Abstract: The outbreak of cucumber downy mildew in greenhouses spreads rapidly and can cause significant yield reduction. The existing disease detection technologies mostly start the identification process after the symptoms of the disease appear, resulting in delayed early warning and difficulty in determining the optimal prevention window period. Constructing a disease prediction model that takes into account the growth status of the crops, environmental conditions, and pathogen information is of great practical significance for achieving early warning of downy mildew and promoting green production of greenhouse vegetables. In response to the industry's current situation of delayed warning and passive prevention measures for cucumber downy mildew in greenhouses, this paper proposes a LSTM-BPNN hybrid neural network prediction method that integrates multi-source information of crop growth, environment, and airborne spores to achieve temporal prediction of the severity of cucumber downy mildew in greenhouses. The experiment was conducted in a Venlo-type greenhouse at Jiangsu University, using the cold-resistant and weak-light-tolerant variety "Jinyou 1" cucumber as the experimental material. Different nutrient solution concentrations were set for treatment, and a continuous 66-day observation experiment was carried out. A total of 198 valid samples were obtained. During the experiment, cucumber plant growth phenotypic parameters such as plant height, stem diameter, and leaf area were collected, as well as environmental data such as average temperature and relative humidity in the greenhouse. At the same time, spore-capturing devices were used to capture spores of the downy mildew pathogen in the air, and the spore concentration data were obtained through microscopic counting. The leaf images were processed using the PP-LiteSeg semantic segmentation network, and the total leaf area and lesion area were calculated without loss, thereby calculating the severity of cucumber downy mildew. All samples were preprocessed using maximum value normalization, and the correlation between each factor and the severity of the disease was analyzed using the Pearson correlation coefficient. The results showed that cucumber plant height, stem diameter, and leaf area were negatively correlated with the severity of the disease, with correlation coefficients of -0.37, -0.37, and -0.52 respectively. The better the plant growth condition, the stronger the disease resistance, and the lower the disease severity. The average temperature, average relative humidity, and the number of spores in the air were positively correlated with the severity of the disease, with correlation coefficients of 0.28, 0.41, and 0.38 respectively. Especially in the early stage of the disease (severity < 20%), the correlation coefficient between spore number and disease severity reached 0.63, which is the core indicator for early warning. To reduce the feature dimension, the normalized plant height, stem diameter, and leaf area were constructed as dimensionless comprehensive growth characteristics Q. Combining the memory mining ability of LSTM for time series data and the excellent nonlinear mapping ability of BPNN, a LSTM-BPNN fusion prediction model was built. To address the problem of gradient disappearance of the Sigmoid function for small disease severity samples, the activation function of the output layer of BPNN was replaced with ELU, and the network hyperparameters were optimized. The number of neurons in the LSTM hidden layer was set to 5, and the number of neurons in the BPNN hidden layer was set to 8. The Adam optimizer was used for training. The dataset was divided into training set, validation set, and test set in a 8:1:1 ratio, and the early stopping strategy was introduced to prevent overfitting. The model test results showed that the average absolute error (MAE) of the LSTM-BPNN model was 0.0133, the mean square error (MSE) was 0.0068, the root mean square error (RMSE) was 0.0825, and the determination coefficient (R²) reached 0.942. All evaluation indicators were superior to those of the single LSTM and BPNN models. Compared with the LSTM-CNN and GRU-BPNN hybrid models, this model achieved a 99.40% accuracy rate for 3-day disease prediction and a 95.44% accuracy rate for 7-day prediction in the future. The prediction performance was better than the comparison models. The study found that the prediction error of the model would accumulate with the increase in prediction time, making it suitable for short-term rolling prediction within 7 days. By iteratively outputting disease risks based on daily updated growth, environmental, and spore data, the prediction accuracy could be guaranteed. This study integrates three types of information - host growth resistance, environmental stress conditions, and pathogen spore sources - into a prediction framework. This compensates for the shortcomings of most existing disease warning studies, which either ignore plant growth indicators or pathogen spore information. It provides a precise early warning technical solution for greenhouse cucumber downy mildew, enabling growers to promptly carry out disease prevention and control, reducing pesticide abuse, and minimizing yield losses caused by the disease. It also offers theoretical references and technical support for the intelligent early warning and prevention of other airborne diseases in facility agriculture.