The demand for fresh water is ever increasing because of the myriad of needs that are encountered daily in different sectors. Its availability and occurrences result from the quantum of precipitation received during a given period. Even groundwater, which forms a significant source of fresh water in many parts of the world, is recharged by rainfall. Therefore, precipitation is the major component of the hydrologic cycle. One peculiar characteristic of precipitation is that its occurrence is uneven over space and time (Sansom et al. 2017). The precipitation occurrence can lead to flood (drought) when it is excess (scanty). For example, the recent occurrences of devastating floods of 2018 in Kerala (Mishra and Nagaraju 2019) and 2021 in Germany (Fekete and Sandholz 2021) remind how costly precipitation can be when it occurs in excess. On the other extreme, some parts of the United States of America (USA) experienced significant drought during 2020 (Yaddanapudi and Mishra 2022), and 2021 also shows the continuation of the drought in other parts of the USA. It is also known that scanty rainfall leads to meteorological drought (Sajeev et al. 2021; Muthuvel and Mahesha 2021). This drought is often the precursor to agricultural drought (Behrang Manesh et al. 2019) and hydrological drought (Hao et al. 2016). This erratic behaviour of rainfall calls for a proper understanding of rainfall distribution in space and time. Moreover, it is imperative to understand the precipitation trend due to anthropogenic activities and climate alteration in the hydrologic cycle (Zhang et al. 2007; Wu et al. 2013; Yi et al. 2016).
One of the ways to understand the distribution of rainfall patterns over an area of interest is the trend analysis that helps us ascertain how much rainfall amount increases or decreases on a particular time scale. The most widely used nonparametric methods to capture monotonic trends are the Mann-Kendall (MK) and Spearman’s rho (SR) methods (Kendall 1938; Mann 1945; Daniel 1990). Some of the recent studies on the use of MK and SR tests for rainfall trend analysis have been reported in the literature (Kalra and Ahmad 2011; Abghari et al. 2013; Gocic and Trajkovic 2013; Formetta et al. 2016; Güner Bacanli 2017; Hajani et al. 2017; Pandey and Khare 2018; Nikzad Tehrani et al. 2019; Raja and Aydin 2019; Gado et al. 2019). However, the MK and SR approaches are limited to the assumptions of the absence of serial correlation (autocorrelation) of a given time series of a variable, normal distribution of the time series, and the sample size of the data. In order to circumvent the requirement of the limiting assumptions, Şen (2012) proposed the innovative trend analysis (ITA) method for carrying out trend investigation of hydrometeorological variables. The ITA method also does not require the prewhitening of time series data prior to applying it. Since then, several studies on the ITA method have been conducted for trend analysis of hydrometeorological variables in different regions. For example, Güçlü (2018a) extended the ITA method to half time series method (HTSM) that could aid the ITA method to detect trend analysis better.
Similarly, the same author proposed double-ITA (D-ITA) and triple-ITA (T-ITA) approaches to use in tandem with ITA to improve trend detection with stability identification (Güçlü 2018b). In yet another study, innovative triangular trend analysis (ITTA) that aids in detecting partial trends within a given time series was applied using the triangular array after splitting a given time series to a pair of equal length sub-series to make a comparison of trends (Güçlü et al. 2020). The extended version of ITA – the Innovative Polygonal Trend Analysis (IPTA) can not only detect trends captured by the traditional methods but also trend transitions of a time scale (weekly, monthly, etc.) of two equal sub-series derived from the original data (Şen et al. 2019). In extending the IPTA, Ceribasi et al. (2021) proposed the Innovative Trend Pivot Analysis Method (ITPAM) to determine the five risk classes using the inherent relationship in data.
Despite the extended versions of the ITA method in literature, the original ITA method (Şen 2012, 2017a) is still widely used for trend analysis of hydrometeorological variables, as evident from recent literature. For instance, Harka et al. (2021) carried out a comparative study of MK, and ITA approaches to detect rainfall trends in Ethiopia's Upper Wabe Shebelle River Basin (UWSRB). The ITA test detected both monotonic and non-monotonic trends that could not have been possible with the MK test. In a similar study of the Bumbu watershed, Papua New Guinea, Doaemo et al. (2022) did a comparative study of rainfall trend analysis using linear regression, Mann-Kendall rank statistics, Sen’s Slope, and ITA. In addition, spectral analysis was carried out to remove cyclic components from the rainfall time series. Their findings, however, indicate that all the four methods consistently indicated decreasing trend of annual rainfall. Several other studies on ITA application are reported in the literature (Danandeh Mehr et al. 2021; Şişman and Kizilöz 2021; Mallick et al. 2021; Phuong et al. 2022; Ay 2022). The extended version of ITA - IPTA is the most widely used method to detect trends and trend transitions of different hydrometeorological variables (Şan et al. 2021; Ceribasi and Ceyhunlu 2021; Ahmed et al. 2021; Akçay et al. 2021; Hırca et al. 2022).
The occurrences of flood and drought events in India (Mishra and Nagaraju 2019; Mishra et al. 2021; Jha et al. 2021) cause economic losses and life. Therefore, the need for flood policy (Jameel et al. 2020) and understanding the effect of historical and future drought on crops (Udmale et al. 2020) is necessary for effective water management. There is an urgent need to address these issues because most of the population depends on agriculture as an occupation. In this regard, rainfall trend analysis will help the country with pragmatic decisions and actionable plans in dealing with both floods and drought. The MK test has been widely used to study monthly, seasonal, and annual rainfall trends in India both at the national (Kumar et al. 2010; Nengzouzam et al. 2020; Kaur et al. 2021) and regional levels (Goyal 2014; Gajbhiye et al. 2016; Chatterjee et al. 2016; Meshram et al. 2017; Pandey and Khare 2018; Mehta and Yadav 2021; Gupta et al. 2021).
Recent studies on precipitation trend analysis in India witnessed the ITA method being increasingly applied. Often the ITA method has been compared with MK, SR, and linear regression methods (Sanikhani et al. 2018; Machiwal et al. 2019; Meena et al. 2019; Praveen et al. 2020; Singh et al. 2021a; Saini and Sahu 2021; Aher and Yadav 2021). A few studies are worth mentioning using the ITA and classical trend analysis methods at the national level. The study by Praveen et al. (2020) presented the rainfall trend analysis of all the meteorological sub-divisions of India from 1901 to 2015 at the seasonal and annual scales. It was noticed that the change detection point conducted using the Pettitt test was mostly found to be after 1960 for the meteorological divisions. The application of the MK test revealed the trend to be positive during 1901–1950; however, the trend reduced after 1951. The ITA method detected mostly negative trends even when the MK test detected no trend. Singh et al. (2021) presented a similar study of the same region using gridded rainfall data (1901 to 2019), both seasonal and annual. The ITA method was compared with MK, modified Mann-Kendall (mMK), and the linear regression analysis (LRA) tests.
Interestingly, the ITA method could detect trends beyond traditional approaches. An increasing trend was observed for the monsoon and annual rainfall in the northwest and peninsular India; however, the northeast central portion of the nation experienced a negative trend. Most of the zones, however, experienced decreasing winter rainfall. Extracting the rainfall events from anomalous rainfall time series (1871–2016), Saini and Sahu (2021) carried out a unique (not raw rainfall time series) study for the same meteorological zones of India using the MK, ITA, and LRA methods. Though the study is unique in terms of data input and detailed, refined trend analysis, the ITA test captured other trends similar to the above mentioned studies.
The ITA method can be used to capture the trend in hydrometeorological time series data overcoming the assumptions of traditional trend analysis approaches. However, the requirement of time series homogeneity cannot be neglected (Şen 2012). However, there are limited studies on using the ITA approach on homogeneous rainfall time series in the Indian context. If homogeneity tests are not carried out, there are chances of accepting false trends, which might not have occurred.
Karnataka stands only next to Rajasthan state in India to be the most drought-prone area (Jayasree and Venkatesh 2015), receiving a little over 700 mm of mean annual rainfall. The northern region of the state, Rajasthan and northern-central Maharashtra constitute 72% of India's total pearl millet production (Singh et al. 2017). The region is also home to some of the largest crop-producing districts in Karnataka. The districts in this region lie on the Western Ghats' leeward side, making them drought-prone. Furthermore, nearly 90% of the inhabitants in the semi-arid region of Karnataka - over and above 11 districts are dependent on agriculture as an occupation (Jayasree and Venkatesh 2015).
From the literature, it is evident that no such study on the ITA method for precipitation trend analysis has been reported for this region. Hence, the present investigation is focused on: (a) to carry out homogeneity tests of seasonal and annual rainfall time series of each district within the region, (b) to determine the serial correlation of each time series, and (c) to compare the trend of seasonal and annual rainfall using the MK, mMK, SR, and ITA approaches.