Next Article in Journal
Landslide Risk Associated with Glacier Tourism in the Mt. Everest Region (Sagarmatha National Park), High-Mountain Nepal
Previous Article in Journal
A Systematic Review of River Discharge Measurement Methods: Evolution and Modern Applications in Water Management and Environmental Protection
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Earth
Volume 7
Issue 2
10.3390/earth7020042
earth-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Analyzing the Effect of the 2015/16 Catastrophic El Niño Event on Wildfire Emissions in Southern Africa Using Lagged Correlation and Interrupted Time-Series Causal Impact Technique

by
Lerato Shikwambana
1,2,
Mahlatse Kganyago
3,* and
Xiang Zhang
4,5
1
Earth Observation Directorate, South African National Space Agency, Pretoria 0001, South Africa
2
Unit for Environmental Sciences and Management, School of Geo- and Spatial Science, North-West University, Potchefstroom 2520, South Africa
3
Department of Geography, Environmental Management & Energy Studies, University of Johannesburg, Johannesburg 2000, South Africa
4
National Engineering Research Center of Geographic Information System, China University of Geosciences, Wuhan 430074, China
5
Shenzhen Research Institute, China University of Geosciences, Shenzhen 518063, China
*
Author to whom correspondence should be addressed.
Earth 2026, 7(2), 42; https://doi.org/10.3390/earth7020042
Submission received: 24 January 2026 / Revised: 19 February 2026 / Accepted: 28 February 2026 / Published: 6 March 2026
Browse Figures
Versions Notes

Abstract

Southern Africa is highly sensitive to climate variability associated with the El Niño Southern Oscillation (ENSO), which strongly influences hydroclimate, vegetation dynamics, and atmospheric composition. This study examined the impacts of the 2015/16 El Niño on vegetation, meteorological conditions, and atmospheric emissions over Southern Africa using satellite observations and reanalysis data. Time-lagged cross-correlation analysis of seasonally adjusted time-series was applied to characterize synchronous and delayed interactions among vegetation indices, hydrological variables, meteorological drivers, and air-quality parameters. Bayesian causal impact analysis was further used to quantify El Niño-induced anomalies by comparing observed conditions with counterfactual scenarios representing the absence of the event. The results showed that vegetation greenness responds primarily to concurrent moisture availability, with strong positive associations between NDVI, precipitation, soil moisture, and canopy water. Moisture-related variables exert delayed influences on atmospheric composition, highlighting the role of wet scavenging and dilution. Carbonaceous aerosols (black carbon [BC] and organic carbon [OC]), particulate matter [PM2.5], and aerosol optical depth exhibit strong synchronous coupling, indicating a dominant biomass-burning source. The causal impact analysis reveals statistically significant and sustained post-2015 increases in fire-related emissions (carbon monoxide [CO], BC, OC, PM2.5, and aerosol optical depth [AOD]), particularly during austral winter and dry seasons. In contrast, precipitation, soil moisture, evapotranspiration, and vegetation greenness show persistent negative anomalies, reflecting widespread drought stress under elevated temperatures. Overall, the findings demonstrate that the 2015/16 El Niño amplified fire emissions while suppressing ecosystem functioning across Southern Africa, underscoring strong climate–fire–vegetation feedback with important air-quality and environmental implications.

1. Introduction

The El Niño Southern Oscillation (ENSO), characterized by El Niño and La Niña cycles, is an important phenomenon affecting weather and climate in various regions [1]. In southern Africa, El Niño is known to bring dry, hot air and cause dry conditions through teleconnections, while La Niña provides moist and cool conditions [2,3,4]. In the years when temperature anomalies in the Niño 3.4 region remain within ±1 °C, conditions are classified as neutral. Among the different cycles, El Niño is the most devastating to southern African ecosystems and food security, causing prolonged droughts, losses in livestock and crops, and floral and faunal biodiversity [5,6]. In particular, the 2015/16 strong El Niño has been dubbed the worst since 1980, causing widespread economic and agricultural losses across southern Africa [7,8]. Mainly, the drier conditions triggered extensive wildfires, and are frequently associated with high atmospheric emissions and poor air-quality [9,10].
While most vegetation types in southern Africa are fire-adapted and can benefit from periodic fires for regeneration, increasingly frequent and intense fires—exacerbated by climate anomalies such as El Niño—can exceed the ecological thresholds of some ecosystems, potentially threatening sensitive species and accelerating ecosystem degradation [10,11]. Moreover, the increase in wildfire events has significant implications for air-quality and climate [12,13,14]. Fires emit a wide range of atmospheric pollutants [15,16], including carbon dioxide (CO2), carbon monoxide (CO), methane (CH4), nitrogen oxides (NOx), and non-methane volatile organic compounds (NMVOCs), as well as particulate matter (PM2.5 and PM10) and black carbon (BC), which are known to cause health and environmental hazards [17,18]. These pollutants contribute to regional haze, reduced air-quality, and adverse health impacts for human populations. Fine particulate matter in particular can penetrate deep into the lungs, causing respiratory and cardiovascular diseases, while ozone precursors from wildfire smoke may lead to elevated surface ozone levels [17].
From a climate perspective, wildfire emissions not only release greenhouse gases that contribute to long-term warming but also produce short-lived climate forcers such as aerosols and black carbon that alter radiative forcing and cloud formation [19]. These emissions can further amplify climate feedback by modifying atmospheric chemistry and energy balance [20,21,22].
Wildfire activity is influenced by several key climatic parameters [23,24], including temperature, relative humidity, wind speed, precipitation, and soil moisture. Under El Niño conditions, southern Africa typically experiences significantly reduced precipitation, higher surface temperatures, lower relative humidity, and increased atmospheric stability—creating ideal conditions for fire ignition and propagation [3]. The combination of hot, dry, and windy conditions accelerates vegetation desiccation, making landscapes more flammable and increasing the likelihood of large-scale fires.
Indeed, several studies have demonstrated increasing wildfire emissions, fire density, and burned area under El Niño conditions, coinciding with lower precipitation and humidity, and elevated temperatures [9,25]. However, there remains a lack of studies that evaluate the lagged correlations and causal impacts of El Niño events on fire outcomes, including atmospheric emissions and meteorological variability.
Lagged correlation analysis involves assessing the time-delayed relationships between climatic variables and fire activity—contrasting with standard correlation, which assumes an instantaneous relationship [26]. This method has been previously applied in hydrological and agricultural studies, but less so in fire–climate interaction research. Meanwhile, causal impact analysis is gaining popularity in environmental science, owing to its ability to estimate the post-intervention effect of a phenomenon (e.g., El Niño) by comparing observed data to a counterfactual scenario—i.e., what would have happened had the event not occurred. Originating in econometrics, causal impact analysis remains underutilized in remote sensing applications related to wildfires and emissions.
Several techniques are available for causal inference, such as difference-in-difference (DiD) [27], geolift, and summation-based methods. These typically require a control or randomized group, which is often infeasible in ecological systems. As an alternative, interrupted time-series (ITS) analysis requires only a historical time-series with seasonal trends and the known timing of an intervention (e.g., the onset of El Niño), making it suitable for climate–fire studies. ITS can estimate a plausible counterfactual trajectory of variables such as wildfire emissions or air pollutants in the absence of El Niño conditions.
It is against this background that this study sought to analyze the lagged correlations and causal impact of the 2015/16 El Niño on meteorological parameters and wildfire-related emissions. Firstly, we performed time-series correlations between key meteorological indicators and various atmospheric emissions linked to fire activity to identify their temporal relationships. Secondly, each meteorological parameter and fire-emitted pollutant was subjected to ITS-based causal impact analysis to reveal the expected trajectories of emissions and weather variables had the El Niño not occurred. This dual approach provides a robust understanding of how extreme ENSO events shape both wildfire dynamics and atmospheric pollution levels, offering insights essential for mitigation planning, early warning systems, and long-term climate adaptation strategies. Among the previously extreme El Niño events, i.e., 1982–1983, 1997–1998, and 2015–2016, the latter was chosen in this study due to its recency and intensity.

2. Materials and Methods

2.1. Study Site

The study area, Figure 1a, encompasses a significant portion of southern Africa, geographically defined by the bounding box coordinates 7.71° S to 45.16° S latitude and 36.00° W to 0.38° E longitude. This extensive region incorporates several developing nations with substantial land areas, including the Democratic Republic of the Congo (DRC), Angola, Zambia, Malawi, Mozambique, Zimbabwe, Botswana, Namibia, and South Africa. A pronounced bioclimatic gradient characterizes the area, driven primarily by precipitation regimes. Countries proximal to or overlapping the equator in the western and central sectors (notably the DRC and northern Angola) are dominated by tropical rainforests, forming part of the Congo Basin, the world’s second-largest contiguous rainforest. These forests are increasingly interspersed with areas of shifting cultivation and permanent agriculture. Progressing southwards, the vegetation transitions dramatically. The arid western part features the hyper-arid Namib Desert, characterized by unique fog-dependent ecosystems and extremely low-growing shrublands. This gives way to the vast, semi-arid Kalahari sands, covered in acacia savannas and grasslands. The central and eastern portions of the region, covering much of Zambia, Angola, Zimbabwe, Botswana, northern South Africa, and Mozambique, are dominated by tropical and mesic savanna ecosystems, including the biologically rich miombo woodlands (dominated by Brachystegia, Julbernardia, and Isoberlinia species) and mopane (Colophospermum mopane) woodlands. These savannas are extensively utilized for both subsistence and commercial crop agriculture and support large-scale livestock production on rangelands. The eastern coastal belt and lower-lying areas, particularly in Mozambique, feature savanna-woodland mosaics, edaphic grasslands, and significant floodplain systems.
Fire constitutes a fundamental ecological process and management challenge across much of the study area, particularly within the savanna and grassland biomes that dominate the landscape. Fire regimes are highly variable, influenced by rainfall, fuel load accumulation, vegetation type, and human activity. Natural ignitions primarily occur via lightning during the late dry season. However, anthropogenic fires, set for purposes including pasture management, crop residue clearing, hunting, and wildfire hazard reduction, are widespread and often dominate the fire pattern. In mesic savannas like the miombo, fires tend to be frequent (often annual or biennial) and predominantly occur late in the dry season (August–October), resulting in high-intensity, stand-replacing burns. In more arid savannas and grasslands, fire frequency is generally lower and more variable, dependent on sufficient fuel accumulation following rainfall events.
Fire plays a complex role in this region (see Figure 1b). It is essential for maintaining grass dominance, nutrient cycling, and habitat heterogeneity in savannas, controlling bush encroachment, and stimulating regeneration in certain plant species. However, alterations to natural fire regimes, particularly excessively frequent late-season fires, and frequent drought conditions, or intense fires in sensitive ecosystems like forest fringes or riparian zones, can lead to biodiversity loss, soil degradation, reduced carbon sequestration, and habitat degradation for fire-sensitive species. Consequently, understanding and managing fire regimes is a critical aspect of biodiversity conservation and ecosystem service provision throughout the southern African region.

2.2. Data Description

2.2.1. Remotely Sensed Data

The Moderate Resolution Imaging Spectroradiometer (MODIS) is flown aboard NASA’s Terra and Aqua satellite platforms, providing near-global coverage with a wide swath of approximately 2330 km. This orbital configuration enables observations of the Earth’s surface at intervals of one to two days. MODIS collects radiometric measurements across 36 spectral channels spanning the visible to thermal infrared regions (0.405–14.385 µm) and delivers data at multiple spatial resolutions of 250 m, 500 m, and 1 km.

2.2.2. Reanalysis Data

The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), represents NASA’s most advanced atmospheric reanalysis designed for the satellite observation period. Developed by the Global Modeling and Assimilation Office (GMAO), MERRA-2 is generated using the Goddard Earth Observing System (GEOS) atmospheric model, version 5.12.4. The reanalysis provides a continuous global record from 1980 onward, with data products typically made available approximately three weeks following the close of each month. Atmospheric variables are distributed on a regular grid with a horizontal spacing of 0.625° in longitude and 0.5° in latitude, and are resolved vertically across 72 hybrid sigma–pressure levels extending from the Earth’s surface to 0.01 hPa. Comprehensive descriptions of the MERRA-2 system and data products are available in the literature [28,29]. For this analysis, wind speed and wind direction fields derived from MERRA-2 were utilized.

2.3. Data Analysis

2.3.1. Exploratory Analysis

Exploratory analysis is an approach used to investigate and understand the fundamental characteristics of a dataset prior to formal modeling or hypothesis testing [30]. The primary aim of exploratory analysis is to summarize the data, identify patterns and trends, detect anomalies or outliers, evaluate data quality, and explore potential relationships among variables [31,32]. This process typically employs descriptive statistical measures, such as central tendency and variability, alongside graphical techniques including histograms, boxplots, scatterplots, and correlation matrices [31]. In contrast to confirmatory analysis, which is driven by predefined hypotheses, exploratory analysis is inherently data-driven and flexible, enabling researchers to generate new hypotheses and gain intuitive insights that inform subsequent analytical or modeling decisions [30,33]. Consequently, exploratory analysis represents a critical initial step in scientific research and data-driven studies, ensuring that modeling assumptions and methodological choices are grounded in an informed understanding of the data structure and behavior. In this study, the summary statistics of the various emission parameters were computed using boxplots to compare conditions pre-2015 (before El Niño, [2010–2014]) and post-2015 (after El Niño, [2015–2017]) periods. Cohen’s d effect sizes were employed to quantify the magnitude and direction of changes in various emission, meteorological, and environmental parameters pre and post the 2015/16 El Niño event. Lastly, we analyzed the volatility of different parameters post-2015 as compared to pre-2015 using percentage change in standard deviation, resulting in two classes, i.e., decreased and increased volatility.

2.3.2. Lagged Correlation Analysis

Lagged correlation analysis is a statistical technique used to quantify the relationship between two time-dependent variables when one variable is shifted forward or backward in time relative to the other. In the context of data analysis, lagged correlation analysis evaluates how changes in one variable may precede, coincide with, or follow changes in another by computing correlation coefficients at a range of time lags. This approach is particularly useful for identifying delayed responses, temporal dependencies, and potential cause–effect relationships in sequential or time-series data, where immediate correlations may not fully capture system dynamics [34]. Examining correlations across multiple lags enables determination of the time delay at which the association between variables is strongest, providing insight into underlying physical, environmental, or socio-economic processes [35]. In this study, the time-lagged Pearson r correlation analysis, varying between 1 (strong positive correlation) and −1 (strong negative correlation) was applied to determine significant lag times when meteorological, vegetation, and emission parameters are highly correlated.

2.3.3. Causal Impact Analysis

In Earth observation (EO), air-quality, and climate impact studies, causal impact analysis is used to quantify the effect of a specific intervention or event on an environmental variable by comparing observed satellite- or model-derived time-series with a statistically constructed counterfactual representing conditions in the absence of that intervention. Typical interventions include the implementation of air-quality regulations, power-plant shutdowns, changes in fuel usage, lockdown measures, or the occurrence of extreme climate events such as heatwaves, droughts, or wildfires. The method is commonly implemented using Bayesian structural time-series (BSTS) models, which integrate long-term trends, seasonal cycles, meteorological covariates, and auxiliary EO datasets (e.g., temperature, wind speed, boundary-layer height) to predict expected pollutant or climate-variable behaviour under normal conditions [36]. Within air-quality applications, causal impact analysis enables the isolation of policy- or event-driven changes in pollutants such as nitrogen dioxide (NO2), sulphur dioxide (SO2), or particulate matter (PM2.5) from natural meteorological variability by explicitly modelling weather influences and background trends. This is particularly valuable in satellite-based studies, where observed concentration changes may otherwise be confounded by seasonal cycles or atmospheric dynamics [37]. Here, we employed Bayesian interrupted time-series (ITS) using CausalPy version 0.7.0 Python package (https://causalpy.readthedocs.io/en/stable/, accessed on 30 September 2025) to assess the impact of the 2015/16 El Niño on fire activity and emissions in Southern Africa. The ITS technique is a quasi-experimental approach that estimates the causal effect of a discrete intervention (i.e., El Niño event) by comparing observed outcomes to a modelled counterfactual based on the pre-intervention trend. In ITS, a linear regression model is fit to the time-series data before the El Niño and then used to forecast what would have happened in its absence (i.e., counterfactual). After the intervention point, deviations of the observed data from this forecasted baseline provide estimates of the causal effect attributed to the El Niño as well as uncertainty (i.e., 95% confidence intervals). CausalPy provides Instantaneous and Cumulative Bayesian Causal Effects, where the former refers to the discrepancy between the empirically observed outcome and the posterior predictive counterfactual distribution at any given time point within the post-intervention period, while the latter is derived through the temporal aggregation of these instantaneous differentials across the entire period of interest. The intervention period was taken as the start of 2015, while the pre- and post-El Niño periods were taken as 2010–2014 and 2015–2017, respectively. The pre-intervention period of 5-years comprised a sufficient time-series to enable robust modelling of the counterfactual, while minimizing the influence of prior strong El Niño events that could otherwise bias the estimates. The post-intervention period, in turn, allowed for the assessment of potential temporal lags between the El Niño episode and the variables under study. It also facilitated the modelling of effects that may persist beyond the recognized duration of the event (i.e., 2015–2016) by one year.

3. Results

3.1. Exploratory Analysis

Figure 2 shows the summary statistics of various emission, meteorological, and environmental parameters before and after the 2015/16 El Niño event. As shown, some parameters, such as black carbon (BC), organic carbon (OC), PM2.5, aerosol optical depth (AOD), and temperature (Temp), were relatively higher post-2015 than before. In contrast, parameters such as CO, SO2, sulfates (SO4), normalized difference vegetation index (NDVI), canopy water (CW), soil moisture (SM), precipitation (PCPN and burnt area (BA) were relatively lower post-2015 than before, while there was no apparent difference between the pre- and post-El Niño periods in evapotranspiration (ET).
Figure 3 shows the results of Cohen’s d effect sizes [38] as a standardized measure of the difference between pre-2015 and post-2015 periods, contributing additional information to the summary statistics explored in Figure 2. Overall, the results show that the response patterns of all parameters considered here were not statistically significant at α = 0.05. Among the parameters that had positive effect size between the periods before and after 2015, Temp had the greatest effect size, i.e., d marginally greater than 0.2, while other parameters, i.e., AOD, PM2.5, OC, BC, and SO4, had a small effect size (i.e., d < 0.2). On the other hand, SM has a medium negative effect size (d ~ 0.4), followed by CW, BA, NDVI, PCPN, ET, and SO2, which all have small negative effect sizes. CO’s effect size is 0, indicating neither a decrease nor an increase.
To further understand the differences in various emission, meteorological, and environmental parameters before and after the 2015/16 El Niño event, percentage changes in the standard deviation (volatility) were calculated (Figure 4). Figure 4 reveals divergent responses across different parameters. The parameters AOD, OC, PM2.5, BC, CO, ET, and Temp show increased volatility, while PCPN, SO4, NDVI, CW, BA, SM, and SO2 show decreased volatility. Comparatively, the AOD exhibits the most pronounced volatility increase, i.e., 17%, indicating amplified variability in the atmospheric particulate loading, probably from the increased biomass burning and dust under El Niño conditions. Similarly, the emission parameters such as OC, PM2.5, and BC also show moderate volatility increases, i.e., >7%, suggesting more episodic pollution events linked to variable fire regimes. On the other hand, ET, Temp, and CO have relatively low volatility increases after 2015, <5%, which could be linked to thermal stress and associated moisture flux variability under extreme conditions. Among the parameters that experienced decreased volatility after 2015, SO2 and SM show the highest declines in volatility, i.e., approximately −10%, followed by BA and CW with approximately 8%, and NDVI and SO4, which show moderate volatility decreases, i.e., <−5%, as compared to a period before 2015. The decreased volatility may be indicative of reduced response of these parameters to the 2015–2016 El Niño event, while variables showing increased volatility may be more sensitive to ENSO events.
Overall, the volatility results reveal generally similar patterns as Cohen’s d effect sizes (Figure 3) but reveal differences in magnitude of shifts. Aerosol-related parameters exhibit the clearest post-2015 intensification. For example, AOD, PM2.5, OC, and BC shows the largest increase in volatility (≈7–17%) and a positive Cohen’s d positive effect sizes, indicating not only greater variability but also a meaningful upward shift in mean conditions and a consistent post-2015 enhancement in aerosol loading. A modest increase in volatility and a positive effect size in Temp reinforces a post-El Niño warming signal. In contrast, several environmental and precursor variables demonstrated reduced volatility and negative effect sizes. The strongest decline was in SO2, showing ≈−10% volatility and ≈0, while the variability in other variables like SM, BA, and CW also declined notably but has small effect sizes, indicating substantial post-2015 reduction. NDVI and SO4 reflect moderate decreases in volatility and small negative effect sizes, suggesting weakened vegetation greenness and secondary aerosol formation relative to the pre-El Niño baseline. Meteorological drivers such as PCPN and ET show relatively small changes in both volatility and effect size, implying limited structural shifts. Overall, aerosol parameters exhibit the most coherent post-2015 amplification in both dispersion and central tendency, whereas sulfur species, soil moisture, and biomass burning indicators show consistent reductions, highlighting a divergent post-El Niño regime shift across emission, meteorological, and environmental variables.

3.2. Time-Lagged Cross Correlation Analysis

Figure 5 shows the results of the time-lagged cross correlation analysis using Pearson’s r correlation based on seasonally adjusted time-series from 2010 to 2017 to understand the complex interactions between vegetation, meteorological conditions, and atmospheric emission parameters. The lagged correlation matrix provides an overview of the statistically significant (α < 0.05) temporal and synchronous relationships among climatic, hydrological, and atmospheric emission variables. NDVI, a proxy for vegetation greenness, displays a strong positive contemporaneous correlation with soil moisture (SM; r = 0.64, lag = 0) and canopy water (CW; r = 0.59, lag = 0), suggesting that vegetation responds to concurrent moisture and energy-related water demands. NDVI also exhibits a moderately strong negative correlation with PM2.5 (r = −0.51, lag = 0) and evapotranspiration (ET; r = −0.49, lag = 0), indicating that greener conditions are associated with lower particulate pollution and atmospheric water loss. The absence of a lag suggests that these relationships are immediate rather than delayed, potentially due to vegetation acting as a sink for pollutants or a modulator of local water fluxes. On the other hand, PCPN shows a positive correlation with SM (r = 0.60, lag = 0) and CW (r = 0.68, lag = 0), emphasizing the role of precipitation in replenishing soil moisture and supporting vegetation water demands. However, it correlates negatively with PM2.5 (r = −0.37, lag = −1), ET (r = −0.57, lag = −1), and AOD (r = −0.29, lag = −2), indicating that rainfall reduces aerosol loading, with maximum scavenging effects occurring 1–2 months post-rainfall. The PCPN–ET relationship is strongly negative (r = −0.57, lag = −1), suggesting that high rainfall periods are followed by reductions in ET, possibly due to cloud cover or soil saturation effects.
Temperature (Temp) has a positive correlation with SO4 (r = 0.36, lag = 0) and CO (r = 0.27, lag = 0), highlighting that elevated temperatures coincide with higher sulfate and carbon monoxide concentrations, possibly via enhanced photochemical reactions and stagnation episodes. while it is inversely related to PM2.5 (r = −0.39, lag = −1) and BA (r = −0.32, lag = −1). The results also indicate that indicates that increased temperatures precede reductions in particulate concentrations, possibly due to increased vertical mixing or reduced biomass-burning activity in warmer months.
SM is closely linked to NDVI (r = 0.64, lag = 0), PCPN (r = 0.60, lag = 0), and CW (r = 0.53, lag = −1), underscoring its sensitivity to concurrent and preceding hydrological inputs. In contrast, it has negative correlations with SO2 (r = −0.47, lag = −1) and PM2.5 (r = −0.43, lag = −1) which further imply that wetter conditions contribute to pollutant dilution or deposition, with effects observed approximately one month later. CW, reflecting vegetation water use, correlates positively with PCPN (r = 0.68) and NDVI (r = 0.59) at lags of 0, and with CO (r = 0.34) with a lag of −1. This suggests that water availability drives CW and that it precedes increases in atmospheric CO by 1 month.
Among emission species, OC and BC are nearly collinear (r = 0.98) at a lag of 0, and both are highly correlated with PM2.5 (r = 0.70 and 0.67, respectively) and AOD (r = 0.81 and 0.80, respectively) also at lag 0, emphasizing their dominance in particulate and optical aerosol burdens. Similarly, these relationships are synchronous, indicating that emissions and optical responses occur simultaneously. SO2 shows strong positive correlations with OC (r = 0.60), BC (r = 0.55), and BA (r = 0.65) at lag 0, suggesting co-emission during biomass combustion events. The other parameters showing tightly coupled relationships with zero lags are PM2.5 and AOD (r = 0.90), confirming that columnar aerosol load is strongly governed by surface-level particulate matter. Biomass burning (BA) is found to correlate positively with OC (r = 0.64) and BC (r = 0.62) at lag 0, reinforcing the shared source profile.
Statistically insignificant lagged correlations occur between SO4 and NDVI, CW and CO, with weak negative r of −0.27 and less. Similarly, PCPN exhibits weak insignificant lagged correlations with OC, BC, and CO. The other similar weak correlations were between these emission parameters and ET as well as ET and SO2. It should be noted that for all the parameters exhibiting statistically insignificant correlations across all lags (indicated by ns), we chose to show lag 0 in Figure 5. In totality, the integration of lag structures reveals that most relationships are immediate, with only select variables—particularly PCPN and SM—exerting delayed effects on atmospheric composition and hydrological dynamics. Further information can be found in the Supplementary Materials (Figures S1–S6) which shows the variations of lagged Pearson’s r correlations over several (i.e., 12) lags.

3.3. Causal Analysis

3.3.1. Analysis of Pre-Intervention Conditions

The results of the causal impact of the 2015/16 El Niño were analyzed using the interrupted time-series (ITS) technique. The period starting from January 2015 was selected to indicate the start of the treatment date. A linear regression model was used to create counterfactual predictions of various parameters under the ‘what if?’ scenario of business as usual, i.e., ‘what would have happened if El Niño did not happen?’. By doing this, we could determine the causal impact and the cumulative causal impact by comparing the observed values to the counterfactual predictions (i.e., business as usual scenario). The results for burned area (BA) and various emission parameters (i.e., CO, BC, OC, SO2, SO4, PM2.5, and AOD) are given in Figure 6, while Figure 7 provides causal analysis results for vegetation and meteorological conditions (i.e., NDVI, PCPN, TEMP, SM, ET, and CW).
Overall, all the ITS models showed excellent pre-event fits (Bayesian R2 from 0.7 to 0.99) (Figure 6 and Figure 7). Hence, we can assume that post-2015 departures reflect the El Niño impact rather than model bias. Before 2015 (pre-2015–2016 El Niño event), the emission parameters such as BC (Figure 6c) and OC (Figure 6d) show generally similar seasonal and temporal patterns. In southern Africa, most biomass burning occur during June-July-August (JJA) and September-October-November (SON) seasons, which constitutes the fire season in this region [39]. Similarly, PM2.5 and AOD exhibit co-varying behavior, where both parameters exhibit bi-modal patterns, with highest peaks occurring during the fire season (i.e., dry season) and relatively low peak in December–January–February (DJF) (Figure 6c,g). There are occasional outliers such as August and September 2010 and DJF of 2012, which may have led to relatively lower fit in the pre-2015 time-series (i.e., Bayesian R2 of 0.87). Under a stable boundary layer, higher AOD tends to correspond to higher PM2.5 [40]. For example, in China, He et al. [40] found highest correlations (i.e., r = 0.61) between the two parameters in winter and at noon, while summer had lowest correlations (r = 0.47), mainly attributed to lower concentrations. These similarities mean that prior to 2015/16 El Niño, these parameters generally followed the annual cycles. Generally, other parameters, i.e., BA (Figure 6a), CO (Figure 6b), SO2 (Figure 6e), and SO4 (Figure 6f) also show annual patterns, with occasional outliers pre-2015. These outliers were less severe (i.e., did not negatively affect seasonal fit and interannual trends) of BA and CO during the pre-intervention period, while they caused reductions in ITS models fit for SO2 and SO4, resulting in Bayesian R2 values of 0.83 and 0.7, respectively (i.e., lowest among all emissions parameters).
The ITS models for vegetation (i.e., NDVI, ET, and CW) and meteorological parameters (i.e., PCPN, ST, and SM) in the pre-intervention period exhibited highest Bayesian R2 of >0.96, indicating that pre-event variability was less affected by outliers. Nonetheless, some years exhibited occasional and minor outliers in both extremes (see Figure 7). Since wildfires are mostly inversely proportional to most of these parameters, the interest is in lower extremes, particularly for NDVI, PCPN, SM, ET, and CW.

3.3.2. Causal Impact of 2015/16 El Niño on Wildfire Emissions

Generally, Figure 6 shows that after the El Niño onset (red vertical line), the observed values of the emission parameters (black points) rise above the counterfactual baseline (orange line). For certain parameters, such as BA, the difference between the modelled counterfactual (i.e., indicating what would have happened in the absence of an El Niño) and observations is marginal. While for others, such as CO, BC, OC, SO2, PM2.5, and AOD, the differences are above the model’s posterior predictive distribution of the counterfactual especially in JJA of 2015 and 2017. Indeed, wildfire’s primary impact is manifested by burned areas as vegetation constitutes fire fuel and its condition and characteristics influence the flammability, intensity, and spread of fires [41]. In certain vegetation types such as grasslands and shrubland, the fire may spread to large areas albeit at low intensity due to their low amount of fuel, i.e., 2–10 Mg/ha. Despite containing the largest amount of fuel (i.e., 10–1500 Mg/ha), forests and other woody vegetation types tend to burn marginally [42], resulting in incomplete combustion. In Africa, broadleaved deciduous trees especially in tropical regions are expansively burned as part of land clearance for agriculture [39,43]. In Figure 6a, El Niño had predominantly positive effect on BA, where the cumulative causal impact indicates that BA anomaly increased steadily but relatively marginal, reaching roughly 3 standardized km2 by the end of 2017. The increase in BA can be linked to similar post-2015 departures from the predicted counterfactual in CO (Figure 6b), BC (Figure 6c), and OC (Figure 6d), reflecting that all three pollutants are co-emitted by biomass burning. The instantaneous Bayesian causal impact in BC and OC show isolated spikes in 2015, 2016, and 2017, which can be associated with emission anomalies driven by drier conditions introduced by the 2015/16 El Niño. In contrast, CO shows consistent rise in emissions post-2015, attributed to El Niño onset. The high Bayesian R2 (≈0.95) and narrow uncertainty intervals (orange shaded region) indicate that these increases are highly unlikely to be due to chance and that it is statistically significant. The cumulative causal impact is positive, reaching ~7.5 standardized kg/m3 for CO, ~4 and ~7 standardized µg/m3 for BC and OC, respectively.
The other emission parameters affected by the anomalies associated with the 2015/16 El Niño include PM2.5 and AOD (Figure 6g,h). However, the highest departures from the counterfactual include JJA 2015 and 2017 as well as DJF 2016 for both emission parameters, resulting in significant spikes in Bayesian causal impact especially around early 2016 and late 2017. These outlier spikes contributed to the overall cumulative causal impact of the 2015/16 El Niño on these parameters, which indicates a steady rise after 2015 (reaching roughly 15 standardized µg/m3 and 15 [unitless] for PM2.5 and AOD, respectively). This indicates a sustained positive anomaly compared to the pre-2015 baseline. The rise in PM2.5 post-2015 is consistent with an increase in precursor gases such as CO, which are emitted by wildfires. This is also in agreement with a recent study by Gong et al. [44] who found a significant positive correlation between PM2.5 and CO concentrations. Overall, the ITS results imply that not only did the area burned increase under the El Niño, but fires were intensely emitted carbonaceous aerosols.

3.3.3. Causal Impact of 2015/16 El Niño on Vegetation and Meteorological Parameters

The results generally show a decreasing effect on the vegetation (i.e., NDVI, ET, and CW) and meteorological parameters (i.e., PCPN, and SM) except for ST, which increased post-intervention. Indeed, PCPN determines the availability of soil moisture and vegetation, and its decline will cause reciprocal decrease in parameters such as the SM and NDVI. Figure 7b shows that PCPN exhibits a pronounced post-2015 deficit relative to the counterfactual. The observed PCPN drops below predictions at the intervention date, and several noticeable low-precipitation anomalies occur on discrete dates thereafter; correspondingly the pointwise causal impact is negative and the cumulative impact shows overall declining trend. This sustained precipitation shortfall resulted in the corresponding NDVI decline (Figure 7a) and explains other hydrological responses observed in parameters such as SM (Figure 7d) and ET (Figure 7e). The NDVI (Figure 7a) departs from its counterfactual post-2015, showing observed values drop below the predicted trajectory on several occasions and the pointwise causal impact is mostly negative for much of the 2015/16 period but positive from 2017 onwards. The cumulative causal impact, therefore, accumulates as a sustained negative anomaly, consistent with reduced photosynthetic activity and canopy greenness during the El Niño event.
The departure in ET from the counterfactual at the beginning of 2015 was positive, and shortly declined sharply, picking up shortly again in 2017. The ET response likely reflects two competing processes. On the one hand, at the beginning of the 2015 crop growing season, higher ST (Figure 7c) may have caused the increase in ET, especially over cultivated areas and shrubs. On the other hand, as the season progressed, reduced SM and stomatal closure under plant stress may have caused ET to decline steadily. The cumulative impact is negative, indicating a net loss of evaporative fluxes over the post-intervention period. The magnitude and persistence of the SM anomaly (Figure 7d), indicate a clear drying of the soil column that is commensurate with the observed PCPN deficits and ST increase. Among all the parameters considered, CW displays a more ambiguous pattern (Figure 7f). While there are some post-2015 departures from the counterfactual, the causal impact and cumulative causal impact are smaller, mostly prominent in 2015. This suggests that CW experienced episodic perturbations, which may have been irrigation-related wetting events and wetland vegetation. However, statistical evidence for a large, persistent El Niño-driven change is weaker. Overall, these results for other parameters indicate a coherent representation of El Niño-associated drought stress and thermal forcing.

4. Discussion

The roles of meteorological conditions in governing soil moisture and vegetation growth are well-established as well as their impact on the wildfire emissions [45,46,47,48]. Studies [44,49] have shown that meteorological parameters, such as precipitation, temperature, relative humidity, wind, and planetary boundary layer modulate the emissions. Using the Partial Least Squares-Structural Equation Modelling (PLS-SEM), Gong et al. [44] found that meteorological conditions exert a direct influence on precursor emissions of PM2.5 such as CO, nitrogen dioxide (NO2), and formaldehyde (HCHO), with a loading/path strength of about 0.918 in the winter season, and overall they contribute to about 89.9% of the variability in precursor emissions. This means that weather conditions not only affect the chemistry and transport of pollutants but also drive changes in the amounts of precursors released into the atmosphere. For wildfire emissions, high precipitation, soil moisture, and relative humidity are often associated with low fire activity, while the deficits in these parameters cause vegetation desiccation, increased fuel load, and higher fire probability and eventually higher release of emissions should the fire occur. For example, the rapid spread and intense burning of biomass during the 2024 wildfires in Canada was attributed to unfavorable El Niño weather conditions, such as high temperatures, dryness, and strong winds [50]. According to Filonchyk et al. [50], these meteorological factors led to high fire radiative energy (FRE) and, consequently, significant emissions of pollutants CO2, CO, CH4, PM2.5, and NOx.
During the 2015/16 El Niño, Southern Africa experienced a pronounced dry–hot anomaly, which caused increased fire activity and related emissions [3,9]. Our results from the interrupted time-series (ITS) causal impact analysis corroborate this, showing that precipitation fell sharply below its pre-2015 (before El Niño) trend, producing a large negative anomaly and a significant cumulative shortfall. Simultaneously, the ITS model indicated that observed temperatures during the El Niño period were anomalously high relative to the counterfactual baseline. These causal findings are strongly supported by the time-lagged correlation analysis, which reveals the mechanistic coupling behind these anomalies. The negative lagged correlations between temperature and vegetation indices, alongside positive correlations with fire emissions, suggest that the thermal forcing identified by the ITS analysis directly stressed vegetation and increased atmospheric evaporative demand. This aligns with previous studies [8,51,52] showing that low rainfall and extreme heat drove a record-breaking seasonal drought, exacerbating moisture loss from soils and vegetation.
The dual stress caused by precipitation decline and high temperatures left soils significantly drier than expected under the counterfactual. The ITS results for soil moisture (SM) showed a strong negative departure from baseline, reflecting both reduced recharge and enhanced drying. Because soil moisture is the primary water reservoir for plants, its decline resulted in an immediate (i.e., within one month) ecological response. Indeed, previous studies [53] underscore that soil moisture regulates vegetation greenness in Southern Africa. In fact, the lagged correlation results indicate that soil moisture typically varies concurrently (i.e., lag = 0) with the normalized difference vegetation index (NDVI) and precipitation (PCPN) (r = 0.64 and r = 0.60, respectively). The synchronicity observed in the lagged correlation analysis explains the rapid divergence of NDVI from its counterfactual in the causal analysis, indicating that the vegetation (dominated by Southern African savanna ecosystems) possesses limited hydraulic buffering capacity, when a monthly-aggregated data is considered. The El Niño-linked drought followed an already dry 2014/15 season [8]; therefore, soils were pre-depleted, and the additional heat and lack of rain in 2015/16 drove soil moisture to record lows, creating ideal conditions for the amplified emissions observed in the causal impact models.
The ITS analysis indicated that while the burned area (BA) showed a steady but relatively marginal cumulative increase, the anomalies in fire-related emissions (CO, BC, OC, PM2.5) were statistically significant at the 95% confidence level. This suggests that the severity of air-quality degradation cannot be explained by the spatial expansion of fires alone. The lagged correlation analysis supports this by demonstrating strong and statistically significant correlation between combustion products (BC, OC) and aerosol loading (AOD, PM2.5), where r = 0.8 and 0.81 at lag = 0 for AOD-BC and AOD-OC correlations and 0.67 and 0.70 at lag 0 for PM2.5-BC and PM2.5-OC correlations. This confirms biomass burning as the dominant source in the post-El Niño period. The BA was strongly correlated with CO, BC, SO2, and OC at lag 0, ascertaining the linkages between these parameters. The findings imply that the El Niño conditions (such as the associated severe drought) likely altered the combustion efficiency of the fuel load. The extreme desiccation of vegetation, resulting from the moisture deficits identified in the ITS, facilitated more complete combustion or intense smoldering phases, thereby enhancing emission factors for particulates and carbon monoxide even where the total area burned did not expand exponentially.
Despite the promising results, the techniques employed in the current study were limited in several ways. First, the ITS assumes that the underlying outcome trajectory would have continued unchanged if the intervention (i.e., El Niño) had not occurred, which may not necessarily hold in practice. Second, it assumes that there are no other simultaneous changes affecting the outcome. For pollutants such as PM2.5 and AOD which may not be necessarily from fires alone, if any external shock such as a dust storm occurs post-El Niño, ITS cannot distinguish its effect from El Niño’s effect. However, in the absence of a control group, ITS is the most promising approach to study the effects of El Niño on wildfires and their related emissions. The counterfactual trajectories were derived exclusively from the pre-intervention time-series in order to characterize the underlying baseline dynamics of the various variables under “business-as-usual” conditions. While many ecological variables exhibit seasonality and excluding such seasonal effects may mistake annual cycles for El Niño impact, the ITS model captured the temporal structure, seasonality, and autocorrelation patterns over the pre-El Niño period of 5-years. Moreover, while the validity of the causal impact results may be questionable, the goodness-of-fit between the modelled and observed values during the pre-El Niño period—quantified using the coefficient of determination (R2)—provided an internal measure of how well the model reproduces the observed dynamics prior to the occurrence of the 2015/16 El Niño. Essentially, a high pre-intervention R2, achieved in this study, strengthens the credibility of the counterfactual projection. Considering the time-lagged correlations, some of the responses may have been masked by monthly-aggregated data, influencing lags of 0 in the results. For example, some vegetation types are sensitive to soil moisture deficits, and their drying out may have happened within weeks. Future studies should consider using weekly or daily data where possible.
Overall, the integration of time-lagged cross-correlation with interrupted time-series (ITS) causal impact analysis provided a robust, multi-dimensional perspective on the Earth system response to the 2015/16 El Niño. While the ITS analysis quantified the magnitude of the divergence from the “business-as-usual” climatology, the lagged correlation analysis elucidated the temporal mechanisms driving these anomalies. These approaches revealed that the observed environmental stress was not merely a function of unpredictable or random variability but a deterministic response to a compound El Niño event that fundamentally altered regional hydro-ecological and atmospheric dynamics.

5. Conclusions

Overall, the time-lagged correlation and causal impact analyses reveal a coherent regional response of Southern Africa to the 2015/16 El Niño, characterized by tightly coupled hydrological, ecological, and atmospheric processes. The lagged correlation results indicate that most vegetation–climate–air-quality interactions in the region are predominantly contemporaneous, while moisture-related variables—particularly precipitation and soil moisture—exert the most pronounced delayed influence on atmospheric composition and surface conditions. This highlights the pivotal role of hydroclimatic variability in Southern Africa in regulating vegetation greenness, evapotranspiration, and pollutant removal, alongside the strong synchronous coupling between carbonaceous aerosols, PM2.5, and AOD that reflects the dominance of biomass-burning emissions.
The causal impact analysis further shows that the 2015/16 El Niño produced statistically significant and sustained positive anomalies in fire-related emissions (CO, BC, OC, PM2.5, and AOD) across Southern Africa, with pronounced departures during austral winter and peak dry-season periods. These patterns suggest that El Niño conditions intensified combustion efficiency and aerosol emissions rather than merely expanding burned area. In contrast, vegetation and hydrometeorological variables display persistent negative deviations from their counterfactual trajectories, indicating widespread drought stress, declining soil moisture, reduced canopy greenness, and suppressed evaporative fluxes under elevated surface temperatures. Overall, the results demonstrate that El Niño acts as a major climatic perturbation in Southern Africa, simultaneously degrading ecosystem functioning and amplifying atmospheric pollution through climate–fire–vegetation feedback. The integrated use of time-lagged correlation and Bayesian causal inference thus provides robust evidence that large-scale climate variability can rapidly propagate through Southern African Earth system components, yielding immediate air-quality impacts and longer-lasting ecological and hydrological consequences.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/earth7020042/s1, Figure S1. NDVI temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05; Figure S2. PCPN temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05; Figure S3. Temperature temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05; Figure S4. SM temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05; Figure S5. ET temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05; Figure S6. CW temporal associations with (a) BA, (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red dashed line indicates zero lag, and the red dots indicate the significant r coefficients and lags, at α = 0.05.

Author Contributions

Conceptualization, M.K. and L.S.; methodology, M.K.; formal analysis, M.K. and L.S.; investigation, M.K., L.S., and X.Z.; writing—original draft preparation, M.K., L.S., and X.Z.; writing—review and editing, M.K., L.S., and X.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This work is based on the research supported in part by the National Research Foundation of South Africa (Ref Number: BRIC231103160523). APC was funded by the South African National Space Agency (SANSA). Xiang Zhang was supported by Natural Science Foundation of China (42461144214).

Data Availability Statement

The data used in this study are freely available from NASA’s Giovanni version 4.40 (https://giovanni.gsfc.nasa.gov/giovanni/ (accessed on 30 September 2025)), while the MODIS Burned Area data can be obtained from Google Earth Engine (GEE, https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MCD64A1 (accessed on 30 September 2025)).

Acknowledgments

The author acknowledges the GES-DISC Interactive Online Visualization and Analysis Infrastructure (Giovanni) for providing the data used in this study. We acknowledge the use of data products from the MODIS Burned Area product (MCD64A1), courtesy of the NASA EOSDIS Land Processes Distributed Active Archive Center (LP DAAC), USGS/Earth Resources Observation and Science (EROS) Center, Sioux Falls, South Dakota. Xiang Zhang was supported by Natural Science Foundation of China (42461144214).

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Li, C.; Wang, X.; Liu, Z.; Cheng, L.; Huang, C.; Wang, J. El Niño southern oscillation, weather patterns, and bacillary dysentery in the Yangtze River Basin, China. Glob. Health Res. Policy 2024, 9, 45. [Google Scholar] [CrossRef]
  2. Chikoore, H.; Jury, M.R. South African drought, deconstructed. Weather. Clim. Extrem. 2021, 33, 100334. [Google Scholar] [CrossRef]
  3. Shikwambana, L.; Kganyago, M. Seasonal Comparison of the Wildfire Emissions in Southern African Region during the Strong ENSO Events of 2010/11 and 2015/16 Using Trend Analysis and Anomaly Detection. Remote Sens. 2023, 15, 1073. [Google Scholar] [CrossRef]
  4. Petry, I.; Fan, F.M.; Wood, A.W. Observed streamflow data shows El Niño–Southern Oscillation increases likelihood of extreme events in South America. Commun. Earth Environ. 2025, 6, 699. [Google Scholar] [CrossRef]
  5. Sazib, N.; Mladenova, I.E.; Bolten, J.D. Assessing the Impact of ENSO on Agriculture Over Africa Using Earth Observation Data. Front. Sustain. Front. Sustain. Food Syst. 2020, 4, 509914. [Google Scholar] [CrossRef]
  6. Mugiyo, H.; Magadzire, T.; Choruma, D.J.; Chimonyo, V.G.P.; Manzou, R.; Jiri, O.; Mabhaudhi, T. El Niño’s Effects on Southern African Agriculture in 2023/24 and Anticipatory Action Strategies to Reduce the Impacts in Zimbabwe. Atmosphere 2023, 14, 1692. [Google Scholar] [CrossRef]
  7. Siderius, C.; Gannon, K.E.; Ndiyoi, M.; Opere, A.; Batisani, N.; Olago, D.; Pardoe, J.; Conway, D. Hydrological Response and Complex Impact Pathways of the 2015/2016 El Niño in Eastern and Southern Africa. Earth’s Future 2018, 6, 2–22. [Google Scholar] [CrossRef]
  8. Kolusu, S.R.; Shamsudduha, M.; Todd, M.C.; Taylor, R.G.; Seddon, D.; Kashaigili, J.J.; Ebrahim, G.Y.; Cuthbert, M.O.; Sorensen, J.P.R.; Villholth, K.G.; et al. The El Niño event of 2015–2016: Climate anomalies and their impact on groundwater resources in East and Southern Africa. Hydrol. Earth Syst. Sci 2019, 23, 1751–1762. [Google Scholar] [CrossRef]
  9. Shikwambana, L.; Kganyago, M.; Xulu, S. Analysis of wildfires and associated emissions during the recent strong ENSO phases in Southern Africa using multi-source remotely-derived products. Geocarto Int. 2022, 37, 16654–16670. [Google Scholar] [CrossRef]
  10. Jaffe, D.A.; O’Neill, S.M.; Larkin, N.K.; Holder, A.L.; Peterson, D.L.; Halofsky, J.E.; Rappold, A.G. Wildfire and prescribed burning impacts on air quality in the United States. J. Air Waste Manag. Assoc. 2020, 70, 583–615. [Google Scholar] [CrossRef]
  11. Sintayehu, D.W. Impact of climate change on biodiversity and associated key ecosystem services in Africa: A systematic review. Ecosyst. Health Sustain. 2018, 4, 225–239. [Google Scholar] [CrossRef]
  12. Bolaño-Diaz, S.; Camargo-Caicedo, Y.; Tovar Bernal, F.; Bolaño-Ortiz, T.R. The Effect of Forest Fire Events on Air Quality: A Case Study of Northern Colombia. Fire 2022, 5, 191. [Google Scholar] [CrossRef]
  13. Zhang, Q.; Wang, Y.; Xiao, Q.; Geng, G.; Davis, S.J.; Liu, X.; Yang, J.; Liu, J.; Huang, W.; He, C.; et al. Long-range PM2.5 pollution and health impacts from the 2023 Canadian wildfires. Nature 2025, 645, 672–678. [Google Scholar] [CrossRef]
  14. van Wagtendonk, J.W. The History and Evolution of Wildland Fire Use. Fire Ecol. 2007, 3, 3–17. [Google Scholar] [CrossRef]
  15. de Souza Fernandes Duarte, E.; Salgueiro, V.; Costa, M.J.; Lucio, P.S.; Potes, M.; Bortoli, D.; Salgado, R. Fire-pollutant-atmosphere components and its impact on mortality in Portugal during wildfire seasons. GeoHealth 2023, 7, e2023GH000802. [Google Scholar] [CrossRef]
  16. Voulgarakis, A.; Field, R.D. Fire Influences on Atmospheric Composition, Air Quality and Climate. Curr. Pollution Rep. 2025, 1, 70–81. [Google Scholar] [CrossRef]
  17. Marais, E.A.; Van Damme, M.; Clarisse, L.; Wiedinmyer, C.; Murphy, K.; van der Werf, G.R. Subtropical southern Africa fire emissions of nitrogen oxides and ammonia obtained with satellite observations and GEOS-Chem. Environ. Sci. Atmos. 2025, 5, 906–920. [Google Scholar] [CrossRef]
  18. Shetty, S.S.; Deepthi, D.; Harshitha, S.; Sonkusare, S.; Naik, P.B.; Suchetha Kumari, N.; Madhyastha, H. Environmental pollutants and their effects on human health. Heliyon 2023, 9, e19496. [Google Scholar] [CrossRef]
  19. Liu, Y.; Goodrick, S.; Heilman, W. Wildland fire emissions, carbon, and climate: Wildfire–climate interactions. For. Ecol. Manag. 2014, 317, 80–96. [Google Scholar] [CrossRef]
  20. Zhang, A.; Wang, Y.; Zou, Y. Positive feedback to regional climate enhances African wildfires. iScience 2023, 26, 108533. [Google Scholar] [CrossRef]
  21. Beerling, D.J.; Fox, A.; Stevenson, D.S.; Valdes, P.J. Enhanced chemistry-climate feedbacks in past greenhouse worlds. Proc. Natl. Acad. Sci. USA 2011, 108, 9770–9775. [Google Scholar] [CrossRef]
  22. von Schneidemesser, E.; Monks, P.S.; Allan, J.D.; Bruhwiler, L.; Forster, P.; Fowler, D.; Lauer, A.; Morgan, W.T.; Paasonen, P.; Righi, M.; et al. Chemistry and the Linkages between Air Quality and Climate Change. Chem. Rev. 2015, 115, 3856–3897. [Google Scholar] [CrossRef]
  23. Kouassi, J.-L.K.; Wandan, N.E.; Mbow, C. Assessing the Impact of Climate Variability on Wildfires in the N’Zi River Watershed in Central Côte d’Ivoire. Fire 2018, 1, 36. [Google Scholar] [CrossRef]
  24. Wasserman, T.N.; Mueller, S.E. Climate influences on future fire severity: A synthesis of climate-fire interactions and impacts on fire regimes, high-severity fire, and forests in the western United States. Fire Ecol. 2023, 19, 43. [Google Scholar] [CrossRef]
  25. Chantelle, B.; Betts, R.A.; Jones, C.D.; Feldpausch, T.R.; Cardoso, M.; Anderson, L. Niño Driven Changes in Global Fire 2015/16. Front. Earth Sci 2020, 8, 199. [Google Scholar] [CrossRef]
  26. Feng, S.; Zhang, Z.; Zhao, S.; Guo, X.; Zhu, W.; Das, P. Time lag effect of vegetation response to seasonal precipitation in the Mara River Basin. Ecol. Process 2023, 12, 49. [Google Scholar] [CrossRef]
  27. Nogueira, A.R.; Pugnana, A.; Ruggieri, S.; Pedreschi, D.; Gama, J. Methods and tools for causal discovery and causal inference. Wiley Interdiscip. Rev.-Data Min. Knowl. Discov. 2022, 12, e1449. [Google Scholar] [CrossRef]
  28. Buchard, V.; Randles, C.A.; da Silva, A.M.; Darmenov, A.; Colarco, P.R.; Govindaraju, R.; Ferrare, R.; Hair, J.; Beyersdorf, A.J.; Ziemba, L.D.; et al. The MERRA-2 aerosol reanalysis, 1980 onward. Part II: Evaluation and case studies. J. Clim. 2017, 30, 6851–6872. [Google Scholar] [CrossRef]
  29. Randles, C.A.; da Silva, A.M.; Buchard, V.; Colarco, P.R.; Darmenov, A.; Govindaraju, R.; Smirnov, A.; Holben, B.; Ferrare, R.; Hair, J.; et al. The MERRA-2 aerosol reanalysis, 1980 onward. Part I: System description and data assimilation evaluation. J. Clim. 2017, 30, 6823–6850. [Google Scholar] [CrossRef]
  30. Tukey, J.W. Exploratory Data Analysis; Addison-Wesley: Boston, MA, USA, 1977. [Google Scholar]
  31. Chambers, J.M.; Cleveland, W.S.; Kleiner, B.; Tukey, P.A. Graphical Methods for Data Analysis; Chapman and Hall/CRC: New York, NY, USA, 1983. [Google Scholar]
  32. Han, J.; Kamber, M.; Pei, J. Data Mining: Concepts and Techniques; Morgan Kaufmann: Francisco, CA, USA, 2012. [Google Scholar]
  33. Wickham, H.; Grolemund, G. R for Data Science: Import, Tidy, Transform, Visualize, and Model Data; O’Reilly Media: Sebastopol, CA, USA, 2017. [Google Scholar]
  34. Box, G.E.P.; Jenkins, G.M.; Reinsel, G.C.; Ljung, G.M. Time Series Analysis: Forecasting and Control, 5th ed.; Wiley: Hoboken, NJ, USA, 2015. [Google Scholar]
  35. Chatfield, C. The Analysis of Time Series: An Introduction, 6th ed.; Chapman & Hall/CRC: New York, NY, USA, 2004. [Google Scholar]
  36. Brodersen, K.H.; Gallusser, F.; Koehler, J.; Remy, N.; Scott, S.L. Inferring causal impact using Bayesian structural time-series models. Ann. Appl. Stat. 2015, 9, 247–274. [Google Scholar] [CrossRef]
  37. Scott, S.L.; Varian, H.R. Predicting the present with Bayesian structural time series. Int. J. Math. Model. Numer. Optim. 2014, 5, 4–23. [Google Scholar] [CrossRef]
  38. Cohen, J. Statistical Power Analysis for the Behavioral Sciences, 2nd ed.; Routledge: New York, NY, USA, 1988. [Google Scholar]
  39. Kganyago, M.; Shikwambana, L. Assessing Spatio-Temporal Variability of Wildfires and their Impact on Sub-Saharan Ecosystems and Air Quality Using Multisource Remotely Sensed Data and Trend Analysis. Sustainability 2019, 11, 6811. [Google Scholar] [CrossRef]
  40. He, Q.; Wang, M.; Yim, S.H.L. The spatiotemporal relationship between PM2.5 and aerosol optical depth in China: Influencing factors and implications for satellite PM2.5 estimations using MAIAC aerosol optical depth. Atmos. Chem. Phys. 2012, 21, 18375–18391. [Google Scholar] [CrossRef]
  41. Koutsias, N.; Xanthopoulos, G.; Founda, D.; Xystrakis, F.; Nioti, F.; Pleniou, M.; Mallinis, G.; Arianoutsou, M. On the relationships between forest fires and weather conditions in Greece from long-term national observations (1894–2010). Int. J. Wildland Fire 2013, 22, 493–507. [Google Scholar] [CrossRef]
  42. Stavi, I. Wildfires in Grasslands and Shrublands: A Review of Impacts on Vegetation, Soil, Hydrology, and Geomorphology. Water 2019, 11, 1042. [Google Scholar] [CrossRef]
  43. Varma, A. The economics of slash and burn: A case study of the 1997-1998 Indonesian forest fires. Ecol. Econ. 2003, 46, 159–171. [Google Scholar] [CrossRef]
  44. Gong, D.; Du, N.; Wang, L.; Deng, X.; Zhang, X.; Yang, L. Impacts of meteorological and precursor emission factors on PM2.5 and O3 from 2019 to 2022: Insights from multiple perspectives. Atmos Res 2025, 315, 107933. [Google Scholar] [CrossRef]
  45. Aguilera, R.; Luo, N.; Basu, R.; Wu, J.; Clemesha, R.; Gershunov, A.; Benmarhnia, T. A novel ensemble-based statistical approach to estimate daily wildfire-specific PM2.5 in California (2006–2020). Environ. Int. 2013, 171, 107719. [Google Scholar] [CrossRef]
  46. Fang, K.; Yao, Q.; Guo, Z.; Zheng, B.; Du, J.; Qi, F.; Yan, P.; Li, J.; Ou, T.; Liu, J.; et al. ENSO modulates wildfire activity in China. Nat. Commun. 2021, 12, 1764. [Google Scholar] [CrossRef]
  47. Jin, H. Drivers of Global Wildfires—Statistical Analyses. Master’s Thesis, Lund University, Lund, Sweden, 2010. [Google Scholar]
  48. Rihan, W.; Zhao, J.; Zhang, H.; Guo, X.; Ying, H.; Deng, G.; Li, H. Wildfires on the Mongolian Plateau: Identifying Drivers and Spatial Distributions to Predict Wildfire Probability. Remote Sens. 2019, 11, 2361. [Google Scholar] [CrossRef]
  49. Ding, J.; Dai, Q.; Fan, W.; Lu, M.; Zhang, Y.; Han, S.; Feng, Y. Impacts of meteorology and precursor emission change on O3 variation in Tianjin, China from 2015 to 2021. J. Environ. Sci 2023, 126, 506–516. [Google Scholar] [CrossRef]
  50. Filonchyk, M.; Peterson, M.P.; Zhang, L.; Zhang, L.; He, Y. Estimating air pollutant emissions from the 2024 wildfires in Canada and the impact on air quality. Gondwana Res. 2025, 140, 194–204. [Google Scholar] [CrossRef]
  51. Blamey, R.C.; Kolusu, S.R.; Mahlalela, P.; Todd, M.C.; Reason, C.J.C. The role of regional circulation features in regulating El Niño climate impacts over southern Africa: A comparison of the 2015/2016 drought with previous events. Int J Climatol 2018, 38, 4276–4295. [Google Scholar] [CrossRef]
  52. Russo, S.; Marchese, A.F.; Sillmann, J.; Immé, G. When will unusual heat waves become normal in a warming Africa? Environ. Res. Lett. 2016, 11, 054016. [Google Scholar] [CrossRef]
  53. Ugbaje, S.U.; Bishop, T.F.A. Hydrological Control of Vegetation Greenness Dynamics in Africa: A Multivariate Analysis Using Satellite Observed Soil Moisture, Terrestrial Water Storage and Precipitation. Land 2020, 9, 15. [Google Scholar] [CrossRef]
Earth 07 00042 g001
Figure 1. (a) Extent of the study area and (b) map showing the fires the El Niño event before and after the fires.
Figure 1. (a) Extent of the study area and (b) map showing the fires the El Niño event before and after the fires.
Earth 07 00042 g001
Earth 07 00042 g002
Figure 2. Boxplots of various emission, meteorological, and environmental parameters pre- and post-2015/16 El Niño event.
Figure 2. Boxplots of various emission, meteorological, and environmental parameters pre- and post-2015/16 El Niño event.
Earth 07 00042 g002
Earth 07 00042 g003
Figure 3. Cohen’s d standardized effect sizes pre and post the 2015/16 El Niño. The up (▲ increase) and down (▼ decrease) arrows show the direction of significant changes, while small (i.e., d = 0.2), medium (i.e., d = 0.5), and large (i.e., d = 0.8) annotations indicate the standard effect sizes.
Figure 3. Cohen’s d standardized effect sizes pre and post the 2015/16 El Niño. The up (▲ increase) and down (▼ decrease) arrows show the direction of significant changes, while small (i.e., d = 0.2), medium (i.e., d = 0.5), and large (i.e., d = 0.8) annotations indicate the standard effect sizes.
Earth 07 00042 g003
Earth 07 00042 g004
Figure 4. Variability (standard deviation) changes of different parameters after 2015, as compared to the period before 2015.
Figure 4. Variability (standard deviation) changes of different parameters after 2015, as compared to the period before 2015.
Earth 07 00042 g004
Earth 07 00042 g005
Figure 5. Pearson’s r correlation coefficient and lag (months), based on seasonally adjusted time-series (2010–2017), where the strongest significant relationship occurs. ns denotes no significant relationship found across the 6 lags at α < 0.05; hence, the zero-lag correlation is shown instead.
Figure 5. Pearson’s r correlation coefficient and lag (months), based on seasonally adjusted time-series (2010–2017), where the strongest significant relationship occurs. ns denotes no significant relationship found across the 6 lags at α < 0.05; hence, the zero-lag correlation is shown instead.
Earth 07 00042 g005
Earth 07 00042 g006
Figure 6. Impact of the 2015/16 El Niño on (a) BA, and emission parameters: (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red solid line denotes the start of El Niño in 2015. The blue and orange shades indicate the 95% confidence intervals pre- and post-El Niño periods.
Figure 6. Impact of the 2015/16 El Niño on (a) BA, and emission parameters: (b) CO, (c) BC, (d) OC, (e) SO2, (f) SO4, (g) PM2.5, and (h) AOD. The red solid line denotes the start of El Niño in 2015. The blue and orange shades indicate the 95% confidence intervals pre- and post-El Niño periods.
Earth 07 00042 g006
Earth 07 00042 g007
Figure 7. Impact of the 2015/16 El Niño on (a) NDVI, (b) PCPN, (c) TEMP, (d) SM, (e) ET, and (f) CW. The red solid line denotes the start of the El Niño in 2015. The blue and orange shades indicate the 95% confidence intervals pre- and post-El Niño periods.
Figure 7. Impact of the 2015/16 El Niño on (a) NDVI, (b) PCPN, (c) TEMP, (d) SM, (e) ET, and (f) CW. The red solid line denotes the start of the El Niño in 2015. The blue and orange shades indicate the 95% confidence intervals pre- and post-El Niño periods.
Earth 07 00042 g007
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Shikwambana, L.; Kganyago, M.; Zhang, X. Analyzing the Effect of the 2015/16 Catastrophic El Niño Event on Wildfire Emissions in Southern Africa Using Lagged Correlation and Interrupted Time-Series Causal Impact Technique. Earth 2026, 7, 42. https://doi.org/10.3390/earth7020042

AMA Style

Shikwambana L, Kganyago M, Zhang X. Analyzing the Effect of the 2015/16 Catastrophic El Niño Event on Wildfire Emissions in Southern Africa Using Lagged Correlation and Interrupted Time-Series Causal Impact Technique. Earth. 2026; 7(2):42. https://doi.org/10.3390/earth7020042

Chicago/Turabian Style

Shikwambana, Lerato, Mahlatse Kganyago, and Xiang Zhang. 2026. "Analyzing the Effect of the 2015/16 Catastrophic El Niño Event on Wildfire Emissions in Southern Africa Using Lagged Correlation and Interrupted Time-Series Causal Impact Technique" Earth 7, no. 2: 42. https://doi.org/10.3390/earth7020042

APA Style

Shikwambana, L., Kganyago, M., & Zhang, X. (2026). Analyzing the Effect of the 2015/16 Catastrophic El Niño Event on Wildfire Emissions in Southern Africa Using Lagged Correlation and Interrupted Time-Series Causal Impact Technique. Earth, 7(2), 42. https://doi.org/10.3390/earth7020042

Article Metrics

Back to TopTop