3.1 Characteristics of heavy and low rainfall zone of India
The percentage distribution of various rainfall amount have been shown in Fig. 1a, the rainfall more than 80% i.e. HRZ is shown in Fig.1b, and less than 40% i.e. LRZ is shown in Fig. 1c. Out of 34 met-subdivisions over the Indian main land, on an average 7-8 come under HRZ category and 13-14 are LRZ category (Fig. 1b and Fig. 1c). It may be noted that some met-subdivisions are experiencing both the categories i.e high and low rainfall, however the locations are different. Therefore, we carry rainfall analysis based only on HRZ and LRZ regions instead of state or met-subdivision level. The contribution of various rainfall events to the seasonal rainfall is shown in Fig. 1d and Fig. 1e. It is seen that MR is the major contributing event for both HRZ (43%) and LRZ (48%). Apart from MR event, the heavy rainfall category rainfall events such as RHR, HR, VHR, and EHR have a substantial contribution in seasonal rainfall for the HRZ (Fig. 1d), however, LR and MR have more contribution to the seasonal rainfall for the LRZ (Fig. 1e).
To get a clear picture of the changes in ISMR distribution, it is very important to identify the area where the heavy or more intense rainfall events (HRm) and the total number of dry days (DD) are increased and/or decreased. Here, we have examined four combinations with increasing/decreasing of HRm and DD. Fig. 2 shows the spatial pattern of the different combination for the increased/ decreased in DD and HRm events over India by using Man-Kendall test with a 90% confidence level, very limited study is available for combination of DD and HRm events that’s why we choose 90% confidence level. It is found that the North Western Ghats which shows the increased in seasonal rainfall have experienced decreasing in HRm and increasing in DD, hence, the intense rainfall events are concentrated over smaller region and experience more severe over that region. Apart from that region, Madhya Maharashtra, north-central part of India and some parts of north-west India also has experienced decreasing in HRm and increasing in DD (Fig. 2a), this analysis confirms the result presented in the previous study by Prathipati et al. (2019). Over central India (from east of Gujarat to Odisha), some parts of Uttarakhand, northeast India and the south of Western Ghats, HRm and DD both are increased (Fig. 2b). Furthermore, along east-coast, parts of the Indian peninsular region, parts of north-east India and western Rajasthan, HRm events are increased and DD is decreased (Fig. 2c), while over south Adhara Pradesh coast, parts of Tamil Nadu, Arunachal Pradesh, and parts of Rajasthan both DR and HRm events are decreased (Fig. 2d). It is also observed that major parts of HRZ come under the increased in DD and decreased in HRm events (more than half) and a very small parts are experiencing increased HRm events (Fig. 2a, Fig. 2b); while the major parts of LRZ are experiencing increased in HRm events (Fig. 2b, Fig. 2c). This implies, HRZ and LRZ both are losing their identity based on heavy or more rainfall events. Thus, it is interesting to analyze the change in overall seasonal rainfall for these regions.
For the single change-point detection, the Standard Normal Homogeneity Test (SNHT) is used to calculate the change point detection at the starting and the endpoint of data-series, however, BR and Pettit test are used to detect change point in the middle of the data series (Martinez et al., 2010). But SNHT shows more than one point for the inhomogeneity or change point because of various climatic or non-climatic facto and it gives a poor performance in detection of change-point at the beginning of the series (Toreti et al., 2011). So, BR test is mostly used for any distribution data (Taxak et al., 2014). Fig. 3 shows the BR and Pettit’s test which are used for the single change point detection and multiple change-point detections over HRZ and LRZ. The HRZ shows a less variability in data (Fig. 3a) and more fluctuations are observed for LRZ (Fig. 3b) as data is not equally spread over space, and the number of stations is also get changed with time. The year 1961 is a change-point year for HRZ and 1930 is for LRZ by using the BR test. The single change-point year by using Pettit’s test in terms of change in mean depicts in Fig. 3c and Fig. 3d. For the HRZ, 1961 is the change in mean rainfall year and the rainfall over this region get reduced after 1961 (Fig. 3c), similarly, for the LRZ, 1930 is a change in mean rainfall year, rainfall gets increased after this year (Fig. 3d).
The multiple change-point detection technique is used to find changes in rainfall with a small time interval (Fig. 3e and Fig. 3f). For the multiple point detection technique, two algorithms are used, offline and online. In the offline algorithm, entire data is checked once and find the change points within it and the online algorithm (which is also known as real-time change point detector) run simultaneously with a process to check the change in current point with the previous one (Aminikhanghahi and Cook, 2017). In this study, the offline method is used to detect change points in the time series. The R-software’s change-point package is used to calculate the change in the mean of yearly seasonal rainfall.
We found that 1920, 1930, 1950, 1952 and 1999 are change points for the mean rainfall over HRZ (Fig. 3e) and 1905, 1915, 1917, 1930 and 1964 (Fig. 3f) are the change points for the LRZ, interestingly, the year 1930 is a changing point and similar with the single point detection as noticed in the previous section. It is also noted that after 1960 both regions i.e. HRZ and LRZ are experiencing the change in mean rainfall. To understand the change in the standardized rainfall anomaly for both region, we divide study time into two equal halves i.e. P1 (Period 1: 1901 to 1958) and P2 (Period 2: 1959 to 2016). The midpoint of this division is near to change the point of HRZ (1961 by single and multiple change-point detections) and LRZ (1964 by multiple change-point detection). Over HRZ the difference between changing point is large compare to LRZ.
3.2 Observed climate change over HRZ and LRZ
The special report on “Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation” of the Intergovernmental Panel on Climate Change (IPCC) provides definition for climate change, according to the report the climate change can be identified by using various statistical test such as changes in the mean, changes in variability and that change must be continue for the decades or more (Field et al., 2012). The Report also provides a definition for the climate extremes as “the occurrence of a value of a weather or climate variable above (or below) a threshold value near the upper (or lower) ends of the range of observed values of the variable” (Field et al., 2012). The Probability Density Function (PDF) is a mathematical function and widely used to identify climate change (Field et al., 2012; Paeth et al., 2013; Mattews et al., 2016).
The PDF of seasonal average rainfall for each LRZ and HRZ has been computed for both the periods to understand whether there has been any shifting (change in per mm) in the mean rainfall in the recent period compared to the earlier period. The Gaussian curve of PDF for ISMR during P1 and P2 has shown in Fig. 4, the mean rainfall of HRZ shows a shift to the left side in P2 and distribution extends beyond 25th percentile of P1 which imply that seasonal mean is decreased (Fig. 4a). At the same time, the PDF curve gets fatter in P2 indicates that the variance of ISMR over HRZ is increased in P2 (Fig. 4a). On the other hand, the mean rainfall of LRZ has shifted towards right side that suggesting the mean seasonal rainfall is increased and variance is also slightly changed (Fig. 4b). In the recent period P2, the mean rainfall is decreased by 50 mm over HRZ and increased over LRZ by 10 mm. It is also noticed that in P2, climate extreme (excess and deficit) are increased over both zones, over HRZ deficit rainfall condition is increased and over LRZ excess rainfall condition is increased.
In order to understand further in deep, 5-years mean series and standardized anomaly are computed. Trend analysis with a binned of 5-year mean suggests an increasing trend in ISMR over HRZ during the P1, and a decreasing trend during P2, while for LRZ, ISMR shows an increasing trend in P1 (Fig. 4c) and no trend in P2 (Fig. 4d). The trend of rainfall averaged over the individual five years (pentad) starting from 1901-1905 has been presented in Fig. 4. The long-term (1901-2015) trend for the HRZ shows a decreasing trend (Fig. 4c) and LRZ shows an increasing trend (Fig. 4d). It is also found that number of excess (deficit) is increased (decreased) in P2 over LRZ, while a reverse picture in depicted for HRZ. Supplementary Fig. 1 shows the standardized anomaly for HRZ and LRZ. It is observed that 11 excess, 7 deficit and 40 normal rainfall years have occurred during P1 and 8 excess, 13 deficits and 37 normal monsoons took place during P2 over HRZ (Supplementary Fig. 1a). When analyzing the monsoon conditions for LRZ during two periods, it is observed that 4 excess, 10 deficit and 43 normal rainfall years have been ensued during P1, while 13 excess, 8 deficit and 37 normal rainfall years are observed during P2 (Supplementary Fig. 1b). It is fascinating to notice that the total number of normal rainfall years are changed and total number of climate extreme (excess + deficit) is increased over both the regions during the recent Period. Further, our findings are supported by Duncan et al. (2013), where they have shown that inter-annual ISMR variability shows a decreasing trend over eastern (which is HRZ) and an increasing trend over west and south India (which is LRZ) during 1951 to 2007 period.
Spatial distribution of the various rainfall events is presented in Fig. 5. The left side column (Fig. 5a, c, e, g, i, k) represents the spatial distribution of various rainfall events for HRZ and right side column (Fig. 5b, d, f, h, j, l) represents the spatial distribution of various rainfall events for LRZ. The details analysis of figures suggest that VLR and LR events has increased over northeast India (Fig. 5a,c) and LR to MR events are decreased over HRZ (Fig. 5c, e) On the other hand, over LRZ, VLR and LR events are decreased in P2 (Fig. 5b, d), while MR to VHR events are increased over LRZ (Fig. 5f, h , j). Recently, Prathipati et al. (2019) pointed out an inconsistency in the frequency of rainfall event over India during ISM season, they have shown that the number of HR events are increased over most parts of Madhya Maharashtra, Jammu & Kashmir, west central India, which are the parts of LRZ. The trends for various rainfall events is checked by using the Mann-Kendall trend test and presented in Table 1. It is found that the MR and RHR events are significantly decreased over HRZ and significantly increased over LRZ.
Table 1
Mann-Kendall trend test for various rainfall events during P1 and P2 for seasonal rainfall. The value greater than 0.05 is confident at 95 % level
Rainfall events
|
HRZ
|
LRZ
|
Z for P1
|
p-value
|
Z for P2
|
p-value
|
Z for P1
|
p-value
|
Z for P2
|
p-value
|
VLR
|
2.415
|
0.0157
|
-1.328
|
0.184*
|
2.052
|
0.040
|
-0.021
|
0.983*
|
LR
|
2.414
|
0.0157
|
-1.113
|
0.265*
|
2.066
|
0.038
|
0.0353
|
0.971*
|
MR
|
2.320
|
0.0202
|
-1.046
|
0.295*
|
1.945
|
0.051
|
0.129
|
0.896*
|
RHR
|
2.079
|
0.0375
|
-0.402
|
0.687*
|
1.73
|
0.083
|
0.275
|
0.782*
|
HR
|
1.757
|
0.0788*
|
-0.509
|
0.610*
|
1.086
|
0.277*
|
-0.550
|
0.582*
|
VHR
|
0.911
|
0.0559*
|
1.060
|
0.951*
|
1.012
|
0.311*
|
1.254
|
0.209*
|
EHR
|
0.2171
|
0.537*
|
0.947
|
0.882*
|
0.7848
|
0.432*
|
0.482
|
0.629*
|
Dry spells having different time-length and the total number of dry days for HRZ and LRZ have presented in Fig. 6. During 1901-2016 time-span, north-west and southern part of LRZ has experienced the number of dry days (Fig. 6a). The difference in dry days number between P1 and P2 reveals an increased in dry days over HRZ in P2 and decreased over north-west LRZ (Fig. 6b). Also, the different consecutive dry days (CDD) trends are presented in Supplementary Fig. 2. The increased in the total number of dry days is consequences of decreased in seasonal rainfall over HRZ, similarly, over LRZ seasonal rainfall is increased and the total number of dry days has decreased. Several studies have shown that monsoonal rainfall has been decreased over India especially monsoon core zone due to land use land cover, reduction in moisture supply from the Bay of Bengal, weakening of tropical easterly jet which are consequences of global warming (Rao et al., 2004; Naidu et al., 2011; Kulkarni 2012; Panda and Kumar, 2014; Paul et al., 2016). Here, we found that the rainfall is reduced over HRZ and rainfall is increased over LRZ. In P2, major contributing events (MR) and HRm which account combined effect of heavy rainfall events (HR, VHR and EHR) have decreased over HRZ except September (Fig. 7 a,b) and in revere, all these events are increased over LRZ (Fig. 7c, d).
The detailed analysis of the number of various rainfall events for each month of June-September (JJAS) has shown that the decreased in percentage (%) of the contribution of MR and HRm events (Fig.7) over HRZ is due to decreased in MR and HR events almost every month in P2 (Supplementary Fig. 3). In June and September, the number of low-intensity rain events (VLR and LR) has decreased (more than 200 events) along with MR events (Supplementary Fig. 3), that’s why the percentage (%) of the contribution of HMR events get increased in July and September (Fig. 2b). Over the LRZ, the MR and VHR events have increased for each month in P2 (Supplementary Fig. 3) which results in increased in percentage (%) of the contribution of the MR and HR events at seasonal and monthly scale (Fig. 7c, d). The difference in mean rainfall between HRZ and LRZ get reduced since 1901 (Supplementary Fig. 4). Over HRZ, low and medium intensity rainfall events have shown a decreasing trend and VHR and EHR has shown an increasing trend at 95% confidence level. Over LRZ, the LR, MR, RHR and VHR have shown an increasing trend at 95% confidence level. Overall, HRZ has experienced a decrease in seasonal rainfall and slightly increased in heavy rainfall events where LRZ has experienced an increased in both seasonal and heavy rainfall events in P2.