A multivariate multiscalar approach for drought quantification: Development and evaluation of the novel drought monitoring index using climatic and hydrological variables

Abstract

Droughts are long lasting and critical environmental concerns that cause severe effects on water and agricultural resources worldwide. The use of a single variable for detecting and monitoring droughts may not effectively represent the complex and multifaceted nature of droughts. Nevertheless, most of the existing drought indices rely on only one or two climatic variables, limited their ability to effectively represent the hydro-climatic interaction for assessing drought behavior. Therefore, there is a need to use a multiple climatic and hydrological factors for quantifying droughts and ensuring sustainable water management practices. To address this limitation, this research has proposed a novel multivariate index called the Multivariate Multiscalar Standardized Combinative Index for Climatic Extremes (MMSCICE), which combines seven critical climatic and hydrological factors such as precipitation (mm), temperature (degrees C), wind speed (m/s), relative humidity (%), groundwater level (m), streamflow (m3/s), and lake water level (m)) for the Beys & cedil; ehir region of T & uuml;rkiye. Each variable is standardized using the K-Component Gaussian Mixture Distribution (K-CGMD) model, which is identified by the Bayesian Information Criterion (BIC). K-CGMD consistently shows the highest flexibility in modeling multimodal and skewed hydroclimatic data by achieving lower BIC values. To further validate the MMSCICE model in terms of its ability to provide accurate forecasts, the Autoregressive Integrated Moving Average (ARIMA) and the Multilayer Perceptron (MLP) models are employed and compared over various time scales against the commonly used indices such as the Standardized Precipitation Index (SPI), the Standardized Precipitation Temperature Index (SPTI), and the Standardized Composite Index for Climate Extremes (SCICE). The study employed Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) as quality measures for comparison purposes. The results show that proposed index (MMSCICE) achieves the lowest error values (RMSE) compared to other comparative indices. For instance, MMSCICE gives the lowest error values (RMSE) (0.662 and 0.176) compared to SPI (0.878, 0.297), SPTI (0.838, 0.249), and SCICE (0.664, 0.235) using ARIMA and MLP models, respectively across the training phase at a 1-month time scale. Similarly, in the testing phase, MMSCICE yields the lowest error values (RMSE) (0.57099, 0.1153) compared to SPI (0.9407, 0.8464), SPTI (0.9493, 0.8464), and SCICE (0.6594, 0.6538) using ARIMA and MLP models, respectively. The accuracy and reliability of MMSCICE in characterizing drought behavior are evident in its ability to generate improved results over different time scales (1, 3, 6, 9, 12, 24) using different forecasting model. Thus, MMSCICE provides a more reliable and comprehensive tool in identifying and characterizing drought behavior in terms of its joint characteristics.