MODELING TEMPERATURE IN THE ECUADORIAN PARAMO THROUGH DEEP LEARNING

This study focuses on predicting temperature using multivariate time series at two locations in Tungurahua province, Ecuador: Mula Corral, situated in the high-altitude grassland ecosystem known as paramo, and Airport Ambato, located in the city center. The prediction integrates key variables such as humidity, precipitation, and wind speed through multivariate neural networks. Six neural network architectures were evaluated: LSTM, Bidirectional LSTM, LSTM with Attention Mechanism, GRU, CNN, and CNN-LSTM. At the Airport Ambato station, the LSTM with Attention Mechanism was the most effective, achieving an RMSE of 0.82 and an R-squared of 0.66. For Mula Corral, the LSTM and GRU networks excelled, with an RMSE of 0.63 and an R-squared of 0.70 for the LSTM, while the GRU achieved an RMSE of 0.89 and an R-squared of 0.60. Conversely, the CNN-LSTM demonstrated the lowest performance, particularly in Mula Corral, with an RMSE of 0.75 and an R-squared of 0.53. These outcomes highlight the superiority of LSTM networks, especially those equipped with an Attention Mechanism, and reveal a preference for Recurrent Neural Networks (RNN) over Convolutional Neural Networks (CNN) for predicting temperatures in sensitive areas such as paramos. The ability to predict temperature with multivariate time series is critical for planning and implementing restoration techniques in the paramos, adapting to climate changes, and maximizing success in the conservation and recovery of these vital ecosystems. The neural network analysis underscores significant climatic differences between the paramo and the city. Mula Corral exhibits lower and more stable temperatures, consistent with the cold, uniform conditions of high-altitude grasslands, whereas the Ambato airport station reflects higher temperatures with greater variability, likely due to urbanization and human activity. These distinctions highlight the importance of accurate neural network models for environmental management, strategic planning, and climate change adaptation.