Response of remote sensing-based climatic indicators to drought conditions during El Niño events and implications for water–food–energy nexus

ENSO

El Niño Southern Oscillation

NDVI

Normalised Difference Vegetation Index

ONI

Oceanic Niño Index

SPI

Standardised Precipitation Index

SVI

Standardised Vegetation Index

VCI

Vegetation Condition Index

WEF

Water–energy–food nexus

Droughts are a recurring, complex, creeping hydrometeorological phenomenon, which unpredictably occurs as a result of water availability deficiency from rainfall amounts below the usual average (Dai 2011; Van Loon et al. 2016; Mikaili & Rahimzadegan 2022). During droughts, the manner in which dry conditions propagate deficits in soil moisture, runoff, and recharge is complex and heterogeneous across geographical domains (Raposo et al. 2023). Future climate change predictions suggest that many areas will start to experience more frequent and intense dry conditions with irreversible impacts for people and ecosystems (IPCC 2014). Many countries, particularly those whose economies rely significantly on rain-fed agriculture, are vulnerable to the effects of climate variability and change. This is the situation in most African countries (Niang et al. 2014). Unfortunately, the majority of these countries are extremely vulnerable to climate change and have limited adaptive capacity to cope with the impacts of climate change (Dibi-Anoh et al. 2023; Taye & Dyer 2024).

Droughts are broadly categorised into four groups: meteorological, agricultural, and hydrological. Meteorological droughts are quite common, and they are primarily classified by the extent of dryness in a given location and the length of the dry period. Although agricultural droughts are linked to a lack of water needed to support crops, the drought does not always coincide with meteorological drought. On the other hand, hydrological drought is limited to the level of streamflow that can meet the demand. A study by Wilhite & Glantz (1985) gives a detailed description of this specific drought phenomenon. Although droughts can end by sudden extreme precipitation, the precise termination point is contested, making the phenomenon difficult to quantify and analyse (such as West et al. 2019; Ayugi et al. 2022).

For detection and determination of drought, different methods are proposed, of which some methods are based on using information of meteorological stations. Based on this information, some meteorological drought indices are calculated such as Palmer Drought Severity Index (PDSI), Surface Water Supply Index (SWSI), Percent of Normal (PN) Index, and Standardised Precipitation Index (SPI). The last one is used more frequently (Bonaccorso et al. 2003; Zhang et al. 2012; Bhunia et al. 2020). The SPI (McKee et al. 1993) was developed for the purpose of assigning a single value to precipitation that can be compared across regions with different climates. The limited number of hydrometeorological stations in developing countries leads to data gaps restricting usage of ground-based data in drought risk decision support in developing countries. Another category of drought monitoring methods are those based on the calculation of vegetation indices by satellite imagery (Heim 2002; Bhuiyan et al. 2006; Ezzine et al. 2014; West et al. 2019; Fatras et al. 2021; Juntakut et al. 2021; Mikaili & Rahimzadegan 2022). Some other remote sensing drought assessment methods used soil moisture data (Souza et al. 2018; Fatras et al. 2021; Souza et al. 2021). Furthermore, using remote sensing datasets, evapotranspiration models or water balance models (Khalili et al. 2011; Zhang et al. 2019; Tam et al. 2023) have been used to assess droughts.

Numerous studies have been conducted on droughts using satellite-derived indices under various climatic and environmental conditions. Ji & Peters (2003) derived a Normalised Difference Vegetation Index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR) and SPI to monitor moisture-related vegetation conditions. They found that the 3-month SPI had the best correlation with the NDVI, indicating the effects of lag time and cumulative precipitation on vegetation. The lag time was likely due to variations in the vegetation type and soil properties. These findings corroborate those of Wilhite et al. (2000) and Morid et al. (2006), who showed that the SPI is more reliable at detecting emerging droughts than other meteorological indices, and is thus a useful tool for initiating mitigation and response actions. Ji & Peters (2005) investigated the relationship between NDVI and precipitation to evaluate the vegetation-climate interactions on a large scale. They found that the time lag was shorter in the early growing season but longer in the mid- to late-growing season. Jain et al. (2010) used the SPI, NDVI, Water Supply Vegetation Index (WSVI), and Vegetation Condition Index (VCI) for drought monitoring in the three districts of Bhilwara, Kota, and Udaipur in India using AVHRR satellite data. They showed that NDVI and WSVI were highly correlated with SPI at different timescales (1–9 months) across districts. Gebrehiwot et al. (2011) used SPI and VCI for the spatial and temporal assessment of meteorological and agricultural drought indices. The results of the analysis revealed the occurrence of drought in southern and eastern Ethiopia. Furthermore, a lag time was observed between the VCI peak and rain gauge stations; the monthly VCI and precipitation data 2 months later had a good correlation.

Thavorntam et al. (2015) used the SPI and VCI to assess and evaluate drought indices in vegetated areas. They used precipitation data from 1980 to 2009 and MODIS-Terra vegetation (MOD13Q1) datasets from 2001 to 2009. SPI analysis revealed drought events at 3 and 6 months in the central and northeastern parts of Thailand. They showed that the relationship between the VCI and SPI can be used to forecast drought in the subsequent few months. Ghazaryan et al. (2020) evaluated the effects of drought on remotely sensed parameters at various stages of crop growth. They calculated several indices, including the NDVI, Normalised Difference Moisture Index (NDMI), Land Surface Temperature (LST), and Tasseled Cap, using Sentinel-1-based backscattering intensity. They revealed that all the remotely sensed variables respond differently to drought conditions. Recently, Mikaili & Rahimzadegan (2022) used satellite vegetation indices, including the Modified Perpendicular Drought Index (MPDI), VCI, Normalised Difference Vegetation Index Anomalies (NDVIA), and Standardized Vegetation Index (SVI), to monitor agricultural drought at a local scale in semi-arid Iran. The VCI exhibited the highest capability for investigating agricultural droughts in different climatic regions. Furthermore, the vegetation indices had the highest correlation with 3-month lag time on the SPI timescale. Arab & Ahamed (2023) demonstrated the use of the SVI and SPI for near real-time drought monitoring and assessment.

Previous studies have shown that although numerous satellite indices have been introduced to assess drought parameters, SPI- and NDVI-related indices are predominantly used. Further, the performance of different satellite indices and the lag time between meteorological drought as well as its effect on vegetation cover (agricultural drought) varies across different climatic and environmental conditions. Although the connection between droughts and climate variability and change has been examined in previous studies (e.g. Zhai et al. 2010; Zhang et al. 2012; Asadi-Zarch et al. 2015), the spatial-temporal and areal extent characteristics of rainfall anomalies during periods categorised as drier than normal (induced by the El Niño Southern Oscillation (ENSO) Phenomena) have not been thoroughly analysed (Hernández & Heslar 2019; Bhunia et al. 2020; Nikraftar et al. 2021; Dibi-Anoh et al. 2023).

The main contribution of this study was the use of common satellite indices (SPI and NDVI-based SVI) to evaluate the effects of El Niño on drought (onset, development, and termination) at a local scale. This assessment was conducted in Southern Province, Zambia, a semi-arid region which is highly vulnerable to drought. Adaptive actions under droughts are also proposed. The main objective of the study was therefore to investigate the response of remote sensing-based climatic indicators towards extreme drought conditions driven by El Niño. Specifically, it investigates (i) the trends of precipitation and vegetation using the satellite indices, (ii) evaluates how satellite indices respond spatio-temporally during extreme conditions, and (iii) explores actions to ensure water and food security during droughts.

The primary data source was the MOD13Q1.006 Terra Vegetation Indices 16-Day Global 250 m product from the Moderate-Resolution Imaging Spectroradiometer (MODIS). The data were based on NDVI and ranged from the 18th of February 2000 to the 23rd of April 2021. We utilised the NDVI data for trend analysis and to determine the SVI. The data were accessed using the Google Earth Engine (GEE) Platform. Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) data (1 January 1981 to 31 March 2021) were accessed from GEE. The data had a daily temporal resolution and a 0.05° spatial resolution. The CHIRPS data incorporate in-situ meteorological station data and satellite imagery. The data were used to calculate the SPI and for trend analysis.

All the data processing and analyses were performed using the GEE Platform. The SPI and SVI were calculated using MOD13Q1.006 (NDVI) and CHIRPS, respectively, using the UN-SPIDER recommended GEE practices for SVI and SPI, following the recommended GEE practices of the United Nations Platform for Space-based Information for Disaster Management and Emergency Response. Trend analysis was performed for NDVI, precipitation, SPI, and SVI. Furthermore, statistical analysis was conducted using Minitab Version 18 software. All datasets used are available via the Earth Engine Data Catalogue (https://developers.google.com/earth-engine/datasets/catalog: Accessed 8th August 2022).

In this study, the predefined time steps chosen were the SPI at 1 month and a 16-day SPI to match the MODIS vegetation index temporal resolution. However, because a lag exists between rainfall conditions and vegetation response (Thenkabail et al. 2004), the 16-day moving window was set to begin 7 days before the vegetation index collection dates. Therefore, the SPI began calculating 7 days before the MODIS start dates and ended 7 days earlier than the MODIS end dates. As cumulative rainfall for a period of less than 1 year tends to not be normally distributed, the rainfall data were normalised using the gamma probability density function. A fitted function was used to calculate the cumulative distribution of the data points. Finally, the data points were transformed into standardised normal variates using Equation (1). The SPI was classified into different drought categories of drought (Table 1) as prescribed by Stricevic et al. (2011).

According to previous studies (Wu et al. 2007; Mishra & Nagarajan 2011), the first limitation of the SPI is the length of the precipitation record, which has a significant impact on the SPI values. Therefore, the SPI can produce different results with different lengths of precipitation records. The second limitation is the probability distribution. Different probability distributions have been used in SPI calculations, namely gamma and Pearson Type III distribution, lognormal, extreme value, and exponential distributions, and have been widely applied to simulations of precipitation distributions (Wu et al. 2007). Another disadvantage is that when applying SPI at short timescales (1, 2, or 3 months) to regions with low seasonal precipitation, misleadingly large positive or negative SPI values may result.

The SVI is calculated based on probability estimate to predict and visualise the relative vegetation greenness with reference to the greenness probability for each pixel, which infers a temporal comparison (Peters et al. 2002). As the index is NDVI-based, it is related to the red and near-infrared reflectance wavebands sensed and recorded by several Earth observation satellite platforms. Therefore, data to derive the SVI are readily available and can be accessed in near real time, making it suitable for monitoring temporal phenomena such as drought. As SVI is a temporal comparison of NDVI, a maximum-value composite approach based on time-series NDVI was used. Before calculating the SVI, data preparation and preprocessing were conducted, which involved the selection of a timeframe, cloud masking, resampling and rescaling of data, and clipping of the imagery to the boundaries of Southern Province. The z-values and SVI of each pixel were calculated using Equations (2) and (3), respectively.

The Mann–Kendall trend test and Sen's slope method were used to perform the trend analysis. The Mann–Kendall trend test is used to determine whether a time-series dataset has a monotonic upward or downward trend. It is a nonparametric test; therefore, it does not require the data to be normally distributed or linear. Meanwhile, Sen's slope is a nonparametric test based on Kendall's rank correlation, tau. It is a point estimator for the median of a set of slopes (Y_j – Y_i)/(t_j – t_i) joining pairs of points t_j ≠ t_i (Sen 1968). The Mann–Kendall trend test only provides an indication of an existing trend (upward or downward), whereas Sen's slope provides the magnitude.

Trends in rainfall, meteorological drought (SPI), and relative vegetation conditions were checked for each season. Despite conducting an analysis of the entire rainfall dataset (1981 to 2021), trend analysis was performed only from 2000 to 2021 to match with the timeframe of the MODIS-Terra vegetation indices. Furthermore, rainfall and SPI were aggregated into 16-day values (with start dates 7 days prior to accommodate the lag of the vegetation response) to match the NDVI and SVI dates. Trend analysis of rainfall was performed on a monthly basis; only the 16-day rainfall data of a particular month were considered in the trend analysis to compensate for seasonality. Particular attention was paid to drought seasons with medium to strong El Niño patterns to check whether there were any differences between such droughts and those with an absent or weak El Niño pattern.

The higher magnitudes of certain trends (and hence, change) can be attributed mainly to changes in land cover and land use and only partially to climate change. This was evidenced by the larger ranges of slope medians in the vegetation indices than in the SPI. Changes in land cover and land use occur at a faster rate than changes in climate. Although climate change is a slow and steady process, land cover and land use changes can occur abruptly as forests, grasslands, wetlands, and other vegetated land cover types are converted into agricultural fields, urban areas, and other anthropogenically dominated landscapes. Notably, Southern Province is one of the main hotspots of deforestation in Zambia due to agricultural expansion, infrastructure development, wood extraction, and fires (Vinya et al. 2011). Therefore, it has been subject to land use change during the study period.

Although the Mann–Kendall trend test and Sen's slope method yielded similar results for the NDVI and SVI, there was still a large difference in the ranges of the slope medians. The smaller range of the SVI may have been attributable to it being an indicator of the deviations of the vegetation condition at a specific time from the mean vegetation condition in a specific pixel, derived from the z-score. The mean vegetation condition of each pixel is influenced by all vegetation conditions over time, and the mean vegetation conditions affect both the past and present values when calculating the SVI. Furthermore, seasonality is absent in the SVI. However, the NDVI value indicates the vegetation condition at a discrete point in time (not relative), and seasonality is present, resulting in larger changes. Additionally, NDVI values are related in one direction of time (when performing trend tests), that is, from the past to the present.

Droughts in southern Zambia can be classified into two categories: regressive droughts and aggressive droughts. We determined that regressive droughts coincided with moderate to strong El Niño conditions and were characterised by the decrease in drought intensity as a rainfall season progressed. Aggressive droughts were characterised by an increase in drought intensity as the season wore in and the absence of moderate to strong El Niño conditions. Recessive droughts tend to dissipate by February and hence rainfall would be available thereafter. We postulate that this is likely the case for the Southern Africa. Therefore, even though El Niño events greatly undermine water security, it means delayed planting to ensure maximum crop productive during strong El Niño events. Soil moisture also tends to be adequate to support the vegetation cover. Further investigations are required to evaluate variability in both evaporation and soil moisture over this region as demonstrated by studies such as Gushchina et al. (2020). However, during the weak to moderate El Niño events, the droughts seem to be persistent. Investigations are therefore required to determine suitable sites for other technologies such as managed aquifer recharge (MAR). MAR would provide a better climate buffer in the region and assure water security under water stress. Drought-prone areas such as Southern Zambia have a number of dams, however, storage capacity of these dams require further investigation due to the burden of sediments (Winton et al. 2021). Furthermore, this region tends to have high evaporation rates (Hamududu & Ngoma 2020). Rainwater harvesting would be an option for the urbanised parts of the region although limited due to spatial variability.

Provided that water is conveyed to agricultural fields/farms, energy (electricity) provision must be assured to support food security amid climate extremes, such as droughts (Taguta et al. 2022). A water–energy–food (WEF) nexus framework is required to respond to drivers of change, such as climate variability and change. The WEF nexus is broadly defined as an approach that considers the interactions, synergies, and trade-offs of WEF when managing these resources (Mabhaudhi et al. 2016; Ololade et al. 2017; Nhamo et al. 2020). WEF securities are inextricably linked, with usage within one sector influencing use and availability in adjacent sectors. Unlike Integrated Water Resource Management (IWRM), which is water-centric in nature, the goal of the WEF nexus is to approach resource management more holistically using a multi-centric philosophy. The WEF nexus presents an opportunity for different actors to integrate sectors to optimise the use of the resource base, maximise synergies, and minimise trade-offs and conflicts.

Several studies have proposed various WEF frameworks. Ringler et al. (2013) presented the concept of the water–energy–land and food (WELF) nexus. The WELF nexus framework evaluates the linkages among the water, energy, land, and food sectors. The direct and indirect drivers of change affecting these linkages are clearly depicted in the framework. Conway et al. (2015) examined Southern Africa's nexus from the perspective of climate and a modified Hoff's nexus framework (Hoff 2011), which integrates global trends (drivers) with fields of action, to highlight the role of climate as a driver. The framework in this study considered the main elements of intra-regional links that occur in WEF sectors at a national level, while highlighting connections at the river basin scale and drawing attention to case studies of many examples of specific trade-offs and synergies.

Smajgl et al. (2016) presented a sectorally balanced, dynamic WEF nexus framework in which sectoral objectives were given equal weightings. The analyses show that this type of framework reveals the emergence and/or changes in cross-sectoral connections because of single-sector interventions. Karabulut et al. (2018) proposed a synthesis matrix system that describes the complex and closely related relationships between the natural resources used for food (specifically water and land), energy (defined as ecosystem service flows in the matrix system), and ecosystems within the WELF concept. The matrix system can be defined at different scales (from global to local) and includes the impacts and nexuses of climate change. Martinez-Hernandez et al. (2017) proposed a simulation and analytics framework and a concomitant Nexus Simulation System termed ‘NexSym’. This study aimed to develop a framework or tool for integrated resource assessment, accounting for integration within and across WEF sectors, ecosystems, and consumption components that interact with a local system. Martinez-Hernandez et al. (2017) indicated that there is a need for a nexus tool on a local scale as solutions are better tailored to local conditions and it becomes easier to achieve synergistic techno-ecological interactions.

This study demonstrated that droughts in Southern Province, Zambia are generally aggressive; however, during strong El Niño events, they tend to be regressive. Regressive droughts dissipated by February; subsequently, precipitation ensured water availability, as demonstrated by the SPI. Further investigation of vegetation cover using SVI showed that during El Niño events, vegetation cover tends to be resilient owing to residual soil moisture. Further investigation is required to evaluate the dynamics of vegetation cover in relation to soil moisture across regions. The SPI and SVI are appropriate for tracking the onset, development, and termination of droughts induced by ENSO in this region. Furthermore, this study demonstrated that water and food security must be achieved using a WEF nexus approach. Innovations such as MAR and renewable energy options must be used for sustainable resource utilisation. In addition, delayed planting coinciding with an increase in rainfall during regressive droughts is a potential strategy for ensuring food security. The results of this study elucidate the trends in drought during El Niño events and serve as a valuable reference for drought adaptation strategies. Using the WEF nexus approach, prospects for sustainable environmental management and improved livelihoods should be investigated using a modelling framework in the future.

This work received support from the Tipping Points Explained by Climate Change Project with funding from SASSCAL (The Southern African Science Service Centre for Climate Change and Adaptive Land Management). Partial funding was also received from NEPAD-SANWATCE and implemented through the WAFSA Fund.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.