fernandes CD, Charchar MJ, and Kelemu S. 2004. Weather-based prediction of anthracnose severity using artificial neural network models. Plant Pathol 53: 375-386.
    CrossRef
  8. Collins SN, James RS, Ray P, Chen K, Lassman A, and Brownlee J. 2013. Grids in numerical weather and climate models. In: Climate change and regional/local responses, eds. by Y. Zhang, and P. Ray , pp. 111-128. Intech, Rijeka, Croatia.
  9. Cullen MJP, and Davies T. 1991. A conservative split-explicit integration scheme with fourth-order horizontal advection. Q J R Meteorol Soc 117: 993-1002.
    CrossRef
  10. Darolt JC, Rocha Neto AC, and Di Piero RM. 2016. Effects of the protective, curative, and eradicative applications of chitosan against Penicillium expansum in apples. Braz J Microbiol 47: 1014-1019.
    Pubmed KoreaMed CrossRef
  11. De Wolf ED, and Isard SA. 2007. Disease cycle approach to plant disease prediction. Annu Rev Phytopathol 45: 203-220.
    Pubmed CrossRef
  12. Do KS, Kang WS, and Park EW. 2012. A forecast model for the first occurrence of Phytophthora blight on chili pepper after overwintering. Plant Pathol J 28: 172-184.
    CrossRef
  13. Duthie JA. 1997. Models of the response of foliar parasites to the combined effects of temperature and duration of wetness. Phytopathology 87: 1088-1095.
    Pubmed CrossRef
  14. Fernandes JMC, Pavan W, and Sanhueza RM. 2014. In: SISALERT: a generic web-based plant disease forecasting system, eds. by M. Salampasis, and A. Matopoulos , pp. 225-233. Proceedings of the 5th International Conference on Information and Communication Technologies for Sustainable Agri-production and Environment (HAICTA 2011). CEURWS, Aachen, Germany.
  15. Firanj Sremac A, Lalić B, Marčić M, and Dekić L. 2018. Toward a weather-based forecasting system for fire blight and downy mildew. Atmosphere 9: 484.
    CrossRef
  16. Gleason ML, Duttweiler KB, Batzer JC, Taylor SE, Sentelhas PC, Monteiro JEBA, and Gillespie TJ. 2008. Obtaining weather data for input to crop disease-warning systems: leaf wetness duration as a case study. Sci Agric 65: 76-87.
    CrossRef
  17. González-Domínguez E, Armengol J, and Rossi V. 2014. Development and validation of a weather-based model for predicting infection of loquat fruit by Fusicladium eriobotryae. PLoS ONE 9: e107547.
    Pubmed KoreaMed CrossRef
  18. Ham JH, Melanson RA, and Rush MC. 2011. Burkholderia glumae: next major pathogen of rice?. Mol Plant Pathol 12: 329-339.
    Pubmed KoreaMed CrossRef
  19. Hirschi M, Spirig C, Weigel AP, Calanca P, Samietz J, and Rotach MW. 2012. Monthly weather forecasts in a pest forecasting context: downscaling, recalibration, and skill improvement. J Appl Meteorol Climatol 51: 1633-1638.
    CrossRef
  20. Hollomon DW. 2015. Fungicide resistance: facing the challenge. Plant Prot Sci 51: 170-176.
    CrossRef
  21. Horsfield A, Wicks T, Davies K, Wilson D, and Paton S. 2010. Effect of fungicide use strategies on the control of early blight (Alternaria solani) and potato yield. Australas Plant Pathol 39: 368-375.
    Pubmed CrossRef
  22. Huber L, and Gillespie TJ. 1992. Modeling leaf wetness in relation to plant disease epidemiology. Annu Rev Phytopathol 30: 553-577.
    CrossRef
  23. Jeong Y, Kim J, Kim S, Kang Y, Nagamatsu T, and Hwang I. 2003. Toxoflavin produced by Burkholderia glumae causing rice grain rot is responsible for inducing bacterial wilt in many field crops. Plant Dis 87: 890-895.
    Pubmed CrossRef
  24. Kang WS, Hong SS, Han YK, Kim KR, Kim SG, and Park EW. 2010. A web-based information system for plant disease forecast based on weather data at high spatial resolution. Plant Pathol J 26: 37-48.
    CrossRef
  25. Kim J, Kang Y, Kim J-G, Choi O, and Hwang I. 2010. Occurrence of Burkholderia glumae on rice and field crops in Korea. Plant Pathol J 26: 271-272.
    CrossRef
  26. Kim S, Kim HM, Kay JK, and Lee S-W. 2015. Development and evaluation of the high resolution limited area ensemble prediction system in the Korea Meteorological Administration. Atmosphere 25: 67-83 (in Korean)..
    CrossRef
  27. Kurita T. 1967. On the pathogenic bacterium of bacterial grain rot of rice. Ann Phytopathol Soc Jpn 33: 111 (in Japanese)..
  28. Lalic B, Francia M, Eitzinger J, Podraščanin Z, and Arsenić I. 2016. Effectiveness of short-term numerical weather prediction in predicting growing degree days and meteorological conditions for apple scab appearance. Meteorol Appl 23: 50-56.
    CrossRef
  29. Lee D-B, and Chun H-Y. 2015. Development of the Korean Peninsula-Korean Aviation Turbulence Guidance (KP-KTG) system using the Local Data Assimilation and Prediction System (LDAPS) of the Korea Meteorological Administration (KMA). Atmosphere 25: 367-374 (in Korean)..
    CrossRef
  30. Lee YH, Ko S-J, Cha K-H, and Park EW. 2015. BGRcast: a disease forecast model to support decision-making for chemical sprays to control bacterial grain rot of rice. Plant Pathol J 31: 350-362.
    Pubmed KoreaMed CrossRef
  31. Magarey RD, and Isard SA. 2017. A troubleshooting guide for mechanistic plant pest forecast models. J Integr Pest Manag 8: 3.
  32. Magarey RD, Seem RC, Russo JM, Zack JW, Waight KT, Travis JW, and Oudemans PV. 2001. Site-specific weather information without on-site sensors. Plant Dis 85: 1216-1226.
    Pubmed CrossRef
  33. Magarey RD, Sutton TB, and Thayer CL. 2005. A simple generic infection model for foliar fungal plant pathogens. Phytopathology 95: 92-100.
    Pubmed CrossRef
  34. Mesinger F. 1981. Horizontal advection schemes of a staggered grid: an enstrophy and energy-conserving model. Mon Weather Rev 109: 467-478.
    CrossRef
  35. Mihailović DT, Koči I, Lalić B, Arsenić I, Radlović D, and Balaž J. 2001. The main features of BAHUS - biometeorological system for messages on the occurrence of diseases in fruits and vines. Environ Model Softw 16: 691-696.
    CrossRef
  36. Nandakumar R, Shahjahan AKM, Yuan XL, Dickstein ER, Groth DE, Clark CA, Cartwright RD, and Rush MC. 2009. Burkholderia glumae and B. gladioli cause bacterial panicle blight in rice in the southern United States. Plant Dis 93: 896-905.
    Pubmed CrossRef
  37. Olatinwo R, and Hoogenboom G. 2014. Weather-based pest forecasting for efficient crop protection. In: Integrated pest management: current concepts and ecological perspective, ed. by DP. Abrol , pp. 59-78. Academic Press, Amsterdam, Netherlands.
    CrossRef
  38. Orlandini S, Magarey RD, Park EW, Sporleder M, and Kroschel J. 2017. Methods of agroclimatology: modeling approaches for pests and diseases. In: Agronomy monograph, No. 60. Agroclimatology: linking agriculture to climate, eds. by JL. Hatfield, MVK. Sivakumar, and JH. Prueger , pp. 1-36. American Society of Agronomy, Madison, WI, USA.
  39. Park EW, Seem RC, Gadoury DM, and Pearson RC. 1997. DMCAST: a prediction model for grape downy mildew development. Vitic Enol Sci 52: 182-189.
  40. Russo JM. 2000. Weather forecasting for IPM. In: Emerging technologies for integrated pest management: concepts, research, and implementation, eds. by GG. Kennedy, and TB. Sutton , pp. 453-473. American Phytopathological Society, APS Press, St. Paul, MN, USA.
  41. Sokal RR, and Rohlf FJ. 1973. Introduction to biostatistics . W. H. Freeman, San Francisco, CA, USA. 368 pp.
  42. Staniforth A, Melvin T, and Wood N 2014. Gungho! a new dynamical core for the unified model. , pp. 15-29. European Centre for Medium-Range Weather Forecasts, Reading, UK.
  43. Walters D, Baran AJ, Boutle I, Brooks M, Earnshaw P, Edwards J, Furtado K, Hill P, Lock A, Manners J, Morcrette C, Mulcahy J, Sanchez C, Smith C, Stratton R, Tennant W, Tomassini L, Van Weverberg K, Vosper S, Willett M, Browse J, Bushell A, Carslaw K, Dalvi M, Essery R, Gedney N, Hardiman S, Johnson B, Johnson C, Jones A, Jones C, Mann G, Milton S, Rumbold H, Sellar A, Ujiie M, Whitall M, Williams K, and Zerroukat M. 2019. The Met Office Unified Model Global Atmosphere 7.0/7.1 and JULES Global Land 7.0 configurations. Geosci Model Dev 12: 1909-1963.
    CrossRef
  44. Webster RK, and Gunnell PS. 1992. Compendium of rice diseases . American Phytopathological Society, St. Paul, MN, USA. 62 pp.


April 2020, 36 (2)
  • DOAJ
  • ORCID