The Kerala state is gateway of Indian monsoon (Joseph et al., 1994). The southwest monsoon irrigates the majority of Kerala’s cropland. The onset of the monsoon dictates the timing of crop sowing because the amount of rainfall influences productivity. Because of the higher geographical variability of rainfall in a tropical region like Kerala, rainfall measurement, monitoring, and analysis are essential. However, the current rainfall observation network is insufficient to monitor the rainfall. It might be challenging to set up and maintain weather stations in remote locations. The highly dynamic nature of precipitation in terms of amount and intensity over a given area makes determining its geographic distribution difficult (Sharma et al., 2012). However, meteorological satellites can provide a precise spatial distribution of rainfall. TRMM (Tropical Rainfall Measurement Mission) is a research satellite built by the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA) as part of their joint space mission (Bookhagen, 2010; Sharma et al., 2012). It uses space-borne radar to detect the precise estimations of near-surface rainfall. The TRMM intended for better recognizing the distribution and variability of rainfall in the tropics as part of the current climate system. It can provide reliable rainfall estimates with adequate validation of rainfall (Gu et al., 2010; Chang et al., 2013; Tarek et al., 2017). David (2020) found that comparing annual TRMM data to average rainfall readings from the India Meteorological Department (IMD) yields a 0.85 correlation coefficient, indicating that the estimation is trustworthy.

Studies on rainfall variability and trends have been conducted by various researchers worldwide (Nicholson and Grist, 2001; Murphy and Timbal, 2007; Nouaceur and Murarescu, 2020). Parthasarathy et al. (1995) studied the Indian monsoon rainfall for long-term trends over the country and the subdivisions. Thomas et al. (2015) and Sreelash (2018) studied seasonal and annual rainfall trends over Kerala during the last century. The long-term tendencies of precipitation over Kerala state were studied by Soman et al. (1988), Krishnakumar et al. (2009), Pal and Al-Tabbaa (2009) and Thomas and Prasannakumar (2016). The yearly and southwest monsoon season rainfall amounts in Kerala have dropped over the previous century (Guhathakurta and Rajeevan, 2008; Krishnakumar et al., 2009). Nikhil Raj and Azeez (2010) showed that climate change affects rainfall variation locally and globally. They evaluated the region’s annual rainfall, southwest monsoon and pre-monsoon rainfall, indicating a significant decline in rainfall over the last years. The annual and seasonal rainfall trends in most parts of Kerala are also showing a considerable decline. According to Nair et al. (2014), the declining trend in rainfall and global anomalies linked to greenhouse gas emissions from increased petroleum consumption, land-use variation due to developmental activities and deforestation, and atmospheric pollutants linked to transportation facility development.

Researchers are facing the scarcity of rainfall data due to unavailability and higher costs especially for weather stations in remote areas. The TRMM data can generate the daily rainfall data. Therefore, this study tries to estimate the variability and trends over Kerala during 1998-2019 using TRMM data. The observatory data used to find out the dependability of the TRMM data. We used both IMD and TRMM data simultaneously to compare the trends and variability. Statistical techniques, the Mann-Kendall test and Sen’s slope used to evaluate annual, monthly, and seasonal rainfall variability and trends. Numerous researchers have studied the trends and variability of rainfall over Kerala based on data for last 100 years or longer but not for the recent years. Therefore, the present study focused on analysis of monthly, yearly and seasonal variability and trends in rainfall during the last 22 years (1998-2019) over Kerala state. It will be beneficial to understand the most recent climate changes, and it will be helpful to the state government for adaptation of management strategies to ameliorate the harmful effects of climate change.

The study depended solely on TRMM and IMD rainfall data. The daily accumulated precipitation data results from the TRMM Multi-Satellite Precipitation Analysis (TMPA) data collected every three hours (3B42) (Huffman et al., 2016). It covers daily rainfall data from 50º S to 40ºN at a 0.25º x 0.25º spatial resolution. The 3B43 dataset is a monthly product of the 3B42 dataset. This product derived using TRMM-adjusted integrated microwave infrared precipitation rates and root-mean-square error estimates. TRMM three hourly and monthly rainfall data can be freely accessible, and it is available on the website maintained by NASA. In this study, the TRMM daily rainfall was calculated using area-averaged gridded data. We have selected the area-averaged gridded rainfall data for the fourteen districts of Kerala where the IMD observation networks are available. It will increase the accuracy of TRMM data while comparing it with IMD observatory ainfall. The area-averaged daily precipitation data for 14 districts retrieved and used to figure out the average monthly, seasonal, and annual rainfall over Kerala. IMD Monthly rainfall data for Kerala was obtained from the IITM publication “Monthly, Seasonal, and Annual Rainfall Time Series for All-India, Homogeneous Regions, and Meteorological Subdivisions: 1871-2016” (Kothawale and Rajeevan, 2017), and for 2017 to 2019 data was collected from the IMD publication “Rainfall Statistics of India - 2012 to 2019”. Statistical analysis employing standard parameters used to evaluate the satellite data to the IMD observatory data. Among these, the Pearson correlation coefficient, which used to determine the linear relationship between two data sets. Additionally, the bias factor (Duan et al., 2012) and the Root Mean Square Error (RMSE) are used to assess the difference in TRMM and IMD observatory data.

Daily rainfall data (1998 to 2019) used to analyze trends, variability, and mean rainfall patterns. The analysis of rainfall data separated into four broad categories: yearly, monthly, seasonal, and decadal time scales. First, the daily rainfall data converted to monthly data. The Pre-monsoon (March to May), Southwest Monsoon (June to September), Post-monsoon (October to November), and winter (December to February) are the prominent seasons considered. Then, monthly rainfall data divided into two groups to determine rainfall variability across two periods: 1998-2008 and 2009-2019. To compare the spatial dependence and station wise performance of TRMM data, we compared the monsoon season rainfall for 14 stations. The mean, standard deviation (SD) and coefficient of variation (CV) used to determine rainfall patterns and variability. The Mann-Kendal test and Sen’s slope used to determine the magnitude of annual, monthly, seasonal, and decadal rainfall trends. For all rainfall trends, the p-value used to determine statistical significance. There is some discrepancy between the results of both the data sets. Therefore, the trend and variability analysis derived from the IMD data considered as acceptable results.

4.1.3 Comparison of Seasonal Rainfall Data

Another comparison made between TRMM 3B42 V-7 rainfall products and observatory data based on seasonal classification. The comparison of the two datasets for Kerala’s four different seasonal average rainfall levels shown in figure 5. The TRMM data for winter, pre-monsoon, southwest monsoon and post-monsoon is more consistent with the observatory data, with correlation coefficients of 0.855, 0.963, 0.899, and 0.834 with RMSE values of 19.57, 50.42, 286.85, and 85.11 mm, respectively. The presence of a high R-value suggests that TRMM can follow the trend, whereas the RMSE shows the presence of under-or over-estimation. TRMM consistently underestimates observatory rainfall data during the southwest monsoon, with an average bias of -0.13.

Figure 5. Seasonal average monthly rainfall recorded at observatory and TRMM data (1998-2019)

4.1.4 Comparison of Highest and Lowest Rainfall Month

Moreover, the TRMM 3B42 V-7 total monthly rainfall data compared to observatory data for the study period’s peak and lowest rainfall months (Figure 6). June is the wettest month, while January is the driest. Thus, TRMM underestimates total rainfall in June and overestimating in January months. The TRMM data for the month with the highest and lowest rainfall values agrees better with the observatory data, with R values of 0.871 and 0.618 with RMSE values of 149.3 and 6.78 mm, respectively. The TRMM delivers more reliable estimated estimations during the month with the highest rainfall than during the month with the lowest rainfall.

Figure 6. Average rainfall recorded at observatory and TRMM data for highest (June) and lowest (January) months (1998-2019)

4.1.5 Comparison of Station Wise Data and Spatial Pattern

Kerala is a complex terrain with significant topographical changes, especially from the West (Arabian Sea) to East (Western Ghat) directions with an average width of approximately 70 km. In this short distance, we can observe seashore, plains and hills. Some regions are below sea level (Kuttanadu region) to 2695 m (Anamudi peak) above sea level. To compare station wise monsoon season rainfall, we selected 14 major IMD stations (Figure 7), representing each district for the last ten years (2009-2019; the year 2011 has no district wise rainfall station data available). Rainfall readings are frequently sensitive to location and altitude. It is decided to explore the TRMM rainfall features in relation to the terrain topography and elevation. The seasonal distribution (South-West Monsoon) and performance evaluated for 14 different topographical gauge locations. District wise prominent station names and their corresponding IMD, TRMM, bias value and elevation from mean sea level described in table 2. While analyzing the spatial pattern, it identified that the amount of monsoonal rainfall increases from South (Thiruvananthapuram -807 mm) to North (Kasaragod -2725 mm) gradually. It indicates that the Northern part of Kerala receives more rainfall than the Southern part. Similarly, Simon and Mohankumar (2004) reported that the North region receives more rainfall than the South region and stated that apart from the hilly region, the coastal region also receives abundant rainfall. The station data and TRMM data comparison obtained reasonable similarity with a correlation coefficient of 0.85, coefficient of determination of 0.72 (Figure 8) and RMSE of 453.02 mm. The bias value indicates that in most cases, TRMM underestimates the station data. A higher bias value was obtained primarily for higher elevation stations of Pathanamthitta (-0.3), Idukki (-0.4), Wayanad (-0.3) and one lower elevation station, Kozhikode (-0.3). The TRMM outcomes are sometimes diverge, especially during the monsoon, and it has several significant underestimations in assessing values at the higher elevation stations viz., Pathanamthitta (101 m), Idukki (900 m) and Wayanad (1018m) compared to the lower elevation stations (<80 m) (Figure 9). However, the research revealed that obtaining rainfall amounts in mountainous areas was challenging using the TRMM satellite data. Both Almazroui (2011) and Milweski et al. (2015) have recognized the significance of undulating topography in influencing satellite and rain gauge accuracy. They all exposed a low level of agreement between TRMM and rain gauges at the high altitudes. While the TRMM satisfactorily predict the rainfall data for entire lower elevation stations except for Kozhikode station (10m). When there is much convection over land, the TRMM precipitation radar algorithm likely underestimates the amount of precipitation (Iguchi et al., 2009; Kozu et al., 2009). Kumar et al., (2020) observed the dynamic characteristics of convective systems that resulted in widespread rainfall over west coast of Kerala. However, dense clouds around India’s West coast (8-20º N) indicate convective activity linked with low-pressure systems and active monsoon conditions in the southwest monsoon zone. Additionally, the higher level of emission from land obstructs the signal received by satellite sensors. As a result, an efficient statistical correction scheme may include stratifications for convective and stratiform precipitation, as well as rainfall over ocean and land (Chang et al., 2013). It may be the reason for the underestimation of station data of Kozhikode and some lower elevation station data by TRMM. However, because only ten years of station wise precipitation data were available, a fair conclusion could not be drawn in this context. Hence, a further detailed study using daily station wise data is necessary. 

Figure 7. Weather stations of IMD in Kerala

Figure 8. Scatterplot between the rainfall recorded at IMD stations and TRMM data in Monsoon season (2009-2019)

Figure 9. Rainfall recorded at IMD stations and TRMM data in Monsoon season (2009-2019) with the elevation

4.2 Rainfall Characteristics

Between 1998 and 2019 (Table 3), the yearly average rainfall in Kerala, according to TRMM, was 2515 mm, with a standard deviation of 351.1 mm. While observatory data indicates it about 2722.3 mm and a standard deviation of 420.9 mm. TRMM measured seasonal rainfall of 410.6 mm, 1557.6 mm, 486.6 mm, and 60.6 mm during the pre-monsoon, southwest monsoon, post-monsoon, and winter seasons, respectively, while observatory data indicates 399 mm, 1794.5 mm, 470.1 mm, and 58.7 mm, respectively. Annual rainfall has a coefficient of variation about 14% for TRMM data and 15.5% for IMD data, indicating that the rainfall very consistently. July has the highest TRMM rainfall (496 mm) and accounts for 19.75% of the yearly rainfall (2515.5 mm), followed by June (18.66 per cent). August and September are the rainiest months, accounting for 14.2% and 9.3% of yearly precipitation, respectively. January has the least rainfall (7.7 mm) and contributes barely 0.3% to the yearly precipitation. According to IMD data, June (599.2 mm) is the wettest month, while January (7.1 mm) is the driest. This discrepancy is due to the underestimation of observatory rainfall by TRMM sensor.

 

Table 3. Monthly and seasonal means of rainfall over Kerala (1998-2019)

IMD-Rainfall					TRMM-Rainfall
Month/ Season	Mean (mm)	SD (mm)	CV (%)	Contribution (%)	Mean (mm)	SD (mm)	CV (%)	Contribution (%)
January	7.1	7.5	106	0.26	7.7	8.3	107.5	0.31
February	18.6	19.6	105.4	0.68	20.8	26	124.9	0.83
March	46.1	48.8	106	1.69	52.3	51.5	98.5	2.08
April	114.2	53.3	46.7	4.2	137.2	59.1	43	5.45
May	238.7	154.6	64.8	8.77	221.1	118.3	53.5	8.79
June	599.2	148.5	24.8	22.01	469.4	108.7	23.2	18.66
July	527.1	169.3	32.1	19.36	496.8	145.9	29.4	19.75
August	399.5	199.2	49.9	14.67	357.5	154.3	43.2	14.21
September	268.7	133.3	49.6	9.87	233.9	117.4	50.2	9.30
October	327	129.5	39.6	12.01	327.6	117.5	35.9	13.02
November	143.1	69	48.2	5.26	159	58.5	36.8	6.32
December	33	29.5	89.3	1.21	32.1	26.8	83.3	1.28
Winter	58.7	38.2	65.1	2.16	60.6	34.6	57.0	2.41
Pre-Monsoon	399	152	38.1	14.66	410.6	117.3	28.6	16.32
SW-Monsoon	1794.5	374.4	20.9	65.92	1557.6	313.7	20.1	61.92
Post-Monsoon	470.1	155	33	17.27	486.6	126.1	25.9	19.34
Annual	2722.3	420.9	15.5	100	2515.5	351.1	14	100

 

 

For IMD data, in January and March (106%), the coefficient of variation was at its maximum, followed by February (105.4%) and December (89.3%), and the lowest in June (24.8%) and July (32.1%). The southwest monsoon (June-September) accounts for 65.92% of yearly precipitation. Pre-monsoon (March-May), post-monsoon (October-November), and winter (December-January) rainfall contribute 14.66%, 17.27%, and 2.16% to the yearly total, respectively. Seasonal rainfall throughout the monsoon season (June-September) is consistent, with a coefficient of variation of 20.1% for both data sets. Meanwhile, rainfall in the winter is unpredictable due to the higher coefficient of variation (65.1%).

4.2.1 Annual Trends

Between 1998 and 2019, the TRMM-derived mean annual rainfall across Kerala exhibited an insignificant decreasing trend. The Mann Kendall test yields a -9 S value. The magnitude of the trend determined by Sen’s slope is of -2.47 mm/year. At the same time, IMD observatory data from 1998 to 2019 over Kerala revealed an insignificant increasing trend. A positive S value of 21 obtained through the Mann Kendall test. The trend’s magnitude determined by Sen’s slope of 10.05 mm/year. The discrepancy is due to the underestimation of rainfall during the rainy months and overestimated during the drier months.

Figure 10 shows that the TRMM under predicts the rainfall in most cases, and the trend line of TRMM data is slightly decreasing, while the IMD data is relatively stable. The primary cause of the disparities in the two data sets is an underestimation of rainfall. It shows that the TRMM data can follow the trend of observatory data in most years. However, it is essential to note that no significant trend observed in the annual scale of rainfall during 1998-2019. As stated above, the validation of the data set is necessary for the TRMM studies in Kerala.

 

Figure 10. Annual rainfall trends over Kerala (1998 to 2019)

 

4.2.2 Monthly Trends

Individual monthly rainfall behaviour examined using the Man-Kendall test and Sen’s slope (Table 4). It is worth noting that rainfall in January exhibited a declining trend that was statistically significant at the 0.05 level for the TRMM data, whereas observatory data showed a decreasing trend but was not statistically significant. March, July, August, September, and December rainfall had an insignificant increase in TRMM and observatory data. January, February, April, May, June, and October rainfall revealed a declining trend that is statistically insignificant for both TRMM and observatory data. In November, the observatory data indicated an increasing trend, while the TRMM data indicated a decreasing trend, and both data indicated an insignificant trend.

 

Table 4. Man-Kendall test Sen’s slope statistics of monthly and seasonal rainfall

	IMD		TRMM
Month/ Season	Mann-Kendall (S)	Sen’s Slope (Q)	Mann-Kendall (S)	Sen’s Slope (Q)
January	-23	-0.1	-76	-0.32*
February	-21	-0.29	-0.29	-0.38
March	49	1.21	39	0.93
April	-23	-1.29	-19	-1.16
May	-15	-1.63	-1	-0.21
June	-13	-1.21	-13	-2.01
July	37	5.45	31	3.15
August	53	9	39	6.1
September	15	4.2	5	1.12
October	-47	-4.37	-59	-6.94
November	39	1.2	-11	-0.79
December	51	1.23	39	0.94
Winter	1	0.13	-31	-0.73
Pre-Monsoon	-25	-6.44	-31	-4.67
SW-Monsoon	41	17.82	29	11.84
Post-Monsoon	-17	-2.02	-65	-7.91**
* Significant values at a<0.05, ** Significant values at a<0.1

 

 

Considering observatory data (IMD) from 1998 to 2019, there was a drop in June’s contribution to yearly rainfall (24 to 22 %) from 1991 to 2005 (Table 5). When comparing to rainfall data for June, the contribution of rainfall is more stable in July. At the same time, rainfall contributions increased in August (12.2 to 14.7%) and September (7.6 to 9.9%). A slight decrease in the yearly rainfall contribution observed throughout October and November. Over Kerala, the percentage rainfall contribution during the southwest monsoon increased while the post-monsoon season is decreasing and relatively consistent during the pre-monsoon and winter seasons. Recent decades have seen the development of these phenomena. 

 

Table 5. Seasonal rainfall contribution to annual rainfall during 1998 to 2019 and 1991-2005

	Krishnakumar et al. (2009)	IMD	TRMM
Month/ Season	1991-2005 (%)	1998-2019 (%)	1998-2019 (%)
June	24.0	22.0	18.7
July	19.7	19.4	19.8
August	12.2	14.7	14.2
September	07.6	09.9	09.3
October	13.3	12.0	13.0
November	06.3	05.3	06.3
Winter	02.4	02.2	02.4
Pre-Monsoon	14.4	14.7	16.3
SW-Monsoon	63.6	65.6	61.9
Post-Monsoon	19.5	17.3	19.3

 

 

4.2.3 Decadal Trends

The monthly rainfall data is divided into two sections to determine the variability and trend of rainfall over two time periods: 1998-2008 and 2009-2019 (Figure 11). Between 1998 to 2008, yearly rainfall decreased at a rate of 3.27 mm/year and 5.50 mm/year for observatory data and TRMM data, respectively, with a Man-Kendall value of -3 and -1. It is also worth noting that rainfall increased between 2009 and 2019 using both observatory and TRMM data, with a Mann-Kendall S value of 7 for both and a rate of 24.58 mm/year and 34.50 mm/year, respectively. However, the trend found throughout these periods is statistically insignificant. Extreme precipitation events have strengthened the frequency and intensity in India over the previous few decades (Roxy et al., 2017) and the frequency of land falling tropical cyclones over the Arabian Sea has been increasing (Mohapatra, 2021). Thus, it could be the rising trend in rainfall over recent years. In addition, Mukherjee et al. (2018) observed an increase in heavy rainfall in India due to anthropogenic global warming, using data from the coupled model inter-comparison project 5 (CMIP5).

 

Figure 11a. Annual rainfall trends over Kerala during 1998 to 2008

 

Figure 11b. Annual rainfall trends over Kerala during 2009-2019

 

To explore the recent rainfall trend over Kerala, the most recent years, viz. 2009 to 2019 monthly, seasonal, and yearly rainfall, were analyzed using mean, SD, CV and percentage of contribution of annual rainfall (Table 6). The percentage contribution of rainfall during southwest rainfall increased from 1871 to 2005 (base period), while the post-monsoon and pre-monsoon showed a slight decrease compared to the base period. Compared to the base period, the months of August and October show a slight decrease in percentage contribution, while the rest of the month shows an approximately equal contribution to rainfall. The vital variation observed in the case of the variability of monthly and seasonal rainfall. The CV during the June, August, October months and the SW-monsoon and post-monsoons are increasing. It indicates that the consistency of rainfall during these months and seasons has decreased in the last decade, which means that uncertainty is increasing. At the same time, it is worth noting that annual rainfall is increasing compared to the baseline period. The frequent drought and flood years are the primary causes of these variations.

 

Table 6. Monthly and seasonal means of rainfall during 1871 to 2005 and 2009-2019

IMD-Rainfall (1871-2005)					IMD Rainfall (2009-2019)
Month/ Season	Mean (mm)	SD (mm)	CV (%)	Contribution (%)	Mean (mm)	SD (mm)	CV (%)	Contribution (%)
January	12	17	146.1	0.26	7.4	7.5	100.6	0.27
February	17	19	115.4	0.68	15	17.8	118.8	0.54
March	36	28	78.5	1.69	46.4	22	47.5	1.68
April	112	52	46.5	4.2	105.2	55	52.3	3.81
May	246	159	64.6	8.77	198.2	86.3	43.5	7.18
June	684	194	28.4	22.01	607.9	196.3	32.3	22.01
July	632	209	33.2	19.36	557	181	32.5	20.16
August	373	157	42	14.67	466.1	255.1	54.7	16.87
September	224	122	54.7	9.87	265.5	113.8	42.9	9.61
October	288	108	37.5	12.01	286.1	122.1	42.7	10.36
November	156	85	54.4	5.26	164	82	50	5.94
December	38	39	102.1	1.21	43.6	26.8	61.6	1.58
Winter	65.3	65.3	14.4	2.16	66	38.4	58.2	2.39
Pre-Monsoon	393.7	393.7	41.5	14.66	349.8	103.9	29.7	12.66
SW-Monsoon	1913.5	377.7	19.7	65.92	1896.5	403.8	21.3	68.65
Post-Monsoon	444.1	138.4	31.2	17.27	450.1	169	37.5	16.29
Annual	2722.3	420.9	14.4	100	2762.4	493.4	17.9	100

 

The increase in southwest monsoon rainfall and decreasing post-monsoon season rainfall during 1998-2019 is a significant finding in this study. The increase in rainfall more pronounced from July to September but not in June during the monsoon season. However, this trend is statistically insignificant. All other studies concur with Kerala’s southwest monsoon rainfall is on the decline (Soman et al., 1988; Rupa Kumar et al., 1992; Guhathakurta and Rajeevan, 2007; Krishnakumar et al., 2009). However, this study shows an increasing insignificant trend. The pivotal factor is that all of these analyses take the long-term climate of the 20^th century into account. Moreover, tropical cyclones have increased in frequency by a factor of two over the Bay of Bengal and Arabian Sea (Singh et al., 2001; Mohapatra, 2021). As a result, it could cause the current insignificant increase in the trend of rainfall.

4.3 Seasonal Trends

4.3.1 Winter (December, January and February)

The TRMM data show a decreasing trend in winter rainfall, while the observatory data show a rising trend, which is not statistically significant (Figure 12a and 12b). The only season in which contradictory results received is the winter season. When observatory data (IMD) from 1998 to 2019 analyzed, the rainfall contribution from 1991 to 2005 remained relatively constant (2.4 to 2.2%) (Table 5). It also identified that TRMM overestimates rainfall in these months.

4.3.2 Pre-Monsoon (March, April and May)

During 1998 and 2019, both observatory and TRMM data indicated a reduction in pre-monsoon rainfall. The rainfall trend observed to be slightly decreasing during this period (Figure 12). However, when comparing winter rainfall (observatory data) from 1991 to 2005 to the current time, the pre-monsoon rainfall appears to be very consistent (14.4 to 14.7%) (Table 5).

 

Figure 12a. Time series of seasonal rainfall derived from TRMM

 

Figure 12b. Time series of seasonal rainfall derived from recorded data at IMD stations

 

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

The authors are thankful to the editor and reviewers for their valuable suggestions and contributions to this article.

CV: Coefficient of Variation; IMD: India Meteorological Department; JAXA: Japan Aerospace Exploration Agency; NASA: National Aeronautics and Space Administration; RMSE: Root Mean Square Error; SD: Standard Deviation; TMPA: TRMM Multi-Satellite Precipitation Analysis; TRMM: Tropical Rainfall Measurement Mission.