Exploring dependence of COVID-19 on environmental factors and spread prediction in India

The study addresses how socio-behavioral measures (notably social distancing and lockdowns) and environmental factors (temperature and relative humidity) influence COVID-19 spread in India. Following the initial outbreak in Wuhan and the first Indian case on January 30, 2020, India implemented nationwide lockdowns beginning March 25, 2020 to curb community transmission. Given evidence that coronaviruses’ survival and transmissibility can be influenced by temperature and humidity, and recognizing India’s challenges with dense populations and social distancing, the research aims to: (1) predict COVID-19 spread and mitigation timelines in India under differing social distancing scenarios using an SEIR model; and (2) evaluate dependence of daily COVID-19 cases on temperature and RH using statistical analyses. The work is positioned to inform public health interventions by quantifying the relative roles of behavior versus environment in shaping epidemic dynamics.

Prior research indicates environmental conditions affect respiratory viruses, including SARS-CoV-2. Studies suggest lower humidity and lower temperatures can enhance viral survival and transmission, while higher temperatures may reduce viability via disruption of viral lipid envelopes. Laboratory studies on SARS-CoV and MERS-CoV indicate enhanced persistence in cooler, drier conditions and potential for aerosol transmission. Emerging evidence also links air pollution to increased susceptibility to SARS-CoV-2. Time-series and cross-regional analyses have reported associations between temperature/humidity and COVID-19 incidence and mortality, though findings are heterogeneous and often confounded by interventions and behavior. These mixed results underscore the need to jointly analyze environmental parameters, socio-behavioral measures, and epidemiological dynamics.

Two complementary approaches were used: (1) SEIR epidemiological modeling for India under contrasting social behavior scenarios; and (2) statistical analysis of environmental dependence using Response Surface Methodology (RSM), ANOVA, Pearson correlation, and two-sample t-tests at state and city scales.

SEIR model: A modified SEIR (Susceptible-Exposed-Infectious-Removed) framework simulated COVID-19 spread in India under two scenarios: Case A (strict lockdown/social distancing compliance; baseline without breaches) and Case B (with community breaches, interstate movements, and phased unlocks). Model parameters included transmission probability β (time-varying, weekly), incubation rate (k), average recovery/removal rate γ=1/14 day⁻¹, and contact/mobility metrics. Inputs were derived from WHO India situation reports and mobility proxies; population data from the Census of India. The model was coded in Python 2.7.5. Simulations spanned from January 31, 2020 through early August 2020, with specific calibration periods (Case A initially through early April; Case B including data through June 9 and unlock phases). Weekly estimates of transmission were used, incorporating reported mass-gathering breaches (Table 3). Model outputs included S(t), E(t), I(t), R(t) trajectories. Short-term validation of predictions versus reported data was performed over four weeks, computing percentage errors.

Environmental/statistical analysis: To assess temperature (T) and relative humidity (RH) effects on daily COVID-19 cases per day (CPD), analyses were conducted at state level for Maharashtra (MH) and Karnataka (KR) and at city level for Mumbai (MUM), Kasaragod (KGD), Srinagar (SNG), and New York (NYK). Environmental data (24-hour averages including day/night) were drawn from CPCB stations and averaged for states; city-level data used local records. A 6-day moving average of T and RH with a 1-day onset was aligned to CPD on the seventh day to reflect incubation and reporting lags. RSM employed a full quadratic model (main effects, squared terms, interaction) to relate coded environmental variables (−1 to +1 per site) to CPD; coefficients were estimated by least squares. Significance was assessed via ANOVA (F- and p-values), with model adequacy evaluated by R-Sq and adjusted R-Sq and residual diagnostics (normality, homoscedasticity). Pearson correlation matrices quantified linear associations among T, RH, and CPD. Two-sample t-tests compared mean environmental conditions between regions/cities to establish distinct climates. Statistical analyses were conducted using MINITAB 14.

SEIR modeling:

• Case A (strict social distancing): Predicted total infected cases would be constrained to approximately 10,050 by mid-June 2020 with a decreasing trend, reflecting the impact of stringent lockdowns and restricted interstate movement (up to June 16). A decline in infections was noted around May 19.
• Case B (with breaches and unlocks): Community spread by late April/early May led to a sharp rise, with infected cases exceeding approximately 550,827 by end of June 2020. On June 10, total infected in Case B were 33.64 times those in Case A. Mass gatherings and policy relaxations (Table 3) contributed substantially to increased transmission.
• Validation: Four-week validation showed percentage errors typically ~1–4% (average ~2%) for infected, recovered, and susceptible counts (e.g., infected on 27 June: actual 529,000 vs predicted 521,703; error 1.2%).

Environmental dependence:

• Distinct climates: Two-sample t-tests showed significant differences in environmental conditions across most state/city pairs (e.g., many comparisons P<0.001), supporting analysis across diverse climates.
• State-level RSM: Maharashtra exhibited strong model fit (R-Sq ≈ 83.63%) with significant linear, squared, and interaction terms of T and RH (multiple terms P<0.001). Karnataka showed a moderate fit (reported around 57% R-Sq) with several significant terms (e.g., RH squared P=0.001; interaction near significance).
• City-level RSM and ANOVA: • Kasaragod: Reasonable fit (R-Sq ≈ 47.8%); temperature showed significant effects (p<0.05) with notable squared and interaction term significance (P<0.001). Directionality indicated temperature’s main negative effect could be offset by positive non-linear terms. • Srinagar: Temperature was the only significant main effect (F=7.18, P=0.009) with R-Sq ≈ 54.4%, positively correlated with CPD. • Mumbai: Main effects of T, RH, and squared T were all significant (P<0.001), with overall fit modest (R-Sq ≈ 24.9%). Main effect of temperature was negative, but a positive squared temperature term dominated, yielding a net positive association in the observed range. • New York: Temperature showed a negative association with CPD; interaction between T and RH was significant (P=0.001) in RSM.
• Pearson correlations (linear): Patterns varied by location—New York CPD correlated negatively with temperature (≈ −0.55). Srinagar showed positive CPD–T correlation (≈ 0.66). Kasaragod showed weak negative CPD–T correlation (≈ −0.23). State-level correlations indicated near-null temperature–CPD in Maharashtra (≈ 0.003) and negative in Karnataka (≈ −0.56). RH relationships varied and were generally less consistent.
• Overall: Temperature displayed positive associations with CPD in the Indian cities studied (Mumbai, Srinagar, Kasaragod) but negative in New York, suggesting context-specific effects and confounding by social distancing and other socio-environmental factors. The role of RH in affecting daily cases remained unclear. Residual analyses indicated acceptable model assumptions at analyzed scales.

The SEIR simulations demonstrate that strict social distancing and lockdowns substantially suppress transmission and accelerate curve flattening, whereas breaches, mass gatherings, and phased reopening markedly elevate case counts and prolong the epidemic. Short-term validation supports model reliability within a dynamic policy environment.

Environmental analyses indicate that while meteorological factors (temperature and RH) can associate with daily case counts, their effects are often overshadowed by socio-behavioral factors. City-level results show heterogeneous relationships: positive temperature–CPD correlations in Indian cities likely reflect confounding by high population density, mobility constraints, and difficulties maintaining social distancing (e.g., Mumbai), whereas New York exhibits the expected negative temperature association. RH effects are inconsistent and less conclusive. Linear correlations may fail to capture non-linear dependencies that RSM reveals, highlighting the need for caution when interpreting simple correlations. State-level averaging can obscure spatial-temporal heterogeneity, motivating city/site-level modeling.

Collectively, findings argue that social distancing remains the primary lever for controlling COVID-19 spread, with environmental factors exerting secondary, context-dependent influences that should nonetheless be incorporated into epidemiological models for better predictive performance and policy planning.

This study integrates SEIR modeling with environmental statistical analysis to quantify COVID-19 spread in India and assess the influence of temperature and relative humidity. Results show that strict social distancing and lockdown policies are decisive in curbing transmission, while breaches and reopening phases can amplify cases by orders of magnitude. Environmental variables exhibit measurable but context-sensitive associations with daily cases; temperature effects vary across regions and can be masked by socio-behavioral dynamics, whereas RH effects remain ambiguous.

The work underscores the need to: (1) incorporate environmental parameters into epidemiological models like SEIR; (2) use non-linear modeling (e.g., RSM) and city-level analyses to capture complex relationships; and (3) undertake coordinated, controlled laboratory experiments and comprehensive modeling to establish causal mechanisms. Future research should leverage de-trended anomaly frameworks, higher-resolution spatiotemporal data, and integrated socio-environmental covariates to refine predictions and support targeted public health interventions.

• State-level environmental analyses rely on spatially averaged temperature and RH, potentially masking intra-state heterogeneity.
• Short-term validation (four weeks) limits assessment of long-term predictive accuracy in a rapidly changing policy and behavioral landscape.
• Confounding by social distancing, mobility changes, and socio-political factors complicates attribution of environmental effects.
• Relative humidity’s role remains inconclusive; linear correlations may be insufficient to capture non-linear relationships.
• Limited set of cities analyzed; generalizability across diverse urban and rural contexts may be constrained.
• Model parameters (e.g., transmission rates) estimated from available data and literature may introduce uncertainty; some reported statistics/tables contain inconsistencies.
• Environmental data aggregation and 6-day moving averages with assumed lags may not fully reflect exposure timing and reporting delays.