To improve the early warning capability of rice blast in Hubei Province, and stabilize rice yield while mitigating disaster losses, this study takes the incidence rate of rice blast across Hubei Province as the research object. Based on rice’s primary growth stages (May to August), 48 monthly meteorological indicators are selected from three dimensions: dynamic condition, radiation and thermal condition, and moisture condition. By comparing multiple machine learning approaches, the XGBoost (eXtreme Gradient Boosting) method is selected to build the rice blast prediction model, and the SHAP (SHapley Additive exPlanations) interpretable framework is employed to systematically explore the key meteorological indicators affecting the incidence of rice blast in Hubei Province. The results indicate that: 1) The XGBoost method outperforms traditional machine learning approaches in terms of accuracy, recall, and F1-score. 2) 15 key meteorological indicators affecting rice blast incidence are identified via SHAP analysis, among which the maximum wind gust in May is the critical indicator, contributing 13.19% to the model’s predictive performance. The dominant indicators also vary across different rice growth stages. During the sowing-seedling stage, maximum wind gust and surface pressure are the primary indicators; during the jointing-booting stage, average minimum temperature and surface pressure are the most influential indicators; and during the heading-flowering stage, maximum sustained wind speed and days of rain emerge as the key indicators. 3) Various meteorological indicators exhibit obvious non-linear threshold effects on rice blast incidence, where some indicators (e.g. soil moisture from June to July) exacerbate risk when exceeding thresholds, while others (e.g. surface pressure from May to July) act as negative indicators. 4) Significant interaction effects are revealed among different meteorological indicators on rice blast incidence. For example, strong wind gust in May weaken the inhibitory effect of high surface pressure on disease incidence. This study combined XGBoost with the SHAP interpretable framework for rice blast prediction, which can effectively predict the risk of rice blast in Hubei Province and provide technical support for disease prevention.
本研究因变量为稻瘟病发病率,定义为稻瘟病发病面积占水稻播种总面积之比。鉴于机器学习方法为二分类方法,本研究参考国际应用生物科学中心(Centre for Agriculture and Bioscience International,CABI)的防治启动阈值,以5%作为判定稻瘟病是否发生的临界标准1。具体而言,当某一县域的年度稻瘟病发病率超过5%时,将其判定为发病年,取值为1,反之则为0。
本研究稻瘟病发病面积数据来源于全国农业技术推广服务中心,水稻播种面积来源于《湖北农村统计年鉴》(2016—2025)2。气象数据方面,降雨天数、降雨量、降雨持续、平均高温、平均低温、地面气压与相对湿度数据来源于国家冰川冻土沙漠科学数据中心发布的融合多源数据的中国高分辨率多要素气象驱动产品(ChinaMet)3,该数据集通过融合多源遥感数据、再分析资料及多个气象站的观测数据构建,提供长时序、高时空分辨率的逐日气象数据,为区域气象研究提供坚实的数据支撑。本研究通过月度平均法对ChinaMet数据集逐日数据进行处理,并经标准化变换构建本研究所需的月度气象特征。太阳辐射、潜在蒸散与土壤湿度数据来源于TerraClimate数据集4,该数据集结合气候常数与时间序列再分析数据,构建兼具高空间精细度与月度长时序动态的气象特征数据集,在农业生产、生态分析等领域具有广泛应用。基于其月度分辨率,本研究提取相关气象特征指标,并进行标准化处理后纳入月度气象特征数据集。最大阵风、最大持续风速数据来源于美国国家海洋与大气管理局(National Oceanic and Atmospheric Administration,NOAA)下设的国家环境信息中心(National Centers for Environmental Information,NCEI)发布的全球逐日气象观测(Global Surface Summary of the Day,GSOD)数据集5,该数据集汇聚了来自全球多个气象站的原始观测记录,经过严格的质量控制与一致性检验。基于各气象站点空间分布特性,本研究采用反距离权重法(Inverse distance weighting,IDW)对GSOD数据进行空间插值与标准化处理,以构建各区县月度气象特征数据。
SeckP A, DiagneA, MohantyS, WopereisM C S. Crops that feed the world 7: Rice[J]. Food Security, 2012, 4(1):7-24
[2]
LuoY H, JiangP, XieK, WangF J. Research on optimal predicting model for the grading detection of rice blast[J]. Optical Review, 2019, 26(1):118-123
[3]
KouY J, ShiH B, QiuJ H, TaoZ, WangW M. Effectors and environment modulating rice blast disease: From understanding to effective control[J]. Trends in Microbiology, 2024, 32(10):1007-1020
[4]
KatsantonisD, KadoglidouK, DramalisC, PuigdollersP. Rice blast forecasting models and their practical value: A review[J]. Phytopathologia Mediterranea, 2017, 56(2):187-216
[5]
GuoF F, LiuW C, LuM H, YouF Z, WuB M. Development of two early forecasting models for predicting incidence of rice panicle blast in China[J]. Phytopathology, 2023, 113(3): 448-459
[6]
GopalakrishnanA M, ChellappanG, PatilS G, RathodS, AyyanarK, RamasamyJ, Nagaranai KaruppasamyS, SwaminathanM. Climate-based prediction of rice blast disease using count time series and machine learning approaches[J]. AgriEngineering, 2024, 6(4):4353-4371
[7]
SahooS, SinghaC, GovindA, SharmaM. Leveraging ML to predict climate change impact on rice crop disease in Eastern India[J]. Environmental Monitoring and Assessment, 2025, 197(4):366
[8]
SriwannaK. Weather-based rice blast disease forecasting[J]. Computers and Electronics in Agriculture, 2022, 193: 106685
HuX M, ZhuZ G, JiangZ Q, JiX, ChenL J, ShiH Z. Construction of a prediction model for Xinyang rice blast based on all subsets regression and BP neural network[J]. Hubei Agricultural Sciences, 2025, 64(12):104-109 (in Chinese)
LiuE L, YueY B, LiY, DingJ S, LiuS F, WangG Q, AnZ Y. Study on the correlation between the occurrence area of rice blast and the key meteorological factors and its prediction model in Mangshi, Yunnan Province[J]. China Plant Protection, 2025, 45(10): 49-52, 101 (in Chinese)
WangX, WuB M. Research on the Spatial Distribution Gradient Model of Conidia of Rice Blast Fungus[C]// In:Proceedings of the 2018 Annual Academic Conference of the China Society of Plant Protection and the Award Ceremony of the Plant Protection Science and Technology Award.Xian: China Agricultural Science and Technology Publishing House, 2018:14 (in Chinese)
[17]
QiuJ H, XieJ H, ChenY, ShenZ N, ShiH B, NaqviN I, QianQ, LiangY, KouY J. Warm temperature compromises JA-regulated basal resistance to enhance Magnaporthe oryzae infection in rice[J]. Molecular Plant, 2022, 15(4):723-739
LiuT H, BaiJ J, LyuD P. A preliminary study on the effect of agro-meteorological factors on molecular mechanism of rice blast occurrence[J]. Chinese Journal of Eco-Agriculture, 2016, 24(1):1-7 (in Chinese)
[20]
QiuJ H, LiuZ Q, XieJ H, LanB, ShenZ N, ShiH B, LinF C, ShenX L, KouY J. Dual impact of ambient humidity on the virulence of Magnaporthe oryzae and basal resistance in rice[J]. Plant, Cell & Environment, 2022, 45(12):3399-3411
WanS Q, ChenX, DengA J, LiuZ X, XuR Q, LiuM. Construction of disease indexes and its application in monitoring/forecasting the meteorological grades of early rice panicle blast in Hubei Province[J]. Journal of Huazhong Agricultural University, 2015, 34(3): 76-81 (in Chinese)
JoachimsT. Text categorization with Support Vector Machines: Learning with many relevant features[C]//In: Machine Learning: ECML-98. Berlin, Heidelberg: Springer, 1998: 137-142
[25]
QuinlanJ R. Induction of decision trees[J]. Machine Learning, 1986, 1(1): 81-106
[26]
CybenkoG. Approximation by superpositions of a sigmoidal function[J]. Mathematics of Control, Signals and Systems, 1989, 2(4): 303-314
[27]
CoverT, HartP. Nearest neighbor pattern classification[J]. IEEE Transactions on Information Theory, 1967, 13(1): 21-27
[28]
DudaR O, HartP E, StorkD G. Pattern classification[M]. John Wiley & Sons, 2001
[29]
RishI. An empirical study of the Naive bayes classifier[C]//In: IJCAI 2001 Workshop on Empirical Methods in Artificial Intelligence.Seattle: IJCAI, 2001, 3: 41-46
[30]
ChenT Q, GuestrinC. XGBoost: A scalable tree boosting system[C]//In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2016: 785-794
[31]
LundbergS M, LeeS I. A unified approach to interpreting model predictions[C]//In: Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 4768-4777
[32]
PowersD M W. Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation [J]. Journal of Machine Learning Technologies, 2011, 2(1): 37-63
[33]
NyawoseT, MaswanganyiR C, KhumaloP. A review on the detection of plant disease using machine learning and deep learning approaches[J]. Journal of Imaging, 2025, 11(10): 326
[34]
KoizumiS, KatoH. Dynamic simulation of blast epidemics using a multiple canopy spore dispersal model[C]//In: Rice Blast Modeling and Forecasting.Seoul: International Rice Research Institute, 1991: 75-88
[35]
张淑春,高英.稻瘟病综合防治技术[J].吉林农业, 2008(5):29
[36]
ZhangS C, GaoY. Integrated control technology of rice blast[J]. Agriculture of Jilin, 2008(5):29 (in Chinese)
KeJ. Effects of meteorological factors on spore release, appressorium formation and infection of Magnaporthe grisea in the field[J]. Journal of Guangxi Agricultural College, 1982, 1(1):125-137 (in Chinese)
HuangZ Z, DuY D, HuangD C, ZhongB Y. Prediction of the occurrence area of early rice ear neck blast in Guangdong based on meteorological factors[J]. Guangdong Meteorology, 2020, 42(2):44-48 (in Chinese)
ZhaoY, ZuY Q, LiY. Effects of enhanced UV-B radiation on growth and sporulation quantity of blast isolate Magnaporthe grisea [J]. Journal of Agro-Environment Science, 2010, 29(S1):1-5 (in Chinese)
[43]
SakamotoM. On the facilitated infection of the rice blast fungus, Piricularia oryzae CAV.due to the wind. I[J]. Japanese Journal of Phytopathology, 1940, 10(2-3): 119-126
[44]
QiuJ H, CaoX X, ShiH B, ChenZ T, ZhangX Y, CaoZ J, HuangD M, WenH, ChenY, KouY J. miR444b.2-HsfA1-AOC1 module mediates heat priming-enhanced blast resistance in rice [J]. Proceedings of the National Academy of Sciences of the United States of America, 2025, 122(36): e2505764122
[45]
TaguchiY, ElsharkawyM M, HyakumachiM. Effect of artificially-generated wind on removing guttation and dew droplets from rice leaf surface for controlling rice blast disease[J]. African Journal of Biotechnology, 2015, 14(12): 1039-1047