Abstract: A convolutional neural network-based long short-term memory network （CNN-LSTM） filling model was proposed to address the problem of soil temperature and humidity sensor data loss caused by harsh environments, battery depletion, hardware failures, and other factors. Using the soil temperature and humidity data from the Shandian River Basin in 2019 as experimental data, five models including CNN, LSTM, TCN, CNN-TCN, and CNN-LSTM were selected to fill in the missing data of the soil temperature and humidity sensor network. The Adam algorithm was used to optimize the model, and the coefficient of determination （R²）, mean square root error （RMSE）, and mean absolute error （MAE） index were used to evaluate the results of the model filling. The results showed that using the linear interpolation algorithm to obtain complete data, the R² of the CNN-LSTM model was 0.999 9, which was higher than that of other models. The MAE and RMSE were 0.001 85 and 0.019 70, respectively, which were much lower than those of other models. The K-nearest neighbor interpolation algorithm was used to obtain complete data. The MAE and RMSE of the CNN-LSTM model were 0.000 12 and 0.000 12, respectively, which were much lower than those of other models. The R² was 0.999 4, which was higher than that of the CNN model, and TCN model;the CNN-LSTM model had the best filling effect on missing values in soil temperature and humidity sensor data. The CNN-LSTM model had good feasibility and accuracy in dealing with the problem of missing data from large-scale soil temperature and humidity sensors.

Key words: CNN-LSTM model, soil, temperature and humidity, missing data filling algorithm