عنوان مقاله [English]
Drought is a critical climate condition affecting many places on Earth. Drought severity is often measured using a combination of different variables including rainfall, temperature, humidity, wind, soil moisture, and steam flow. During the last decades, Iran has suffered from drought conditions and it may suffer more in future. The frequent occurrence of drought in Iran is mainly due to lack of sufficient precipitation and improper water management system. Drought is often categorized into three types: meteorological, agricultural, and hydrological. There are various methods for measuring and quantifying drought severity. The most commonly used ones are Palmer Drought Severity Index (PDSI) and Standardized Precipitation Index (SPI). Remotely sensed data can also be used for monitoring drought condition. The most widely used ones are Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), Vegetation Condition Index (VCI), Temperature Vegetation Index (TVX) and NDVI deviation Index (DEV). Neural Network (NN) and Autoregressive Integrated Moving Average (ARIMA) are two of the most widely applied methods for modeling and monitoring drought severity indices.
In this paper, monthly time series data (2000 to 2014) of three remotely sensed indices (i.e., NDVI, VCI, and TVX) and one meteorological index (i.e., SPI) were applied for modeling drought severity. In addition, the NN and ARIMA were developed for modeling these indices.
Materials & Methods
Data used in this paper were the time series of NDVI, VCI, TVX, and SPI. The study area in this paper was Arak, center of Markazi province. It has cold and wet winters with warm and dry summers. ARIMA and NN were employed for modeling indices.
ARIMA model is generally derived from three basic time series models: Autoregressive (AR), Moving Average (MA), and Autoregressive Moving Average (ARMA). These basic models are used with static time series, i.e., they have constant mean and covariance in relation to time.
Usually, NN method has three layers. The first layer or the input layer introduces data to network. Input data is processed in the second layer or the hidden layer. Finally, the output layer produces the results of the input data. In this paper, single hidden layer feed forward network, which is the most widely utilized NN form, was employed for modeling indices.
Results & Discussion
After implementing NN and ARIMA models on the time series data, the performance of the models was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The RMSE obtained by NN and used for modeling NDVI, VCI, TVX, and SPI indices of Arak were 0.1944, 0.2191, 0.1295, and 0.2990, respectively. In addition, RMSE obtained from ARMIA, and used for modeling these indices were 0.0770, 37.2318, 0.2658, and 1.3370. In another experiment, the correlation between remotely sensed indices and SPI was studied. Among the remotely sensed indices, TVX shows the most powerful correlation with SPI.
In the present study, drought condition in the central region of Markazi province was studied during the 2000 to 2014 period. We used the time series of remotely sensed data (such as LST and NDVI) and meteorological data (such as SPI). Then TVX, VCI, and DEV indices were extracted from NDVI and LST data. NN and ARIMA were applied for modeling time series data. Based on the findings, it is concluded that NN is more successful and efficient than ARIMA for this study area. In addition, TVX, which is built based on NDVI and LST, had the most powerful correlation with SPI. This issue implies that both vegetation index and temperature index had an important role in modeling and monitoring drought condition.