Drought is not a natural calamity, but rather a periodic and climatic occurrence that can occur in every region. Due to the harm droughts impose to the agriculture sector and water supplies in arid and semi-arid regions where water scarcity is a problem, they are frequently regarded as a catastrophe. Drought is one of the most significant natural calamities, particularly for farmers. Drought can happen anywhere in the world since it is triggered by an abnormally low amount of precipitation compared to the average or expected climatic value. Lack of water, plant desiccation, and a drop in surface and subterranean water levels are only some of the insidious secondary repercussions of drought that are harmful to both individuals and the ecosystem. More than 85% of worldwide losses are caused by extreme weather events, with drought being a key factor. Due to drought, the world has lost 722 billion USD and 23 million people in the first decade of the 21st century (Ghozat et al., 2022; Gopalakrishnan, 2013). It is predicted that the global air temperature will rise by 0.78 to 1.5 degrees Celsius (Solomon, S. et al., 2007), which will alter the rainfall patterns and subsequently increase the frequency and severity of droughts (Zarch et al., 2011). This climate change poses significant risks to water resources, environmental sustainability, and social and economic growth (Malik et al., 2020; Rahman et al., 2022). Consequently, monitoring drought and its features, and early warning are essential for mitigating its regional and global effects.
For the study of drought, monitoring systems are utilized. There are a variety of drought classifications based on distinct features. Meteorological drought occurs when the amount of precipitation during a period is less than its typical value. Thus, type of drought is calculated using precipitation-based indices such as SPI or bivariate indices such as SPEI, which are based on precipitation and evapotranspiration. Meteorological drought indices, which are derived from meteorological data, are utilized more frequently than hydrological and agricultural indices due to better data availability, wider spatial distribution, and the ability of atmospheric general circulation models to predict them for future years under various climate change scenarios. Consequently, numerous researchers attempt to establish a connection between these indicators.
(Teimoori et al., 2015) compared meteorological and hydrological drought using the SPI and SSI (Standardized Stream flow Index) indices. The SSI index is calculated similarly to the SPI index, with the exception that former uses the river's flow rate instead of precipitation. According to the findings of this study, there is no perfect match between these two indicators; nevertheless, this match can be improved by incorporating previous time steps. Comparative research of SPI and SWI in the Marand plain of Iran by (Zeinali et al., 2017) found that there is a 1% association between these two indicators and that the subsurface water resources are affected by the drought with a five-month lag. This delay between meteorological and hydrological drought was confirmed by (Maleki nejad & Soleimani-Motlaq, 2011) in Chaghalvandi Basin (located in Lorestan province) and (Salahi et al., 2018) in Marand plain located in East Azarbaijan province by comparing SPI and SWI indices, as well as (Abbasi et al., 2016) in Qorveh and Dehgolan plains by comparing SPI and GRI indices. In a 21-year study conducted in Jordan (Yarmouk Basin), Hind (Mohammad et al., 2018) determined that this basin is especially susceptible to regular droughts and that severe drought occurrences has had a negative effect on the groundwater level. (Kubiak-Wójcicka & Bąk, 2018) evaluated meteorological drought and its effect on hydrological drought in the Vistula basin in Poland using the SPI, SWI, and SRI indices over 29 years with three time lags (12-24-48 months), where highest association was found along the Central Vistula and its tributaries, while weakest correlation was found in the foothills. In addition, the results revealed non-climatic elements could have influenced the correlation between coefficients (such as underground reservoirs, urban and industrial consumption). (Aleboali et al., 2016) studied the impacts of drought on groundwater resources in the Kashan plain of Iran over a 19-year period using the SPI index and concluded that excessive exploitation of groundwater resources is the cause of water level decline in addition to drought. The contribution of overharvesting to the decline of aquifer levels has been significantly greater than that of drought.
As a result, it may be argued that the subsurface water supplies are only partially affected by the climatic drought, and that other causes, such as over harvesting, also have a role. However, excessive harvesting results from the expansion of the cultivated area, the rise in temperature, and the subsequent rise in evapotranspiration and water consumption. The SPI index, which is based on a single precipitation variable, cannot account for this factor. The SPEI bivariate index, based on precipitation and evapotranspiration, was introduced by Winslet (Beguería et al., 2014). Other scholars have examined this index and compared it to the SPI index. In a comparative research of two meteorological drought indicators SPI and SPEI in the province of Golestan, (Rezaei Ghaleh & Ghorbani, 2018) found a stronger link between these two indices at stations with a more humid climate. (Pei et al., 2020) calculated the SPI and SPEI indices at 1, 3, 6, and 12-month intervals for 102 weather stations in Inner Mongolia over 1981–2018 and found that as the time window expanded, the difference between the indices decreased. This difference may become negligible over longer time periods. Compared to drought conditions and plant indices, the SPEI index was deemed more suitable for drought monitoring than the SPI index. (Jipkate et al., 2020) compared the SPI and SPEI indices with the SWI index in the Upper Bhima Sub Basin during 2002–2016 and found no direct linear relationship between SWI and SPI and rainfall in the region. (Fung et al., 2020) analyzed the SPI and SPEI indices for 1, 3, and 6 months in Peninsular Malaysia to determine the significance of temperature in causing droughts. In the study of temperature-induced fluctuations on the SPEI index, two indices demonstrated distinct performances; the analysis of the two indices was conducted by analyzing the spatial variations in drought frequency, average drought period, average drought intensity, and average maximum drought. Due to the relevance of temperature increase in the establishment of drought, the SPI index is more ideal for a shorter time window of 1 month, but the SPEI index is more suitable for a longer time window of 3 to 6 months. By comparing the SPI and SPEI indices during a 46-year period in Ankara Province, Turkey, (Danandeh Mehr & Vaheddoost, 2020) determined that whereas the SPEI index has a declining trend, the SPI index does not exhibit a similar pattern. (Babre et al., 2022) assessed groundwater drought periods in the Baltic States using established drought indices and found that meteorological drought indices (SPI, SPEI, and RDI) were substantially linked with groundwater drought conditions in shallow groundwater wells. (Kubicz & Bąk, 2019) investigated the response of groundwater to a multi-month meteorological drought in Poland. They discovered that there was no significant linear relationship between the SPI index in time windows of 6, 12, and 24 months and the standardized groundwater level index (SGI), and they concluded that the level of underground water is influenced by factors other than precipitation. (Leelaruban et al., 2017) analyzed the association between the SPI index and groundwater level changes in the United States for 6, 9, 12, and 24 months intervals. In addition to these indicators, precipitation and average air temperature data was also utilized. The strongest correlation was found in 17 of the 32 wells with SPI-24, while 12 of the wells had a correlation value of 0.6 or greater and the remaining wells were reasonable correlated.
Modeling and determining the link between data has long been a topic of interest, and several research have been undertaken in the field of analyzing and comparing various modeling methods using data mining. (Nourani et al., 2016) employed two data mining techniques (associative rules and decision tree) to identify the relationship between the highest monthly precipitation at synoptic stations in Urmia and Tabriz and the surface temperatures of the Black, Mediterranean, and Red seas. Association rules were used to the observational data to reveal hidden trends and patterns, and decision tree-based techniques and algorithms were utilized to identify and choose the most effective groups. The subsurface water level in Ardabil plain was forecasted by (Sattari et al., 2018) using support vector regression, the M5 decision tree model, and data from 24 piezometric wells over 17 years (1997–2013). The model's inputs were the underground water level in the previous month, the volume of precipitation input to each cell, and the number of feeding wells, while the model's output was the underground water level in the current month. To measure the effectiveness of the model, the correlation coefficient and the mean square error were calculated. The findings showed that both methods were successful in estimating the groundwater table, but the decision tree method's outputs were more transparent and intuitive to use. (Ghorbani, 2016) compared the data mining models M5 decision tree model and k nearest neighbor (KNN) to the IHACRES hydrological model for predicting the monthly river discharge in Arazkouseh station. Due to modeling transparency and the availability of simple regression equations, they confirmed the data mining models' superiority over the hydrological model and determined that the M5 model was the most accurate. In predicting SSI hydrological drought index based on SPI and SPEI meteorological drought indices with machine learning methods, (Shamshirband et al., 2020) determined that the M5 tree model provides superior results to SVR and GEP. White box models, like the decision tree model, are useful because they produce accurate outcomes while also allowing the user to easily identify the impact of various factors using straightforward regression analysis.
None of the research looking into the cause of the droughts in the study area have used data-driven models, and none of them have looked into splitting the area into a semi-deep aquifer. In this study, we examine the relationship between meteorological drought, using SPI and SPEI indices, and fluctuations in the groundwater level over a 6-month period in semi-deep aquifers in Golestan Province of Iran by comparing the linear multivariate regression method and the decision tree method. The primary purpose is to quantify the impact of air temperature and precipitation on hydrological drought while assessing the efficacy of these two techniques.