Abstract: Based on the meteorological data of Fuqing Station in Fujian Province and the microclimate data of the facility greenhouses from 2017 to 2019, we used multiple linear regression, BP neural network and extreme learning machine (ELM) to construct the simulation models of the facility microclimate. The indoor microclimate data were retrieved from the meteorological data of 18 meteorological stations from 1990 to 2019. Combined with the high temperature and high humidity grade index of facility tomato, we examined the spatial and temporal pattern of high temperature and high humidity of facility tomato in Fujian Province. The results showed that the maximum and minimum indoor air temperatures were best simulated by multiple linear regression model, with RMSE being 2.19 and 1.20 ℃, and R² being 0.87 and 0.96, respectively. BP neural network method was the best one to simulate the maximum relative humidity, with RMSE being 3.77% and R² being 0.65. Combined with facility crop disaster indicators, the frequency of level-2 high temperature and high humidity disasters in the northern and central regions of Fujian Province significantly increased year by year, while the frequency of level-3 disasters in the central region decreased year by year. The probability of level-2 disasters was high across all regions, the probability of level-1 disasters was small in the north and central regions, and the probability of level-3 disasters was small in the south. The meteorological disaster level of facility tomato was high during May to July, and distributed in the northwest of Fujian. The disaster level was the lowest from August to October, with level 2 in most of the whole province and level- 1 only in some coastal areas. The results provide scientific basis for meteorological layout and environmental regulation of facility tomato.

Key words: greenhouse, microclimate, simulation model, high temperature and humidity disaster, spatial and temporal distribution.