Climate scientists are gaining an understanding and data of the past and are projecting what the future climate might be like through applying the climate models. A climate model is like a Virtual Earth, it's designed to mimic the real world, so that scientists can forecast future scenarios of climate changes. Climate models are composed of computerized representations of components that represent the atmosphere, ocean, sea, ice, surfaces and other processes. Climate models do not rely on speculation, they describe the climate system with mathematical equations based on the physics and solved with high advanced computers. This research presents a significantly Climate forecasting model using a deep convolutional Long Short-Term Memory (LSTM) to forecast temperatures world widely. New developments in this model include the next-days prediction with Convolutional LSTMs mapping of past climate change to project future climate change since the observed changes. In addition, the model of unsupervised Deep Learning networks is for tackling climate patterns detection problems and the improvements Recurrent Neural Network-RNN architecture by minimization the loss function over multiple sequence steps. Our model is considered one of the best models comparing with others, due to its high testing accuracy (100%).