Abstract: Abstract: China has the largest waterfowl industry, accounting for almost 75% of the waterfowl production in the world. The air temperature in the lion-head goose breeding environment is crucial to the survival of lion-head goose. The air temperature is susceptible to many factors such as relative humidity, illumination intensity, total suspended particulates, etc. So, it is significant to understand timely and accurately the change of the air temperature which can prevent environment deterioration, disease outbreaks, and optimize breeding management. Grasping the trend of the air temperature value timely and accurately is the key to the high-density healthy lion-head goose culture. The traditional prediction methods in temperature have problems such as low prediction accuracy, poor robustness, and poor generalization ability, etc. To improve the prediction accuracy of the air temperature of the lion-head goose breeding environment, a predicting model of air temperature fusing Principal Component Analysis (PCA), Support Vector Regression (SVR) machine, and Autoregressive Moving Average (ARMA) model was proposed for lion-head goose breeding environment. First, the key impact factors of air temperature in lion-head goose breeding were extracted by principal component analysis, which eliminated redundant information between variables and reduced the complexity of the prediction model structure. Therefore, the key impact factors were selected air temperature, relative humidity, illumination intensity, total suspended particulates, respectively. Then, SVR-ARMA model was used to build the nonlinear prediction model between the temperature and these key impact factors for lion-head goose breeding, in which, first, the air temperature was predicted by SVR, then the air temperature prediction values were corrected by the residual prediction value of ARMA model. Because the kernel parameter and the regularization parameter in the SVR training procedure significantly influence forecasting accuracy, a leave-one-out cross-validation method was utilized to optimize the SVR model parameters and biases under a parameter space design. With this model, the air temperature change had been predicted for the lion-head goose breeding environment from July 21st, 2018 to July 30th, 2018 in Shanwei city, Guangdong province. The experimental results showed that the proposed combination prediction model of PCA-SVR-ARMA had a better prediction effect than the standard BP neural network, standard support vector regression machine, PCA-BPNN, PCA-SVR, and PCA-BPNN-ARMA method. And the relative Mean Absolute Percent Error (MAPE), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for the temperature between the PCA-SVR-ARMA and PCA-SVR models were 29.59%, 40.37%, and 60.75% respectively under the same experimental conditions. The relative MAPE, RMSE, and MAE for the temperature between the PCA-SVR and standard SVR models were 31.78%, 15.89%, and 29.45% respectively. The relative MAPE, RMSE, and MAE for the temperature between the PCA-SVR-ARMA and PCA-BPNN-ARMA models were 55.64%, 35.66%, and 55.26% respectively. The relative MAPE, RMSE, and MAE for the temperature between the PCA-BPNN-ARMA and PCA-BPNN models were 43.16%, 30.63%, and 44.16% respectively. The relative MAPE, RMSE, and MAE for the temperature between the PCA-BPNN and BPNN models were 10.34%, 4.80%, and 7.98% respectively. To sum up, the SVR model had a better prediction performance under the condition of small samples, while the residual error correction method based on the ARMA model effectively improved prediction performance. The temperature prediction model based on PCA-SVR-ARMA exhibited best prediction accuracy and generalization performance when compared with other traditional forecasting models. Therefore, the presented model based on PCA-SVR-ARMA could meet the actual demand for accurate forecasting of temperature and provide a reference for environment control in healthy breeding and seedling breeding of lion-head goose. As well as it also could help farmers make decisions and reduce farming risks.