India's climate is a mosaic of mountain ranges, coastal plains, and inland plateaus, where monsoon winds and ocean currents weave into a complex pattern. Neighboring papers in our archive have already shown how climate sends early signals, how to distinguish weather noise from climate signals, and how old weather records come alive. But they all look at averages or isolated events. This work takes a step sideways: it examines not "what is happening" but "how the dynamics themselves change"—whether the climate becomes more predictable or, conversely, more chaotic. Such a view is especially important for a country where the same monsoon can bring floods to one state and drought to a neighbor.
The researchers took NCEP reanalysis data at 2.5° × 2.5° resolution for 15 locations covering all of India's climate zones according to the Köppen–Geiger classification, from the cold arid deserts of Ladakh to the humid tropics of Kochi. After removing seasonality and smoothing with a moving average, they reconstructed the phase space using Takens' delay embedding. Then, in a sliding window of about 7 years (with a step of 8 months), they built recurrence plots and recurrence networks. From these, they extracted four measures: determinism (DET), laminarity (LAM), link density (LD), and characteristic path length (CPL). High DET and LD indicate regular, nearly periodic dynamics; low values indicate irregular, stochastic dynamics. To identify significant shifts, they used bootstrap confidence intervals.
For temperature, the 15 locations split into 5 clusters that do not coincide with climate zones. For example, Ladakh, Manali, and Tawang—cold, high-altitude sites—formed one cluster, while Mumbai and Bhopal, separated by hundreds of kilometers, fell into another. Nearly all locations showed a drop in DET in 1972–1979: the dynamics became more irregular. This coincides with the strong El Niño of 1972–73 and La Niña of 1973–76. After 2000, DET rose again—dynamics became more regular. For relative humidity, the picture is different: regularity prevailed until 1970, then scattered shifts toward irregularity began, and clusters formed differently (Tawang turned out to be isolated). The CPL and LD measures confirmed these transitions: CPL increased in the 1970s and decreased after 2000.
Unlike neighboring papers that analyze averages or extremes, this article looks at the structure of dynamics. It shows that climate shifts are not just changes in temperature or humidity but a reorganization of the very nature of variability. The clusters derived from recurrence measures do not match the Köppen–Geiger classification: dynamics unite regions that climatologists considered different. This means that understanding climate risks requires accounting not only for geography but also for nonlinear links between the atmosphere, ocean, and land. The work aligns with research on climate variability but differs in applying recurrence analysis to a long, country-wide data series rather than to individual events.
The topic could develop by including additional variables—precipitation, pressure, wind speed—and higher spatial resolution. It would be interesting to compare dynamic clusters with general circulation models to understand which physical mechanisms generate a given complexity. Also promising is the study of links between clusters: how changes in one region propagate to another via atmospheric teleconnections.
The results could be useful for climate monitoring and regional planning in India, especially for agriculture and water resource management. They also provide a methodological benchmark for analyzing climate dynamics in other countries with pronounced monsoon circulation.
The next step could be to test the robustness of clusters using other datasets (e.g., satellite observations) and other complexity measures. It is also important to investigate how local anthropogenic factors—urbanization, irrigation, deforestation—affect dynamic transitions.
The work directly relates to an unresolved problem in climatology: how to distinguish internal variability of the climate system from external forcing (e.g., greenhouse gases). Recurrence analysis provides a tool for detecting regime shifts but does not explain their cause. The question remains open as to why some regions are more sensitive to ENSO than others, and how this sensitivity changes over time.
The analysis window of 5000 time steps corresponds to about 7 years—comparable to the typical period of El Niño. This choice allowed capturing an entire ENSO cycle within one window and seeing its influence on dynamics.