1Department of Mathematics, The ICFAI University Meghalaya, Tura-794 101, Meghalaya, India
2Department of IT, The ICFAI University Meghalaya, Tura-794 101, Meghalaya, India
3Department of Statistics, Assam down town University, Guwahati781 026, Assam, India
4Department of Mathematics, Manipur International University, Imphal West-795 140, Manipur, India
*Corresponding Author: Shonahar Ali, Department of Mathematics, The ICFAI University Meghalaya, Tura-794 101, Meghalaya, India, Email: mdshonaharali@iumeghalaya.edu.in
The paper is an exploration of rainfall variability in the South West Garo Hills, one of the areas that is very sensitive to monsoonal variations of Meghalaya within India.
The monthly rainfall from October 2014 to August 2024 were undertaken to analyses and capture short term variability and localized structural changes. Though this data set duration does not represent long term climatic variability but provides valuable insights into stochastic rainfall behavior and agricultural implications at the district scale. The Shapiro-Wilk, Bartels and Wald Wolfowitz tests were performed to test the normality, randomness and stationarity of the mean respectively. To detect trends, the Mann-Kendall test and its variants, along with Sen’s slope estimator were utilized. Pettitt test, Buishand tests and CUSUM-PELT were used to find the Change-point detection. The visualization of localized changes and residual behavior was made using segment-wise linear regression technique and additive seasonal decomposition.
The test provided rainfall series was non-normal, non-random and non-stationary. The Mann-Kendall and its variants revealed insignificant monotonic trends with Z≈ -0.50, p>0.6, Sen’s slope ≈ -0.012 mm/month. However, the Pettitt’s test and segmented regression detected a short-term structural change between June 2019 and July 2020 along with p = 0.045, R2 = 0.317. Moreover, Seasonal decomposition indicated weak seasonality with stochastic residuals. The rainfall variability showed stochastic nature with localized structural shifts but no persistent long-term trends during the study periods. Only univariate statistical methods are insufficient to capture rainfall anomalies in the study area. Future studies should incorporate machine-learning approaches to improve rainfall prediction and early warning systems in Northeast India, particularly in Meghalaya. The novelty of this research is the combination of multiple statistical methods and visualization techniques.
Change-point detection, Modified Mann-Kendall test, Rainfall variability, Seasonal decomposition, Sen’s slope
