Variation in the Spatial Factors Affecting Precipitation in Relation to the Decadal Changes of Annual Precipitation in Iran

Some mechanisms of climate change, particularly changes in precipitation, are the result of changes in local mechanisms, while some others are caused by the interaction of events on larger scales, e.g., regional, synoptic, hemispherical, or planetary scales. However, in all these changes, the reactions of spatial factors like geographical coordinates (latitude and longitude) and topographic features, including altitude, terrain slope, and terrain aspect, on a local scale can be a proper signal of large-scale changes. In particular, numerous studies have shown that spatial variations, as well as temporal variability of precipitation, are in relation with spatial coordinates (longitude and latitude) and topography (altitude, terrain slope, and terrain aspect). Nevertheless, the fact that the temporal variation of precipitation is in relation with the roles of spatial factors has been neglected.Using the Artificial Neural Network (ANN) technique, the present study aimed to provide inferences about the decadal changes in the overt and covert links of spatial factors with the precipitation climatology of Iran. Thus, using the national network data (Asfazari), 3rd version, the spatial distributions of precipitation for the last four decades were compared based on spatial factors. Also, an attempt was made to show the decadal variation of precipitation in Iran in relation to spatial factors, which could serve as an index of climate change as an essential field of research on precipitation. 

Two datasets were employed to conduct this investigation; the 3rd version of Asfazari Precipitation Dataset and the data of a Digital Elevation Model (DEM) related to Iran. The first dataset with the dimensions of 16801×205×167 and a resolution of 10 km was applied to study the temporal and spatial behaviors of precipitation within Iranian borders. The second dataset with a resolution of 10 km belonged to the US Geological Survey produced via ASTER satellite imagery with a global coverage.Based on the two above-mentioned datasets, the following steps and methods were taken and adopted to conduct the current study:1- The average precipitation for the whole period (1969-2015) was calculated and its spatial relationships were examined. To investigate the variability of decadal precipitation, the average precipitation for each decade up to the decade of 2006-2015 was measured. Thus, the first 6 years (1969-1975) did not fit into the study decades to provide a comparison. Accordingly, the spatial characteristics of precipitation in Iran during the four decades of 1976-1985, 1986-1995, 1996-2005, and 2006-2015 were studied.Precipitation is considered as one of the elements, phenomena, and climatic processes, as well as an important indicator, in climate change tracking. One of the notable features of precipitation is its strong and often nonlinear relationship with geographical coordinates (latitude and longitude) and topographic factors (altitude, slope, and slope direction). There are several ways to study this relationship. In this regard, we can refer to regression methods, control methods, ANN methods, etc. In recent years, the use of regression techniques (for example, Singh et al., 1995; Glazin, 1997; Alijani, 1373; Ghayyur and Masoudian, 1375; Mojarad and Moradifar, 1382; Asakereh, 1384; Razi'i and Azizi, 1387) has been in focus.Modeling the time series of climate like precipitation and chaotic spatial relationships of such nonlinear series are difficult and complex task due to atmospheric dynamics and its nonlinear relationships with spatial variables and since temporal change (variability) of precipitation in a continuous and chaotic system reflects a complex and nonlinear atmospheric behavior in the "geographical space". The spatial analysis showed that the relationships between precipitation and spatial factors had undergone a change on the tempo-spatial scale. Accordingly, complex algorithms, such as ANN methods, were more suitable for modeling these chaotic time series in a broad space like Iran.To study the characteristics of precipitation in Iran and compare the spatial relationships of precipitation in the current research, the spatial distribution of precipitation on the decadal scale and the decadal variability of precipitation were first investigated. Based on the selected spatial-topographic factors in all 16203 cells on the map of Iran as the ANN inputs, a model could be extracted to better fit the data. In this paper, the precipitation in Iran was regarded as the target variable to be compared with the model outputs. 

General characteristics of annual rainfallThe spatial average of precipitation was about 250.5 mm. There was a very large spatial difference of precipitation in Iran. The spatial variability of precipitation was estimated based on geographic coordinates and topographic variables by using the ANN technique. Although the model’s error rate (88809.3) was noticeable, the correlation coefficient (0.95) showed that the estimated spatial distribution pattern of precipitation and the spatial distribution of real precipitation were very similar (90%). The absolute values of the model’s coefficients revealed that longitude, latitude, and altitude played the most important roles, respectively. The terrain aspect played the least important role in justifying precipitation. Decadal changes of precipitationThe average precipitation in the country demonstrated a significant decrease from 268.1 to 220.3 mm from the first to the fourth decade. Nonetheless, the second decade had experienced a relatively significant increase and thus disrupted the general downward trend. The average precipitation anomaly was negative in the last two decades as well. This was evidence of the impact of the decreasing trend of precipitation in all regions of the country. Consequently, in the last two decades, 76.1 and 81% of Iran’s territory had received less precipitation than the long-term average precipitation between 1969 and 2015, respectively. The amounts of precipitation in the models fitted to each decade were compatible with the actual precipitation amounts. Therefore, the role of spatial factors in estimating rainfall had an acceptable capability.Decadal changes in the effects of spatial factorsAssessment of latitude coefficients revealed that both the pattern and coefficient values ​​were corresponding to the first, third, and fourth decades. It seemed that the negative values of latitude increased towards the last decade. For the second decade, which was associated with a relative enhancement in rainfall, the coefficients were different from those of the other decades. In this decade, coefficient variability was higher than those of the other decades. The average longitude coefficients of 10 neurons for the four studied decades were 1.76, 29.35, 0.91, and -1.19, respectively. The average altitude coefficients of neurons for these decades were about -2.87, -7.3, 0.1, and 3.75, respectively. Also, the average slope coefficients for the decades were almost similar to those of the altitude pattern (-2.29, 29.91, 0.3, and -0.22, respectively). However, the degrees of influence (coefficient values) and their signs were highly different for these two factors. Finally, the average coefficients for slope for the mentioned decades were about -0.71, 31.18, 0.34, and -2.83, respectively.

 Conclusion

In this investigation, the diversity of spatial factors, such as geographical coordinates and topographic features, were found to have led to the spatial diversity of climatic elements like precipitation.  In association with the temporal changes of precipitation, spatial factors played different roles in the process. Therefore, despite the relative stability of spatial factors, it could be inferred that these factors played different roles in the context of precipitation changes. To track the roles of geographical coordinates and topographic factors, i.e., altitude, terrain slope, and terrain aspect, in precipitation, the Artificial Neural Network (ANN) model was utilized. The research findings could be presented in two categories as follows