Observed rainfall changes in the past century (1901–2019) over the wettest place on Earth

Several studies have shown the influence of many air, sea and land surface factors that control the rainfall changes in India across the seasons. Therefore, we have used an MLR that includes all known, but important and established drivers of Indian rainfall variability, as described in detail in Nair et al (2018). The major drivers include the sea surface temperature (SST) of the Indian, Atlantic and Pacific Oceans through various atmospheric and oceanic processes such as the Indian Ocean Dipole (IOD), Atlantic Zonal Mode (AZM) and El Niño and Southern Oscillation (ENSO; Seetha et al 2019). The north Indian Ocean surface temperatures are further modified by IOD events and are represented by the Dipole Mode Index (DMI; Ashok et al 2001). Therefore, the temperatures of the Arabian Sea (SSTA) and Bay of Bengal (SSTB) are also considered in the regression. The extratropical SSTs have an appreciable effect on the Indian monsoon and are represented by the Extratropical SST (ESST; Chattopadhyay et al 2015). The changes in equatorial winds (i.e. Equatorial Wind Index or EQWIN over the region 2.5° S–2.5° N and 60° E–90° E (Francis and Gadgil 2013) can influence the SST distribution in the tropical Indian Ocean as the Equatorial Indian Ocean Oscillation (EQUINOO) modulates the monsoon circulation, and is considered in the regression procedure. The connection between ENSO and Indian rainfall is well established, and is represented by the Multivariate ENSO Index (MEI), which is based on six atmospheric and oceanic variables of the Niño 3.4 region (5 °N–5° S and 170°–120 °W) of the eastern Pacific (Trenberth and Hoar 1997). The central Pacific ENSO is considered by taking the El Niño Modoki Index (EMI, Ashok et al 2007). The impact of the Atlantic Ocean surface temperatures on Indian rainfall is represented by two related processes, the Atlantic Multidecadal Oscillation (AMO, Goswami et al 2006) and AZM (Kucharski et al 2008). The index corresponding to AZM is computed from the SST anomaly over 3° S–3° N and 20°–0° W, but the region 0°–60° N and 75°–7.5° W for AMO. In addition, a new Index is introduced in this study, which is the normalized difference vegetation index (NDVI) to represent the changes in LULC and vegetation in the northeast. The NDVI data are taken from the Landsat measurements. The MLR is then constructed for this study in the following form:

$$\begin{align*} {\text{R}}\left( {\text{t}} \right){\text{ = }}&{{\text{C}}_{\text{A}}}{\text{A}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{\text{z}}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{{\text{MEI}}}}{\text{MEI}}\left( {\text{t}} \right) \nonumber\\ & {\text{ + }}{{\text{C}}_{{\text{EMI}}}}{\text{EMI}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{{\text{AMO}}}}{\text{AMO}}\left( {\text{t}} \right)\nonumber\\ & {\text{ + }}{{\text{C}}_{{\text{AZM}}}}{\text{AZM}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{{\text{DMI}}}}{\text{DMI}}\left( {\text{t}} \right)\nonumber\\ &{\text{ + }}{{\text{C}}_{{\text{EQWIN}}}}{\text{EQWIN}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{{\text{ESST}}}}{\text{ESST}}\left( {\text{t}} \right)\nonumber\\ &{\text{ + }}{{\text{C}}_{{\text{SSTA}}}}{\text{SSTA}}\left( {\text{t}} \right){\text{ + }}{{\text{C}}_{{\text{SSTB}}}}{\text{SSTB}}\left( {\text{t}} \right)\nonumber\\ &{\text{ + }}{{\text{C}}_{{\text{NDVI}}}}{\text{NDVI}}\left( {\text{t}} \right){{ \varepsilon }}\left( {\text{t}} \right),\end{align*}$$

where R is the rainfall, C(t) is the linear trend, is the residual, t is the year and A is constant and is considered as 1. The terms with C as the prefix represent the corresponding regression coefficients (e.g. C_MEI is the regression coefficient of time series of MEI, which determines the contribution of the process represented by MEI when the other variables are kept constant in the regression procedure). The trend is considered as statistically significant when the value is greater than twice its standard deviation (i.e. in the 95% confidence level) here. The proxies and residuals are roughly normally distributed and the proxies are not auto-correlated. Further discussion on the model and its performance is given in Nair et al (2018).

In general, the rainfall starts in pre-monsoon by March and peaks during ISM from June–September at all stations, and figure 2 shows the monthly distribution of rainfall at different stations in the northeast. However, there is small spatial inhomogeneity in the peak rainfall, as seven stations show the peak rainfall in June and the remaining in July. The rainfall distribution is about 10 mm in April and peaks to 15–20 mm in June–July, depending on the stations. The CHE station is an exception to this, as it shows about 5 mm in March, about 15 mm in May, peaks to 30 mm in June, and a gradual decrease thereafter. Therefore, it exhibits heavy rainfall throughout the year, which makes the region the wettest place on Earth. As also mentioned in section 2, the time series data of rainfall at each station show a distinct difference in annual rainfall after the mid-1970s (figures 1 and S1) and thus, we have divided our analyses into before and after the year 1973. This apparent shift in the regional climate was also reported in earlier studies (Basistha et al 2009, Deka et al 2013, Goyal 2014) and henceforth, 1973 is taken as the turning year for the recent epoch. The monthly distribution of rainfall is similar at all stations (figure 2). Nevertheless, the rainfall exhibits a significant decrease in the recent epoch at most stations, except for DHU and JOR. It is also interesting to note the change in peak rainfall period in some stations in the recent epoch (1973–2019) compared to the previous epoch, and is explicit at CHE and SHG, where the peak rainfall occurs a month later than in the previous epoch. This demonstrates the spatial and temporal changes in rainfall in northeast India, including the wettest place on Earth.

Zoom In Zoom Out Reset image size

The northeast region experiences distinct seasonal variation in rainfall. Figure 3 shows the average (of all stations) rainfall received in the region in different seasons and each year. The values are computed for each season in a year and are represented by the year of the corresponding season. The winter rainfall shows the lowest among the seasons, about 1 mm, and the rainfall increased to 2 mm in 1917 and 1998, although in many years it was close to zero, such as in 1979 and 1999. The monsoon rainfall is also lower and almost 1 mm higher than that of winter, as the average rainfall in the season is about 2 mm. In some years, the rainfall amounts to 5 mm, as in 1986 and 3 mm in 1928, but hardly any in 1935 and 2015. Pre-monsoon and monsoon are the rainy seasons of northeast India. Average rainfall of about 6 mm during pre-monsoon and twice of that during monsoon are observed there with large inter-annual variability. The pre-monsoon showers show inter-annual changes of between 4 and 10 mm, but some extreme rainfall events such as 2 mm in 1961, 10.5 mm in 1930, 1977 and 2017 are also observed. The highest annual pre-monsoon rainfall occurred in 2010, about 11 mm. The monsoon rainfall oscillates between 9 and 22 mm, and the rainfall scales about 13 mm in most years. The highest monsoon rainfall in the region was recorded in 1974, about 22 mm, and the lowest rainfall was registered in 1963, about 8.5 mm. The amount of annual average rainfall received in the region coincides with the pre-monsoon rainfall, which is about half of the rain received during ISM, about 5–10 mm.

Zoom In Zoom Out Reset image size

Large variability is observed at the stations that experience high rainfall, such as at CHE, SHG, and SLH (figure 1), but little inter-annual variability is found at KLS, TEZ, GHY and IMF. Previous analyses suggest that the abrupt increase in rainfall after the 1970s in the CHE station is due to the anomalous anticyclonic circulation over central India and the northern Bay of Bengal at 200 and 850 hPa levels. The positive OLR anomaly and moisture divergence anomalies show suppression of convection over western India. The convergence of anomalous moisture flux over northeast India coincides well with active convection over northeast India, which drives above normal rainfall there (e.g. Murata et al 2007).

In general, most stations show the largest decreasing trend in rainfall in summer and the lowest in winter (figure 4, table 1). The decreasing trend in the southwest monsoon is about 0.001–0.280 mm dec⁻¹, and is largest at KLS, about 0.28 mm dec⁻¹. Six stations show statistically significant negative trends in the rainfall, i.e. at AIZ, DIB, JOR, KLS, NLP and TEZ. Although not statistically significant, negative trends are estimated at GHY, IMF, ITA, PGT and SLH. However, statistically significant positive trends are observed at CHE, DHU and SHG, but are insignificant at AGT. A clear reduction in rainfall is evident at most stations since the late 1970s (figures 2 and S2). Nonetheless, the estimated trends at CHE and SHG must be interpreted carefully as there is a steep decrease in rainfall at both stations since mid-1980. Detailed trend analyses for the period 1973–2019 are also shown in figure 4 (bottom) and table S2.

Zoom In Zoom Out Reset image size

In winter, insignificant positive trends are found at AGT and SHG, and insignificant negative trends are estimated for AIZ, CHE, DHU, GHY, IMF, ITA, NLP, PGT, SLH and TEZ. Statistically significant negative trends are observed at the remaining three stations. Similar to winter rainfall, the pre-monsoon data show statistically significant positive trends at AGT and SHG in 1901–2019. On the other hand, AIZ, JOR and NLP show statistically significant negative trends, although CHE, DIB, GHY, ITA, KLS and TEZ show statistically insignificant negative trends. The monsoon (June–September) rainfall shows increasing trends at AGT, CHE, DHU and SHG, and except for AGT the remaining stations show statistically significant rainfall trends. On the other hand, decreasing trends are observed at AIZ, DIB, JOR, KLS, NLP and TEZ, which are statistically significant, while the trends at the remaining stations are insignificant. In the case of post-monsoon rainfall, ten stations exhibit a decreasing trend, for which JOR and NLP show statistically significant trends. In contrast, AGT, CHE, DHU, GHY and SHG show positive trends, in which the SHG measurements are statistically significant.

The annual mean rainfall for the period 1973–2019 shows decreasing trends of about 0.42 mm dec⁻¹ and are statistically significant, along with seven other stations, i.e. AGT, CHE, GHY, KLS, PGT, SHG and SLH. The winter rainfall is decreasing in the northeast at all stations (0.11 mm dec⁻¹, from the linear trends calculated for both periods, as shown in the figure 4 bottom panel and table S3) and this reduction in rainfall is statistically significant at NLP, PGT and SHG. In pre-monsoon (March–May), eight stations show a decreasing trend in rainfall in which only one station shows a statistically significant trend, i.e. at SHG. On the other hand, seven stations show an increasing trend in rainfall, but none is statistically significant. The rainfall during the pre-monsoon season, throughout the study area, shows a statistically insignificant decreasing trend (table S3).

The analyses are again cross-checked with the ERA5 rainfall data by applying the same trend detection method (table S4). As discussed, most stations show decreasing trends in rainfall across the seasons, where the trends in winter, pre-monsoon and post-monsoon are dominated by decreasing rainfall at most stations. The trend values are also comparable, such as the estimated annual averaged rainfall trends of −0.14 ± 0.25 mm dec⁻¹ at KLS and −0.13 ± 0.68 mm dec⁻¹ at CHE. It is interesting to note the negative trends estimated for the CHE and SHG data (0.1–0.3 mm dec⁻¹), which are consistent with the surface measurements, implying that the rainfall is decreasing at the wettest place on Earth in all seasons. The negative trends shown at PGT and DIB are statistically significant, and are around 0.62 mm dec⁻¹. Nonetheless, the trends at DHU, GHY, IMF and ITA show positive values as the rainfall received in each season at these stations is higher than that of other stations.

There are two distinct peaks in the observed rainfall measurements at CHE around the mid-1970s and mid-1980s for the period 1901–2019. This is clearly illustrated in figures 1 and S2. The trend estimated for the first epoch has a part of one of these peaks and the data up to 1950 have no clear temporal trends. Therefore, the trend estimated for the first epoch, which is weighted by the measurements during the period 1950–1972, is positive. On the other hand, the trend estimates in the second epoch include a part of the peak of mid-1970, the high rainfall in the mid-1980s, and the lowest rainfall ever recorded at CHE in the late 2000s. These make the trend value negative for the recent decade. This is the reason that we performed a test on the turnaround period for trend estimates, as the initial and end year measurements are very important and would influence the linear trend estimates.

In brief, ISMR at most stations shows a decreasing trend for the period 1973–2019. The analyses show a reduction in rainfall at all stations, in which seven are statistically significant. These analyses clearly demonstrate that the rainfall is decreasing during the recent epoch in northeast India. The decrease in rainfall is also evident at the wettest place on Earth (CHE) and its nearby station (SHG), and both stations are situated at higher elevations (figure 1 and table S1). It is evident from our analysis that northeast India receives a smaller amount of rainfall during ISM in this epoch compared to the previous epoch. This is also consistent with the analysis for the period of 119 years (1901–2019) over the entire northeast region that showed an insignificant decreasing trend of 0.001 mm dec⁻¹. In post-monsoon, all stations show negative trends in rainfall and are significant at two stations, i.e. CHE and PGT. The inter-annual variability is comparatively smaller for the post-monsoon rainfall at all stations (figure 3).

To further assess the estimated trends, we have used another method of trend analysis, the MLR, as discussed in section 2.1. The method also indicates the contribution of different remote and local forces to rainfall change in northeast India. The analyses of 119 years of rainfall data show that there is a significant reduction in the rainfall received in all seasons in both epochs. The decrease in rainfall over the period on the wettest place on Earth is even more evident and alarming. In all seasons, the rainfall shows a substantial reduction, except in pre-monsoon. The smallest trend is observed in pre-monsoon and post-monsoon, around −0.18 mm dec⁻¹, and the largest in the summer monsoon, about −0.785 mm dec⁻¹ (table S5). The pre-monsoon trends show small positive values, implying the changes in rainfall as it is normally infrequent and very light in this season. However, trends in this season also show signatures of significant change. We did a separate analysis for the wettest place, CHE, where all seasons, except pre-monsoon, show a substantial reduction in rainfall in the recent epoch (1973–2019). The estimated trends for the northeast and CHE are statistically significant in all seasons and for the annual averaged data, except for the pre-monsoon showers, where the trends show insignificant positive values. The magnitude of decreasing trends is several times the average for the whole northeast (e.g. seven fold in post-monsoon, about −0.8 ± 0.3 mm dec⁻¹) and are significant too; showing the rapid changes in the rainfall and its intensity in the region. The change in rainfall pattern also moved the rainfall intensity and frequency to further west of CHE to MAW, which makes the latter the wettest place on Earth.

These results are also confirmed by the analyses with annual accumulated rainfall at CHE and MAW (shown in the figure 1 bottom panel). The analyses show that the average rainfall at CHE is about 11 963 ± 2346 mm (i.e. average ± standard deviation) and MAW is about 12 550 ± 2321 mm for the period 1989–2010. The linear trends computed for the same period also show a larger decrease of about 27 ± 39 mm yr⁻¹ at CHE and 6.9 ± 73 mm yr⁻¹ at MAW. The highest ever rainfall in the northeast was measured at MAW, about 26 000 mm in 1985–1986. These rainfall measurements also reveal that the wettest place on Earth moved from CHE to MAW by the late 1970s. Some of the expected signatures of climate change are indeed the changes in regional rainfall pattern and warming of the world oceans. Henceforth, we explore further to understand the causes of the significant changes in rainfall that are clearly visible in northeast India.

We analysed the contributory factors to the rainfall variability in northeast, computed using the MLR and are illustrated by figure 4 (top: northeast, bottom: CHE). The winter rainfall is about 1 mm on average and the contributions from different factors are also lowest in that season. The contribution of different factors is smaller in the post-monsoon season and are mostly negative, but the ENSO and equatorial winds dominate there, and are less than 0.5 mm of rain. In pre-monsoon, most factors favour rainfall and contribute to 0.5–1 mm and the contribution from ENSO is the largest among the factors. The changes in the SST of Arabian Sea and the equatorial Indian Ocean (i.e. DMI and ESST) contribute to the rainfall in pre-monsoon. In addition, the analyses show that the NDVI or the vegetation also plays a significant role in rainfall during pre-monsoon due to healthy leaf senescence there, about 0.7 mm. During monsoon, all factors except equatorial winds (EQWIN) contribute positively to the rainfall, which is about −1.0 mm by EQWIN. The other factors contribute up to 0.8 mm of rain, and the contribution of the equatorial Bay of Bengal and Atlantic SST is noticeable. The NDVI also contributes about 0.4 mm of rain in this season, which is greater than that of four other contributory factors.

The regression with annual averaged rainfall data shows appreciable contribution from ENSO and equatorial ocean temperatures (ESST), about 2.2 mm. The north Atlantic and Bay of Bengal surface temperatures, and vegetation cover contribute about 0.2–0.4 mm. However, as far as the changes in temperatures of the north Indian Ocean are concerned, SSTB has a greater influence than SSTA and its contribution is about −0.5 mm yr⁻¹. The influence of equatorial winds and Arabian Sea is very limited and their contributions to rainfall are negative. This suggests that the changes in moisture transport from the Arabian Sea do not significantly affect the southwest monsoon rainfall in the northeast compared to that from the Bay of Bengal, consistent with the analyses of Nair et al (2018).

Nevertheless, there is a spatial inhomogeneity in the contribution of different factors of rainfall change. For instance, as the rainfall in CHE is the highest in the world, henceforth, proportionately higher values of contributions from different factors are estimated there. Since monsoon rains are highest among the seasons, the largest contribution is found during this season and mostly, the Atlantic and Indian Ocean temperatures decide the amount of rainfall received in this season, about 3–5.5 mm. The changes in vegetation also have a key role in the variability of rainfall there, about 1.5 mm. Conversely, the strength of equatorial winds has a greater role in suppressing the rainfall and its contribution amounts to −8 mm, the largest among the contributory factors. The annual rainfall shows the contribution mostly from the Pacific and equatorial Indian Ocean surface temperature changes. Since vegetation or land cover changes play a predominant role in rainfall in the northeast and CHE regions, we further examine this aspect in detail in the next section using LULC analyses.

There is a significant change in LULC in recent decades in northeast India and the changes are mostly in the forest cover and vegetation in the region. Although the published records may not show the up-to-date extent of deforestation, satellite measurements can pinpoint the actual deforestation and all types of LULC. Therefore, we have used satellite observations for monitoring the temporal changes in LULC in the region. During the past few decades, the LULC of India was changed substantially due to the increase in population from 1951 (361 million) to 2011 (1221 million as per the latest census report of 2011) and due to development activities (Moulds et al 2018). Northeast India has the highest vegetation cover in India and includes 18 biodiversity hotspots of the world, indicating the importance of the region in terms of its greenery and climate change sensitivity. Population growth demands more food production and infrastructure development, and has changed the LULC of the region. Forest cover is the most exposed vegetation type to anthropogenic activities such as the conversion of forestland to mining, agriculture, urban area expansion and railroad construction (Lele and Joshi 2009, Chitale and Behera 2014). Therefore, we have done an analysis on the LULC change in northeast using satellite measurements and the results are shown in figure 5 and table S3.

Zoom In Zoom Out Reset image size

Our analysis (figure 6) shows noticeable reductions in vegetation (−104.5 km² yr⁻¹), barren land (−106.1 km² yr⁻¹) and snow and ice (−3.9 km² yr⁻¹), but significant increases in cropland (182.1 km² yr⁻¹), water bodies (32.1 km² yr⁻¹), urban and built-up lands (0.3 km² yr⁻¹) during the period 2001–2018. The change in percentage area reveals that the waterbodies, urban regions and built-up areas are continuously increasing from the reference year of 2001. However, the decrease in vegetation cover and increase in the areas of cropland are observed from the year 2006 onwards, implying the conversion of forestlands and vegetation cover to croplands. This is also the first step of the urbanization practice, as forest or vegetated areas are converted to croplands and then those lands will be eventually altered to urban settings. The changes in snow cover over the years can be linked to the slow warming in the region, which also increased the barren lands at higher elevations and led to the expansion of waterbodies. The increase in barren lands in the lower elevation areas exhibits its use for urban conversions. It has to be noted that the Indian government has launched a number of socio-economic schemes to enhance the lives of people there, such as the Mahatma Gandhi National Rural Employment Guarantee Scheme, Joint Forest Management, Japan International Cooperation Agency and Indo–German Development Cooperation Project. A number of check dams of small and medium size watersheds have also been constructed as a part of these schemes, and are the other reasons for the small increase in waterbodies in the past two decades (e.g. Behera et al 2018).

Zoom In Zoom Out Reset image size

We also used the CCI data to investigate the MODIS results, which are shown in figure S2 and table S7. Both data show similar changes for most LULC components for the overlapping time period (2001–2018, red trend line in figure S2). However, there is a small difference in snow and ice area analyses, as the MODIS data show a significant decrease in its area, but no detectable decrease is found with the CCI data. Although the waterbodies show a decreasing trend in CCI data, the MODIS data show an increasing trend in their area since 2001. Yet, the area of waterbodies shows an increasing trend from 2005 onwards; confirming the positive trends estimated with the MODIS data. Therefore, both data sets reveal similar changes in most LULC components and suggest significant alterations in LULC in northeast India. Nevertheless, these results also show the uncertainties in different land surface variables of LULC in different satellite data. The changes in LULC have a negative impact on the precipitation through the modification of temperature by evapotranspiration in the region, as shown by the negative trends in rainfall of CHE and SHG, which are the heavy rainfall regions in the northeast and on Earth. Note that the climate change signatures will be first evident for the extreme cases, such as the rainfall received at the then wettest place in the world, and as such, CHE and is clearly exposed in our analyses.