The spatial and temporal stability of temperature correlation network is analyzed based on extreme temperature, El Niňo/La Niňa abnormal year, key areas of temperature change and high frequent year of extreme respectively. Results show that, the connectivity of network decreases by filtering out extreme temperature, property of the network at the inflexion P=0.1 or 0.9, might be defined as the of boundary of extreme temperature. La Niňa network possesses more significan links and higher clustering coefficient than the El Niňo network, which indicates that the former is more communicative and more stable than the latter. And based on these statistical properties, we verified that during La Niňa years predictability is higher compared to El Niňo years and normal years. Furthermore, we also verified that during high frequency years of extreme high temperature predictability is higher than high frequent years of extreme low temperature and normal years. We also come to the conclusion that middle and east tropical Pacific （180°E—80°E，15°N—15°S） are key areas of global temperature changes, and exlernal forces acting on these areas have great effect on global temperature change.