Predicting new COVID-19 cases was, and still is, of paramount importance to decision makers in many countries. Researchers are constantly proposing new models to forecast the future tendencies of this pandemic, among which long short-term memory (LSTM) artificial neural networks have shown relative superiority compared to other forecasting techniques. Due to COVID-19 transmission nature, e.g., sneezing, coughing, and physical contract, researchers have explored the correlation between the spread of COVID 19 and exogenous factors, specifically weather features, to improve the forecasting models. However, researchers have reported contradictory results regarding the incorporation of weather features into their COVID-19 forecasting models. Consequently, this study compares uni-variate with bi- and multi-variate LSTM forecasting models to predict COVID-19 cases, of which the latter models consider weather features. LSTM models were used to forecast COVID-19 cases in the six Gulf Cooperation Council (GCC) countries: the United Arab Emirates, Saudi Arabia, Kuwait, Bahrain, Qatar, and Oman as a group of countries that share similar weather conditions. Moreover, the LSTM models were used to forecast COVID-19 cases in cold-weather countries, such as the United Kingdom, Germany, Canada, and Norway. The root-mean-squared error (RMSE) and coefficient of determination (R2 ) were used to measure the accuracy of the LSTM forecasting models. According to the conducted statistical comparisons, the R2 and RMSE values for cold-weather countries were significantly improved with the inclusion of weather features in the forecasting models. On the other hand, despite similar weather conditions in the GCC countries, the improvements gained by including weather features were insignificant considering the RMSE values and marginally significant considering the R2 values. Therefore, this study concludes that bi variate and multi-variate LSTM models are crucial for forecasting COVID-19 cases in cold weather countries. Ho