Changing Rainfall Patterns in an Era of Climate Change: A Multiparameter Spatiotemporal Analysis of Trends & Impacts for India


 The hydrological cycle that starts with rainfall has been under major threat from the global temperature rise and climatic changes. In India, rainfall changes not only jeopardize water security but also have a major set-back for socio-economic stability. There have been attempts to decode the changing rainfall patterns in India but most of them conducted at wider spatial resolution (such as national, state, or sub-divisional level) fail to capture the essence of spatial variation in rainfall characteristics. To get a clearer understanding of change in key rainfall parameters, this paper analyses more than 197 million 0.25˚ x 0.25˚ gridded rainfall data points. The fine resolution 117 years (1901-2017) of daily rainfall data is utilized to test significant spatiotemporal trends in the quantum of rainfall and other key rainfall parameters such as rainy days, monsoon onset and withdrawal dates, occurrences of extreme rainfall events, and frequency of drought and high rainfall years. With an emphasis on changing climatic patterns since perceived climate change onset in the 1970s, the study identifies the regions with significant changes in rainfall patterns by comparing key parameters pre- & post- 1970s. The paper also highlights the major repercussions and challenges for the identified regions with significant changing rainfall patterns.

between the months of June and September. Except for high-altitude regions of northern and north-eastern parts, the 27 precipitation is normally in the form of rainfall (liquid-state precipitation) in India. (Thus, the words precipitation and 28 rainfall are used interchangeably in this paper). Spatially, India has a great diversity in rainfall conditions, with a few 29 pockets in northeastern hills getting 10,000 mm of annual precipitation on one extreme whereas the western desert 30 region receives less than 500 mm of rainfall annually. Besides quantum of rainfall, the onset of monsoon, number of 31 rainy days, and variation in year-on-year rainfall are key variables of rainfall diversity in India. 32 The substance of good rainfall for the socio-economic stability of India can be gauged from the fact that almost half of 33 in (Yu, Zou and Whittemore, 1993;Yue and Hashino, 2003;Partal and Kahya, 2006). For better Spatial understanding 122 of pockets with significant trends (MK Test, p-value < 0.05) 1 , Maps are produced with Sen's estimate value for pixels 123 observing significant trends. (p < 0.05). 124 Extreme Events Analysis for observing how the composition of extreme rainfall events are changing over the years, a 125 total number of extreme rainfall events (>65 mm in a day) and very extreme rainfall events (>100 mm in a day) are 126 summed for each year across India and plotted with Local Polynomial Regression (LOESS) Regression to understand 127 the trend over the years. Table 1

Analysis 132
The paper aims to take up spatial and temporal analyses of the rainfall parameter to capture the nuances and 133 peculiarity of the diverse climatic regions of India. The wholistic quintessence understanding of rainfall requires several 134 rainfall parameters that are beyond the scope of this paper. To get an as good understanding as possible, the analysis 135 focuses on a few important rainfall parameters such as Annual 2) Compare the long-term trends of these parameters before the climatic change with the era since the beginning of 151 climatic changes. 152 To select a breakpoint, the year considered as the point of significant change in the climatic trends, we use literature 153 as well as a change-point analysis of extreme rainfall events results. To capture a single point for change in the rainfall 154 timeline, we need to use a parameter that can better capture the anomalistic behavior in rainfall when aggregated at 155 the national level. The extreme rainfall events are opted for the change point analysis as they better emulate the 156 changing rainfall pattern which might not be easily captured through the aggregation of the total quantum of rainfall 157 or rainy days across data points. For example, a region may have seen a similar amount of rainfall say 500 mm over 158 the years with an average of 20 rainy days without any significant changes over the years. But there can be a 159 considerable change in extreme events as now the region experiences 5 days with > 65 mm of rainfall as compared to 160 1-2 days a few years back. Also, the sum of extreme events is easier to aggregate over the entire geography of India 161 as compared to aggregating rainy days and better parameters to capture anomalies as compared to the quantum of 162 rainfall. To understand rainfall characteristics and usual spatial distribution, we are starting the analyses by plotting the annual 176 quantum of rainfall aggregated at a national scale (Fig 1 (a)) and ii) plotting monsoonal as well as the annual quantum 177 of rainfall (Fig 1 (b)). This initial analysis shall help to comprehend normal rainfall conditions in India but an aggregation 178 of data at the national scale shall not reflect any intra-annual observable changes across regions. The annual national 179 rainfall average plotted in Fig 1(a) shows the total cumulative volume of the water precipitated each year across all 180 the observational pixels along with the local regression curve (LOESS Regression). The total quantum of rainfall is 181 calculated by taking the cumulative sum of the product of quantum of rainfall (in mm) multiplied by the area of the 182 pixel (km 2 ). (Equation (1))  183 At a national scale, rainfall shows a periodic behaviour with an alternate cycle of higher and lower than normal rainfall. 184 These alternate decadal patterns are suggestive of the absence of significant monotonic trends at the macro scale. 185 The MK-Test for the annual rainfall data series does not show a significant increasing trend (p > 0.05), but a LOESS 186 Regression trendline (red line in Fig 1 (A)) suggests the total quantum of rainfall is on an average 200 BCM less rainfall 187 than its peak in the 1940s. Although the observation cannot be classified as a trend, the plot certainly indicates there 188 have been more frequent drier years in the recent decade as compared to a few decades ago. Fig. 1

(B) shows how the 189
Monsoonal Rainfall contributes primarily to total annual rainfall. Indian subcontinent predominantly receives shower during the months of June-July-August-September (JJAS) with most parts of India experiencing more than 80 percent 191 of the total annual rainfall during these four months. The rest of the seasons are predominantly dry.

Normal Spatiotemporal Variation in Rainfall 223
Taking the first step towards combining the two extents of the analysis, space, and time, we need to understand how 224 the rainfall distribution normally varies over the years as well as space. There is always some year-on-year variability 225 in rainfall within spatial pockets, which is normal in observing any natural phenomenon. Thus, the variation over the 226 years also becomes a key parameter in defining the rainfall characteristics. For instance, the western desert districts 227 of India receive scanty and highly volatile monsoonal rainfall. Every 4 th or 5 th year is a dry year and on the other side, 228 there is also an equal probability of very high rainfall every 4 th or 5 th year in these districts. Whereas central and east 229 to central parts of India don't have that high rainfall variability. They normally experience a dry year once in 10 years. 230 We are quantifying this variability across the pre-1970s and post the 1970s for each pixel using a coefficient of variation 231 [CV = (Standard Deviation)/Mean]. Overall, rainfall aggregated at the national level shows the standard deviation of 232 rainfall as 95.3 mm against the mean rainfall of 1108.7 mm. (Data in Fig 1) The coefficient of variation for the composite 233 rainfall stands at 8.6 percent. 234 The majority of Indian pockets have a CV in the range of 20-40 percent, mostly around 25 percent. (Fig 4) Except a few 235 western hot dessert and regions and norther cold desert areas of Leh and Ladakh that show very high rainfall variability 236 that explains their desert characteristics to some extent. The gradient of annual variation in rainfall is decreasing 237 almost continuously moving eastward. Even a few pockets in the north-eastern regions of India also experience high 238 rainfall and also high variability too. Variation in rainfall is expected but high variability increases the uncertainty about 239 the availability of water for the annual cycle of agriculture and other domestic needs. The first evidence for the 240 changing rainfall patterns can be witnessed from Fig 4. Fig 4 (c)

Trend in Annual Total Quantum of Rainfall 257
The most noticeable and imperative parameter to understand the rainfall trend is the total quantum of rainfall. Figure

Trend in Number of Rainy days 281
It is possible the region has experienced the same number of rainfall but the rainfall is more squeezed in terms of rainy 282 days. As stated earlier, rainy days may help in identifying deviation from normal rainfall. The MK Tests for the annual 283 total number of rainy-days show that most North and Eastern India has experienced a significant declining trend of 284 the total number of rainy-days post-1970 which is inconsistent with the pre-1970 period. (Fig 6)  Due to the reduced number of rainy days, the regions may experience frequent agriculture droughts even if they are 291 not facing the hydrological or meteorological drought. The shorter wet spell followed by a longer dry spell leads to 292 crop failure even during the meteorologically good rainfall year. Shorter rainfall spell shall also lead to less natural 293 percolation of the rainwater and higher surface run-off that may over flood the dam capacities but does not recharge 294 the aquifers efficiently. 295

Trend in Monsoon Onset & Withdrawal 296
A few of the most important rainfall parameters for agriculture activities are the onset and withdrawal dates for the 297 monsoon. Many agrarian decisions such as cropping patterns, use of irrigation, demand for seeds, and fertilizers 298 heavily depend on the onset of monsoon. The anomalistic onset of the monsoon dates can jeopardize agrarian 299 planning for states, markets, and farmers. This can even sink the cost of a decision taken based on the normal rainfall 300 onset time such as the purchase of seeds of the crops to sow. As stated earlier, India's monsoon starts from the 301 southwestern coast of Kerala in the first week of June and spreads across India in the next 30-40 days, and covers 302 major parts of India. (Fig 3) 303 The worrying trend revealed in Figure 7 (B) shows that the trend has been anomalous in a different part of India. 304 Although the onset of monsoon in the Kerala coast has not seen any deviation, the eastern regions are receiving early 305 monsoon onset (red pixels), whereas the western part is experiencing a late arrival trend (blue pixels). As the onset dates are important for agrarian decisions, the withdrawal of rainfall is important to observe for the 314 agriculture output. Very early withdrawal can deprive the crop of the crucial water needs before harvest that can 315 impact the yield to a great extent whereas the delayed showers can be detrimental for harvest-ready crop-fields and 316 spoil the production in the Indian agrarian sector that faces scarcity of storage infrastructure. The end date for the 317 monsoon rainfall map is predominantly shaded-red (early withdrawal) but the trend is statistically inconclusive 318 (p>0.05) post-1970s except for a few pockets. (Fig 8) Although there are significant changes in the onset of monsoon 319 dates, the withdrawal dates do not show any significant change over the years. This supports the phenomenon of 320 shrinking monsoon season altogether in India. 321

Trends in Extreme Events 324
This set of rainfall parameters cannot follow the pattern used for analyses in section 3.3 due to the peculiarity of 325 occurrence of high rainfall events are not continuous and the dry or wet years are identified based on the cumulative 326 annual rainfall and they are mostly understood based on the frequency of their occurrences. 327

Dry and Wet year Frequency 328
The annual rainfall is discretely having different patterns each year, and there are certainly some years having distinctly 329 higher than normal or much lower than the normal rainfall. Classifying the years in which the area has received a 25 330 percent deviation from the normal rainfall as dry (less than normal) or wet (more than normal) years, a definition 331 adopted by NRAA (2020), we have compared the vicennial changes in the frequency of these years are plotted in Fig  332   9. is an important factor for agrarian and livelihood sustenance as it absorbs the negative impacts of the rainfall variability 365 to some extent. Also, the lack of livelihood sustenance reduces the resilience capacity of the population against any shock such as the climatic changes thus it is used as a proxy of the overall vulnerability of the population in different 367 parts of India along with percent rainfed area (complementing the irrigated area). 368 Quantum of rainfall and rainy days trend analyses in sections 3 indicates the northern parts of India majorly comprising 369 of the states such as Eastern Uttar Pradesh, Bihar, West Bengal, Jharkhand, and Orissa that have seen adverse rainfall 370 conditions (See Fig 5-6). These are the states that count for more than 40 percent population of India, and an estimated 371 70 percent of people live on agriculture and allied activities with major populations following substantial agriculture 372 practices. The stark reality is further revealed in Fig 11 (B) that shows the region in yellow and red is a higher priority 373 for livelihood-related interventions due to prevalent poverty and livelihood challenges. The changing rainfall and 374 decreasing water availability shall hamper the already stressed regions of central and northern India. Also, these 375 regions are worst prepared too. The regions are also the ones that are having the least human development index 376 score in India and lack adequate resources to cope up with the risks associated with the changing normality of rainfall. 377 Decreasing quantum of rainfall and changing rainy days, monsoonal patterns are catastrophic for the agrarian well-378 being of the rainfed regions. Irrigation that is a very basic deterrent against falling water availability through rainfall is 379 not adequate and lowest in the regions (except the state of Uttar Pradesh) with the most significant adverse rainfall 380 changes. The changes are also detrimental for the irrigated regions with a lack of irrigation sustainability such as the 381 State of Uttar Pradesh, Haryana, Punjab, etc. The abrupt rainfall events and shrinking rainy-days are detrimental for 382 the groundwater irrigated pockets as the abnormal rainfall events will lead to more surface run-off/less recharge on 383 top of further augmented groundwater extraction. India is already undergoing heavy groundwater stress, and the 384 adverse climatic condition will prove further detrimental. Although India is currently the world's highest groundwater 385 extractor, still there is potential to further develop the groundwater resources for quick and reliable water resources 386 against the adverse climatic conditions in the regions with less groundwater development. Due to less groundwater 387 development and less groundwater exploitation in the north and eastern part of India (highlighted earlier in the section 388 for increasing threats of water security) (also see Fig 12 (A)), developing groundwater irrigation can potentially lead a 389 way for resilience against changing rainfall patterns. But groundwater irrigation has a direct correlation with the 390 accessibility of energy for irrigation. Unfortunately, the regions experiencing the rainfall changes are also the ones 391 with the highest energy cost for irrigation (coloured in Yellow in Fig 12). There is an urgent need to make irrigation 392 more accessible so that the impacts of variability in the rainfall can be mitigated. Also it is tested hypothesis that the 393 investment in Managed Aquifer recharge work can yield better result for regions with less groundwater availability 394 and rainfall volatility. (Patel, Saha and Shah, 2020) Regions experiencing the better quantum of rainfall but increased 395 variability (Western and Peninsular Regions, see