干旱区生态环境知识资源中心

Relationships between precipitation and elevation are difficult to model for mountainous regions, due to complexities in topography and moisture sources. Attempts to model these relationships need to be tested against long-term location specific meteorological data, and hence require a case-study approach. This study uses artificial neural networks to model these relationships for the Middle of Zagros region, in semi-arid western Iran. Precipitation data for the region were collected for 1995-2007. Annual precipitation was designated as the target variable for the network, which additionally included variables significantly related to precipitation for the region, including longitude, latitude, elevation, slope, distance from the ridge, and relative distance from moisture. Long-term changes in annual precipitation for the region are investigated for 1961-2010. The artificial neural network (ANN) model explains 76% of the spatial variability of precipitation in the Middle Zagros. Precipitation predominantly increases with elevation on the windward slope, to a maximum height of 2500 m.asl, and thereafter either remains constant or decreases slowly to the ridge. Precipitation in the region has decreased significantly over the study period, with fluctuations driven by AO, NAO, ENSO and variability in the strength of pressure centers. Spectral analysis reveals significant oscillations of 2-4 and 5 yr periods, which correspond temporally with cycles in macro-scale circulation, ENSO and the Mediterranean Low pressure.