The Application of Earth Observation Data to Desert Locust Risk Management: A Literature Review

Abstract

The desert locust is documented as one of the most destructive polyphagous plant pests that require preventive or proactive management practices due to its phase polyphenism, rapid breeding, transnational migration, and heavy feeding behaviour. Desert locust situation analysis, forecasting and early warning are complex due to the systemic interaction of biological, meteorological, and geographical factors that play different roles in facilitating the survival, breeding and migration of the pest. This article seeks to elucidate the factors that affect desert locust distribution and review the application of earth observation (EO) data in explaining the pest’s infestations and impact. The review presents details concerning the application of EO data to understand factors that affect desert locust breeding and migration, elaborates on impact assessment through vegetation change detection and discusses modelling techniques that can support the effective management of the pest. The review reveals that the application of EO technology is inclined in favour of desert locust habitat suitability assessment with a limited financial quantification of losses. The review also finds a progressive advancement in the use of multi-modelling approaches to address identified gaps and reduce computational errors. Moreover, the review recognises great potential in applications of EO tools, products and services for anticipatory action against desert locusts to ensure resource use efficiency and environmental conservation.

1. Introduction

The desert locust (Schistocerca gregaria Forskål, 1775), which belongs to the class Insecta in the order Orthoptera of the family Arcrididae, is documented as one of the most destructive polyphagous plant pests [1,2]. The pest requires preventive or proactive management practices due to its phase polyphenism, rapid breeding, transnational migration and heavy feeding behaviour. Historically and most recently during the 2019–2022 upsurge, desert locusts have been known to destroy vegetative biomass in agricultural farms, rangelands, ranches and natural forests, leading to massive damage and great risk to people and the environment [3]. Unlike other grasshoppers, the desert locust exhibits polyphenism, which is a transition from the solitary to gregarious phases [4]. This plastic reaction to population density is linked to the pest’s transformations in behavioural, morphological and physiological traits [5]. When the numbers of desert locusts are low, the pests exist as harmless, solitarious insects, and when the population is high, they behave as very destructive, gregarious hopper bands and swarms that pose a severe threat to plants [6]. The transition from harmless and solitarious to destructive and gregarious insects starts with an outbreak within a limited area [4]. This phase transition is triggered by high rainfall that leads to increased green vegetation, which, in turn, causes the rapid breeding of desert locusts [7].

Low-density populations of solitary desert locusts always exist naturally without posing any danger in the extensive recession area composed of approximately 16 million square kilometres that stretches from Mauritania in western Africa to India in southwest Asia [8,9]. However, the pest migrates to invaded countries during upsurges and plagues. Desert locust invasion area is estimated to cover up to 31 million square kilometres, affecting about 20% of the earth’s agricultural land [10]. In northern Africa, the Middle East and southwestern Asia, as well as western and eastern Africa, the desert locust is considered the most destructive agricultural pest in more than 60 countries [11]. The biology, behaviour and migratory patterns of desert locusts are influenced by ecological factors especially vegetation greenness and density, meteorological elements such as precipitation patterns, temperature dynamics, wind direction and speed, edaphic factors such as soil texture and water content and geomorphological features, especially topography and altitude [10]. These factors either inhibit or stimulate desert locust breeding, gregarisation and migration. For example, high amounts of precipitation contribute to vegetation boom, moist sandy soil is necessary for oviposition, while winds influence gregarisation and migration [12]. Ineffective monitoring due to natural or humanmade challenges leads to outbreaks, upsurges, plagues and the subsequent migration of swarms across the Sahel-to-Maghreb region of northern Africa or over the Arabian Peninsula to East Africa or Indo-Pakistan regions [13].

The desert locust has three distinct development stages of incomplete metamorphism: egg, nymph/hopper and adult [11]. Under optimal conditions, the lifespan of gregarious desert locusts is about 3 to 4 months, and hence, 3 to 4 progenies can emerge annually during upsurges and plagues [5]. Gregarious desert locusts can remain sexually immature for several months, awaiting conducive breeding conditions [1]. Typically, immature swarms require more than 20 mm of rainfall to initiate sexual maturity, copulation and oviposition [11]. According to a publication by the World Meteorological Organization (WMO) and Food and Agriculture Organization (FAO) [14], after oviposition, the hatching of eggs can take place within 2 weeks, hoppers can fledge in 6 weeks and adults can mature sexually in 4 weeks, on average. However, desert locust outbreaks mostly occur undetected, and the geospatiotemporal trends of invasions are, therefore, challenging to predict with precision [9].

Desert locust oviposition commences within 2 days after copulation [15]. The female desert locusts lay eggs in 3–4 intervals of 7–10 days in pods which are strategically placed 10–15 cm below the surface of moist sandy soils. One female desert locust typically oviposits in 3–4 pods of 60–80 eggs each, with up to 20% viability [1]. Egg development takes approximately 2 weeks, but this highly depends on soil moisture content and temperatures; thus, below 15 °C, egg development aborts. At optimum temperatures of 32–35 °C, the incubation period takes 10–12 days [1]. After incubation, each successful egg hatches into a wingless nymph/hopper, which undergoes 6 or 5 instars during the solitary and gregarious phases, respectively, with each phase having a unique growth rate, as well as morphological and behavioural changes [4].

The gregarious hoppers take approximately 6 weeks from hatching to fledging, depending on the prevailing temperatures. This stage of development can take 22 days at about 37 °C and may lag for up to 70 days at temperatures below 22 °C [6]. The 5th instar nymph fledges into a sexually immature adult, which is the most injurious to green vegetation. Generally, it takes about 4 weeks before an adult desert locust is sexually mature, but the development is similarly highly dependent on temperatures and rainfall. Under warm and wet conditions, with plenty of green vegetation, an adult desert locust may become sexually mature within 3 weeks [15]. Under cooler and drier conditions with dry or no vegetation, sexual maturity can take up to 8 months. Male desert locusts mature before females, and their bright yellow colouration is an indicator of this maturity.

To ensure survival, desert locusts migrate at night to their traditional winter, spring and summer breeding areas during the solitarious phase [1,14]. However, during upsurges and plagues in the gregarious phase, desert locusts migrate during the day to invasion areas in search of food and suitable breeding grounds. Gregarious swarms can migrate at a speed of 16–19 kilometres per hour, depending on the wind speed and direction, and they can travel 150–200 kilometres daily [16]. Before embarking on a migratory journey, a desert locust swarm warms up by basking in the sun shortly after sunrise to benefit from thermodynamics, and hence, sunny days allow desert locusts to fly for long hours [9]. The rapid reproduction and long-distance migratory behaviour pose a great threat to human beings and natural ecosystems. As such, earth observation (EO) data have become important in desert locust management, especially the prediction of suitable habitats where continuous monitoring should be conducted to identify early signs of outbreaks [17,18,19,20].

2. Methodology

This review article seeks to outline the progress that has been made in the utilisation of EO data to complement field observations in desert locust management. The review focuses on the use of EO technology in mapping the spatiotemporal distribution of the desert locust, an analysis of the factors that affect the breeding and migration of the pest, an assessment of the impact through vegetation change detection, the modelling of the prevailing situation and the forecasting of future risk for early warning. The authors used the following structured query with keywords and Boolean operators that filtered all English publications from 1973 to January 2025: Title-Abs-Key ((“desert locust” or “Schistocerca gregaria”) and (“earth observation” or “satellite” or “remote sensing” or “Landsat” or “MODIS” or “NDVI” or “geographic information system”)) from the Scopus database. The search returned 69 documents from 56 sources published by 206 contributing authors.

This automated selection of the most relevant publications was then followed by a manual “snowballing” process where reference lists from each of the previously identified publications were used to confirm that no important study had been missed out on. This process found an additional 21 important documents that were reviewed. The authors read all abstracts, methodology sections and concluding notes and sometimes the entire text of the identified publications to understand how the content addressed the use of EO tools, products and services in desert locust management. The results of the review are presented thematically and chronologically with brief highlights of the advancement in the application of EO data in desert locust management. The authors also identified some gaps from each publication, as well as how these were addressed in subsequent studies, and finally outlined some potential opportunities for further research.

3. Results

The prospects of using EO technology in desert locust management were foreseen as early as the 1960s [16]. However, initial studies documenting the application of EO products in desert locust management were conducted using Landsat multi-spectral data to detect the green vegetation in the recession region of northwest Africa [21,22,23]. In cognizance of the potential of satellite remote sensing (RS), the subsequent two decades included several studies published on how satellite imagery could be utilised in desert locust control operations to provide valuable information. In 1981, satellite RS was tested in monitoring suitable desert locust habitats [24]. Later, advanced very-high-resolution radiometer (AVHRR) and Landsat satellite images were tested in forecasting desert locust outbreaks [25].

Even with these early and promising results, Klein et al. [18] noted that studies on the application of RS in desert locust management were rare until the early 2000s. The authors attributed the increased studies on EO in desert locust management to collaborative efforts among research agencies, institutions of higher learning and the FAO, rather than government initiatives or the motivation of practitioners in the affected countries. Moreover, the increase in research activities could have been influenced either by desert locust upsurges, rather than technological advancements in EO technology, or the improved availability of RS data. For example, the 2003–2005 plague and the 2019–2022 upsurge triggered an increase in research studies that used EO data. The EO technology has been used to map the spatiotemporal distribution of the desert locust, analyse factors that affect the breeding and migration of the pest, assess its impact through vegetation change detection, model the prevailing situation and forecast future risk [26].

3.1. Mapping of Desert Locust Distribution Using Eearth Observation Data

Solitary desert locusts live and breed in arid and semi-arid lands (ASALs) that span from west Africa through the Horn of Africa and the Middle East to southwest Asia [27]. As such, countries along the Sahel belt from Mauritania and through Mali, Niger and Chad to Sudan, those in the Horn of Africa, especially northeastern Ethiopia, Eritrea, Djibouti and northern Somalia, are susceptible to desert locust outbreaks [28]. The Arabian Peninsula states of Oman, Yemen and Saudi Arabia, as well as Pakistan and India, are also vulnerable. However, in the 1980s, most studies on the application of EO in determining the spatiotemporal patterns of desert locust infestations focused on China, Australia, Mauritania, Uzbekistan and Kazakhistan [18]. In the 1990s, there were studies that produced spatiotemporal maps of potential desert locust vulnerable locations in recession areas of Sahel using Landsat data [29]. Lazar et al. [30] integrated 43 years of field data in combination with selected Landsat images to identify desert locust breeding hotspots during the solitary phase. Triggered by the 2003–2005 plague in west Africa, there was an increase in research studies that used EO data to explain the spatiotemporal patterns of the desert locust, albeit with limited local specificity. During the recent 2019–2022 upsurge, there were research studies focusing on eastern Africa, the Arabian Peninsula and southwest Asia that made use of EO datasets [12,31,32,33,34].

In addition, very-high-resolution (VHR) EO data from unmanned aerial vehicle (UAV) and very-high-spatial-resolution satellite sensors such as the WorldView-3, GeoEye and SuperView could be able to detect accumulation of locusts [18,35,36]. However, there is little evidence of the application of these EO resources in the tracing and real-time tracking of desert locust hopper bands or swarms on static roosts or in motion. In an earlier review, Latchininsky [16] noted that radio detection and ranging (RADAR) technology had shown potential to track desert locusts as early as 1968. However, the author acknowledged that the high installation and maintenance costs of RADAR, coupled with cumbersome data analysis procedures, made it impractical, and hence, the technology remained underutilised. However, Anjita and Indu [37], as well as Anjita et al. [38], explored the potential use of Doppler weather radar (DWR) to identify and track desert locust swarms in near-real time through single- and dual-polarisation techniques. The findings revealed that DWR had the capacity to forewarn of an imminent threat of desert locusts several hours prior to their observation within a geographical scope of about 100 kilometres.

3.2. Analysis of Factors That Influnce the Distribution of Desert Locusts

The effective use of EO technology in desert locust management requires the acquisition of multiple datasets from different sources [39,40]. These datasets carry information such as the status of vegetation, soil characteristics, precipitation patterns, temperature variations, wind parameters, elevation, geomorphological characteristics and desert locust presence or absence [41,42,43]. In addition, Latchininsky [16] reported that desert locusts require solar energy, which can be monitored using EO technology.

3.2.1. Habitat Suitability Mapping Through Vegetation Change Detection

Green vegetation is necessary for the survival of desert locusts [14]. Some studies in the 1980s and 1990s demonstrated the potential of the EO-normalised difference vegetation index (NDVI) in monitoring desert locusts’ preferred habitats [23,25,44,45,46]. The Satellite Pour l’Observation de la Terre Vegetation (SPOT-VGT) spectral bands’ integration of 10 days’ NDVI at a 1-square-kilometre spatial resolution identified favourable conditions for desert locust survival [47]. The study also highlighted errors of commission and omission and recommended the addition of red, near-infrared (NIR) and shortwave infrared (SWIR) spectral bands to enhance the validity and reliability of results. In addition, the study proposed the addition of moderate-resolution imaging spectroradiometer (MODIS) data to reveal sparsely distributed vegetation cover, which could easily be omitted due to the coarse spatial resolution of SPOT-VGT NDVI. Later Ceccato et al. [48] published research on the application of 16-day MODIS NDVI to assess weather variability, which could be integrated into early warning systems to monitor preferred desert locust habitats and breeding areas.

Errors of omission and commission associated with the inaccurate identification of sparsely populated vegetation in ASALs, as noted by Ceccato [47], were partially addressed by Pekel et al. [49], who proposed a more reliable MODIS-based multi-temporal technique. The authors also developed a colour transformation procedure that helped identify areas with green vegetative biomass in near-real time. The procedure converted the red, green and blue (RGB) spectral bands to hue, saturation and value (HSV) parameters. Hue was used as a qualitative spectral index, whereas its temporal variations could be interpreted as a change in vegetative biomass. Cressman [8] confirmed that 11 periods of a 10-day vegetation greening EO satellite composite proposed by Pekel et al. [49] almost corresponded with the duration of an entire progeny of the desert locust life cycle.

Working towards the verification of Pekel et al. [49], Waldner et al. [50] assessed the reliability of 10-day dynamic greenness maps and confirmed their reliability in identifying desert locust summer breeding areas (F-score of 0.64 to 0.87). However, the findings were unreliable in winter breeding zones (F-score of 0.28 to 0.40). The authors also reported that the reliability of the MODIS-based multi-temporal technique was anchored on geomorphological fragmentation (R² = 0.9). The authors acknowledged that MODIS spatial resolution was too coarse to address the challenges of complex landscape patterns, which accounted for 60% of commission and omission errors documented by Ceccato [47] To address this seemingly unique and important gap, Waldner et al. [50] compared a project for on-board autonomy–vegetation (PROBA–V) using 100 m resolution data and reported that a higher spatial resolution reduced the aforementioned errors in fragmented areas by 20% and hence increased the vegetation categorisation quality.

Renier et al. [51] tracked the onset of vegetation senescence by integrating temporal NDVI trajectories and the normalised difference tillage index (NDTI), which was sensitive to both green and dry vegetation. The authors used MODIS SWIR bands to calculate 11 different metrics, which were used to categorise vegetation into three phenological classes: growth, density reduction and drying. This categorisation was adopted and expanded by FAO to explain vegetation status as greening, green, drying and dry and density as high, medium and dense [52]. However, this categorisation could be unreliable in invasion countries where the agroecological zones are diverse with perennial vegetation in some locations and in areas where anti-desertification programmes have introduced evergreen invasive species such as Prosopis (Prosopis juliflora), which belongs to Fabaceae family of plants. Moustafa and Cressman [53] utilised MODIS satellite images to compute enhanced the vegetation index (EVI) and developed risk maps of desert locust breeding areas. The authors proposed a simple reclassification procedure which was based on an analysis of EVI output and locust survey data collected in Egypt during the 2011 desert locust outbreak. The findings revealed that areas with EVI < 0.05 had low risk, and areas with EVI = 0.05–0.1 had large, low to medium-density infestations, while areas with EVI > 0.1 had large, medium- to high-density infestations. This EVI reclassification proved accurate during the 2013 desert locust outbreak [53].

Eltoum et al. [41] made use of serial satellite images to produce a desert locust vegetation risk map and observed significant correlations between meteorological, edaphic, ecological and other geographical parameters. Kimathi et al. [32] used a 10-day RS composite from SPOT-VGT and MODIS to assess the status of vegetation development. However, the authors acknowledged that the insufficient assessment of temporal variations in vegetation type, density and prevalence, indicating that these limitations could have had restrictions in their findings. Klein et al. [18] recommended the use of Sentinel-1 and Sentinel-2 data to address the challenge of spatial and temporal resolution. However, Piou and Marescot [54] cautioned scholars from focusing more on reducing computational errors to improving the quality of field survey data and considering a comprehensive understanding of population dynamics. The crowd-sourcing of data from untrained citizens using a GPS-enabled application on smartphones could also compromise the quality of data [55].

To address concerns raised by Piou and Marescot [54], a study by Lawton et al. used spatiotemporal hierarchical patch dynamics theory to investigate the influence of preceding vegetation growth on desert locust outbreaks [56]. The study identified three spatial levels (species range > geographic region > land unit) and a temporal scale between seasons. The authors used MODIS NDVI data as a measure of vegetation growth in hierarchical generalised additive models at different scales, and the results revealed that the observation of desert locusts in the field was preceded by vegetation growth 32 and 20 days before their appearance. The study demonstrated that, although vegetation growth depicted the possibility of outbreaks, the temporal pattern of NDVI differed between the spatial and temporal levels. The model selection criteria for a similar spatial hierarchy, geographic region > land unit, supported the hierarchical patch dynamics paradigm. By acknowledging the spatiotemporal patterning of desert locust densities, the study accounted for the heterogeneity of population dynamics, and the results demonstrated the importance of incorporating spatiotemporal variation in modelling population dynamics.

3.2.2. Soil Feature Analysis to Identify Desert Locust Breeding Zones

Desert locusts require moist, sandy soils for breeding [57]. However, Latchininsky et al. [58] acknowledged the limitations of using satellite soil-based parameters to explain the suitability of desert locust breeding sites. Escorihuela et al. [59] derived soil moisture estimates at 100 m spatial resolution by synergising Sentinel-1 EO data with soil moisture and ocean salinity (SMOS) and developed a product that could be applied to desert locust management. Gómez et al. [60] evaluated the influence of soil characteristics on solitarious desert locusts using the European Space Agency Climate Change Initiative (ESA-CCI) soil moisture product with a 0.25° spatial resolution. The authors analysed the relationship between the presence of desert locusts and soil moisture at different time intervals and concluded that shorter time intervals of 6 days produced the best results. According to the study, the period between 72 and 95 days before desert locust nymphs were observed in the field was the most critical in forecasting and early warning.

Gómez et al. [61] utilised EO data to reveal that soil moisture at root zones under different time scenarios was an important variable when assessing desert locust breeding zones. In addition, Kimathi et al. [32] made use of the average soil moisture and sand content at 5–15 cm below the general ground surface to determine potential breeding hotspots in East Africa. Dandabathula et al. [62] assessed soil texture and moisture to identify the sites that were preferred for oviposition by sexually mature female desert locusts. The study extracted sand–silt–clay percentages at breeding sites for 2017–2021 using SoilGrids ver. 2.0 from the World Soil Information Service database and Soil Moisture Active Passive (SMAP) mission extracts at Level-4 data products for all the identified locations. The results revealed that female desert locusts preferred sandy loam-textured soils for oviposition with 5–10% and 10–20% surface and sub-surface soil moisture, respectively. Dandabathula et al. [62] concluded that damp soil at the ground surface level was required to initiate oviposition.

3.2.3. Precipitation Estimation to Identify Potential Desert Locust Habitats and Breeding Zones

Precipitation (rainfall) triggers vegetation growth and, hence, inhibits or stimulates desert locust breeding, gregarisation and migration [14]. Ceccato et al. [48] published a report on the application of 10-day satellite precipitation estimates to assess whether weather variability could be integrated into early warning systems to monitor preferred desert locust habitats and breeding areas. Ceccato et al. [63] assessed the desert locust plague of 2003–2005 in west Africa using precipitation estimates to forecast future outbreaks. However, the authors observed that the seasonal prediction of rainfall in north Africa is affected by unpredictable mid-latitude storms and suggested that long-term rainfall forecast could be enhanced in areas where oceanic atmospheric circulation conditions evolved slowly. Dinku et al. [64] evaluated the precipitation detection ability of satellites over the desert locust recession ASALs that cut across northwest Africa to northwest India. The study revealed that the greatest limitation of RS precipitation data was over-estimation bias, especially in favour of dryness.

Most studies on the application of EO data focused on the analysis of precipitation patterns to forecast potential gregarisation and guide the planning of field surveys [65]. Kimathi et al. [32] utilised 1970–2000 WorldClim2 precipitation and desert locust field data from the early stages of the 2019–2022 upsurge to determine potential breeding hotspots in East Africa. However, the study did not assess how precipitation could have affected the migration of desert locusts and the persistence of infestations. Wang et al. [66] used precipitation to identify potential desert locust habitats and determine China’s risk of desert locust invasion from India and Pakistan following the 2019 outbreaks. The authors noted that, although rainfall helped identify potential habits, it could not be used without wind patterns to determine the probability of invasions. Ellenburg et al. [57] observed that sudden high amounts of precipitation contributed to a vegetation boom, resulting in an increased desert locust population due to rapid breeding. However, while the use of long-term climatic data to derive precipitation means for suitable desert locust habitats may be useful in recession areas during the solitary phase, it might be unreliable during the gregarious phase, given the shorter life cycle, change in behaviour and long-distance migration to more heterogenous areas.

3.2.4. Temperature Trend Assessment to Determine Suitable Desert Locust Habitats

The biology, behaviour and migratory patterns of desert locusts are influenced by temperatures [6]. To demonstrate the importance of temperature in desert locust management, Gómez et al. [61] utilised MODIS land surface temperature (LST), air temperature data and other bioclimatic variables to develop a desert locust habitat suitability map during the solitary phase in the recession period. However, the authors observed that the time of temperature data retrieval is critical in ASALs with high day–night temperature ranges. Wang et al. [66] used temperature to determine suitable desert locust habitats in China but concluded that the risk of invasion was negative due to unfavourable wind patterns. Kimathi et al. applied 1970–2000 WorldClim2 temperature and other bioclimatic datasets to identify potential breeding areas for the desert locust in eastern Africa [32]. However, the use of long-term climatic data to evaluate suitable temperature and precipitation for desert locusts may be unreliable in invasion countries where weather patterns are heterogenous and given the short lifecycle of the desert locust, as well as changes in biology and behaviour during the gregarious phase.

3.2.5. Wind Pattern Analysis to Predict Desert Locust Movements

Wind direction and speed play an important role in the gregarisation and migration of desert locusts [27]. Wang et al. [66] simulated windborne movements of the desert locust at different altitudes to determine China’s risk from India and Pakistan following the 2019 outbreaks, and the authors revealed a negative possibility of invasion. Boultif et al. [67] applied spatial interpolation of wind direction and speed to calculate an RS index that helped categorise desert locust-suitable habitats in Algeria. The authors concluded that wind only contributed to movement and gregarisation. Mitra et al. [42] found that wind direction contributed to desert locust distribution, but wind speed had a minor influence. Retkute et al. [43] developed a comprehensive framework that revealed that short- and long-term swarm movements were influenced by turbulent wind flow as a product of atmospheric dispersion. Although wind plays a critical role in desert locust movement, there was limited research evidence on the use of EO products and services to strengthen its application in situation analysis and forecasting the future risk of desert locust infestations. Only a few studies included either wind direction or speed in modelling desert locust habitat suitability [42,43,67].

3.2.6. Other Factors That Influence Desert Locust Infestations

Apart from vegetation, soil, precipitation, temperature and wind, several other factors were mentioned by authors as having contributed to the distribution of desert locusts in either recession or invasion areas. For example, Piou et al. [68] observed a non-linear causal association between mean vegetation density and desert locust presence, without considerations of geomorphologic variables such as wadis, sea beaches, lake shores and river lines, which served as desert locust breeding sites. Waldner et al. [50] reported that the MODIS-based multi-temporal technique, which had been used extensively to determine suitable desert locust habitats, was anchored on geomorphological fragmentation. Boultif et al. [67] identified proximity to water bodies and topographical elevation as additional important factors in predicting suitable desert locust habits in Algeria.

Liu et al. [69] reported that altitude and geomorphological features were important factors, reporting that the Himalayan mountains in southern Tibet provided unconducive weather with low temperatures and high humidity, where desert locusts could neither live nor breed. As a result, the mountains served as a natural barrier against farther northward migration. Mitra et al. [42] revealed that the presence of cloud cover, coupled with low to moderate temperatures, had a significant impact on locust occurrence and migration. Although several publications documented that desert locusts had to bask in the sun to activate their flight [9,16], there is hardly any evidence of studies that attempted to include the intensity of solar radiation in determining persistence or early departure from an infestation site.

3.3. Desert Locust Impact Assessment Through Vegetation Change Detection

Desert locust impact can be assessed using EO data by monitoring changes in the vegetation cover. This is possible because stressed or damaged vegetation is characterised by a difference in reflectance compared to healthy vegetation [70]. According to Klein et al. [18], stressed vegetation can be detected through RS imagery using the red edge of the electromagnetic spectrum due to a loss of chlorophyll. The authors reported that an extreme loss of green vegetative biomass is usually visible through computing a change in vegetation index (VI). Through the use of spectral reflectance and high-resolution synthetic aperture radar (SAR), changes in canopy cover and structure are detectable [67]. For instance, Ma et al. [71] assessed the relationship between vegetation biomass measurements, the leaf area index (LAI), the Landsat-based NDVI, the atmospherically resistant vegetation index (ARVI) and locust presence, obtaining positive results (R² = 0.6474).

In a different study, MODIS multi-spectral indices were used to carry out temporal filtering, and the results showed that NDVI was the best damage assessment index for locust risk [72]. Liu et al. [73] and Tian et al. [74] used Landsat NDVI to compute the differences before and after outbreaks. Zha et al. [75] introduced the locust density index (LDI), which computed the difference between the initial and destroyed vegetation after infestation to estimate losses. Hunter et al. [76] utilised airplane EO data to analyse vegetation damage with hopper bands, which were clearly visible in RGB images. However, Weiss [77] unsuccessfully used a MODIS 1-kilometre square temporal satellite RS composite to estimate vegetation damage by locust hopper bands. Song et al. [78] analysed drone EO data to assess the loss of reeds due to locusts. Based on the aforementioned studies on other locust species, there is evidence that EO could help resolve challenges in the manual assessment of crop damage by desert locusts. However, the use of EO products to assess economic losses from desert locusts in monetary terms has rarely been demonstrated.

The hidden Markov model (HMM) was utilised by Shao et al. [79] to predict the severity of desert locust infestation in croplands using time-series dynamic change features extracted from EO data. The authors also assessed crop damage using change detection methods by comparing the crop spectrum before and after desert locust infestations from two-phased hyperspectral images covering a sub-study area of northern Narok in Kenya. The results indicated that the severity of crop damage varied with an irregular trend in April, May, June and July (0.78, 0.71, 0.74 and 0.72), respectively. The land cover classification of the Kenyan sub-study area for the two-phase images was 97.45 and 96.14. The authors concluded that cropland changes using hyperspectral images could be used to quantitatively compute the damage caused by desert locusts. However, Narok was one of the least desert locust-affected counties in Kenya [3], and hence, results from hyperspectral image analysis could have been unreliable. In a similar study, Adams et al. [80] analysed time-series data from MODIS, the harmonized Landsat and Sentinel-2 product, and C-band radar data from Sentinel-1. The authors distinguished between desert locust crop damage from either senescence or other confounding factors such as water stress. The results, however, did not accurately quantify damage from the desert locust due to the sporadic and localised nature of infestations, which could have had an influence on the timing, scale and unexpected vegetation conditions. The authors recommended the use of higher spatial resolution satellite products such as the Planet Labs protocols for plant damage assessment.

Laneve et al. [70] assessed desert locust crop damage by integrating remotely sensed indicators such as precipitation, vegetation greenness, NDVI, air temperature, LST and soil moisture. The authors used a weighted method to construct a habitat suitability index and to quantitatively extract large areas of locust breeding areas before extracting the major vegetation types such as farmland, grassland and shrubland in the study area.

“For the vegetation with stable periodicity growth curves, the damaged area monitoring of the desert locust was conducted by comparing the vegetation index after the infestation and the average vegetation index over the past years. For other vegetation, the damaged area monitoring was conducted by simulating the vegetation growth index of the same meteorological conditions and the same growing period and comparing it with the actual situation after the infestation.”

Although the study provided a detailed procedure for vegetation change detection after desert locust infestations, it did not demonstrate how to remediate plant effects from water stress, physiological disorders and pathological attacks.

Alemu and Neigh [81] assessed desert locust cropland damage by analysing 5–10 m spatial resolution EO data using 121 swarm presence points and 94 random absence locations within 20–25 km from the infestation epicentre. The findings showed that vegetation health indices (VHIs) and integrated drought condition indices (IDCIs) computed for the affected sample sites from 2000 to 2020 were strongly correlated (R² > 0.90) with those of the corresponding unaffected locations. Drought indices were strongly correlated with the computed standardised precipitation evapotranspiration indices (SPEIs), indicating that, since 2000, 2020 was the wettest year. The authors concluded that the slightly wider but significant gap of cropland phenologies between desert locust-infested versus uninfected cropland sites based on NDVI and backscatter coefficient was likely due to desert locust damage and not drought.

The utilisation of VHR data from UAV and advanced satellite sensors such as WorldView-3, GeoEye and SuperView can be used to monitor damaged plants [18]. Unfortunately, there was little evidence of the application of these EO resources to assess vegetation losses due to desert locust infestations. In addition, Piou and Marescot [54] noted that empirical studies that explained a reliable procedure to assess desert locusts’ impact on the environment were lacking. The authors recommended that impact assessment should go beyond the estimation of vegetation damage to associated ecosystem biodiversity losses. To help in biodiversity loss assessment, hyperspectral EO data from PRecursore IperSpettrale della Missione Applicativa (PRISMA) and the Environmental Mapping and Analysis Program (EnMaP) can produce unique spectral signatures for the affected plants [82,83]. Since desert locusts are polyphagous pests foraging on diverse plant species, it is important to explore how hyperspectral EO data can be used to produce plant species-specific risk maps.

3.4. Desert Locust Risk Modelling Throuth Earth Observation Data Fusion

Through big data fusion, desert locust survey reports have been integrated with EO data such as vegetation indices, precipitation estimates, temperature dynamics, soil characteristics and wind parameters to explain the breeding, migration and impact of the pest. For instance, Tratalos and Cheke [84] determined whether NDVI time series were related to either phase polyphenism, locust population dynamics or variability in precipitation. Due to the limitations associated with the temporal resolution of EO data, the authors recommended more research studies covering extended time periods to provide additional insights into the relationship between different ecological characteristics, meteorological factors and desert locust population dynamics. Deveson [65] discovered that most EO data fusion in desert locust studies was applied in the analysis of precipitation patterns and vegetation change detection to forecast potential zones of gregarisation, guide the planning of field surveys and assist in risk assessment.

A forecasting model that incorporated historical field survey data and MOD13Q1 NDVI for 16 days at a 250 m spatial resolution to determine the reliability of vegetation change detection as an indicator of desert locust habitat suitability in Mauritania was published by Piou et al. [68]. The study revealed that the maximum NDVI followed the topographical structures at the localised scale. The authors observed a non-linear causal association between mean vegetation density and desert locust presence. However, this could not be analysed without considerations of geomorphologic variables such as wadis and areas with water accumulation including beaches, lake shorelines and riverbanks, which serve as desert locust breeding sites. The authors concluded that desert locust populations grew with the development of vegetation. In addition, they clarified that rainfall and the time delay between vegetation change detection and desert locust observation in the field were useful parameters in predictive modelling. Recognising the limitations of using vegetation indices in the identification of suitable desert locust habitats, Latchininsky et al. [58] presented a MODIS EO data fusion approach in Mauritania.

A machine learning algorithm was applied by Gómez et al. [61] to integrate MODIS NDVI, LAI, SMAP at the root zone, surface soil moisture, MODIS LST, surface temperature and desert locust field survey data. The study revealed that surface temperature, NDVI and soil moisture at the root zone under different time scenarios were the most critical variables when assessing desert locust habitat suitability. The authors concluded that LST from SMAP was the most important parameter, but they acknowledged that the inclusion of more environmental variables increased the predictive performance of the machine learning algorithm (Kappa = 0.901; ROC = 0.986). Piou et al. [85] modelled the relationship among the temporal development of NDVI, soil moisture, rainfall and LST around survey points in recession areas using statistical analysis to assess their partial contribution to desert locusts’ presence. The findings revealed that the greatest contributor of desert locusts’ presence was NDVI (AUC = 0.7264) and then soil moisture (AUC = 0.6280), followed by LST (AUC = 0.6201) and rainfall, which was the weakest dependent variable (AUC = 0.5797). The random forest forecasting model which combined soil moisture and NDVI data obtained positive observations (AUC = 0.761).

Desert locust field data from the early stages of the 2019 outbreaks were used by Kimathi et al. [32] to determine potential breeding hotspots in eastern Africa. The authors integrated the 1970–2000 WorldClim2 temperature and precipitation data, as well as soil moisture and sand content from 5 to 15 cm below the ground surface. The study also included a 10-day vegetation greening EO composite from SPOT-VGT and MODIS data to assess the vegetation development status. The authors, however, acknowledged a lack of detailed assessment of temporal variation in the vegetation type, density and prevalence as restrictions that could have affected the findings. Using a similar simulation approach, Wang et al. [66] determined China’s risk of desert locust invasion from India and Pakistan following the 2019 outbreaks. The authors used precipitation and temperature to identify potential desert locust habitats in China, simulated windborne movements of the pest at different altitudes and observed a negative possibility of an invasion. Wang et al. [12] modelled the causal association among precipitation, temperature, vegetation cover and soil moisture data from EO images, together with long-term meteorological observations, to explain outbreaks across eastern Africa and western Asia. The results revealed that increased precipitation in the Arabian Peninsula during 2018 resulted in increased soil moisture and lush vegetation. The boom in vegetative biomass promoted the breeding, multiplication and gregarisation of desert locusts, leading to an upsurge. In addition, the study revealed that regions affected by heavy rainfall in southwest Asia and northeast Africa attracted migrating swarms from the Arabian Peninsula, confirming that suitable soil moisture and enhanced NDVI coincided with desert locust movements.

Chen et al. [31] fused NDVI, LAI, soil moisture, rainfall between 2005 and 2020, and desert locust field reports to develop a time-series simulation model for extrapolating the potential geographic distribution of the pest in Africa, Asia and Europe. The authors observed that LST and LAI were the two greatest influencing factors of desert locust distribution with 27.02% and 25.63% contributions, respectively. Soil moisture was, however, observed as the weakest contributor (2.7%). This anomalous overall contribution is a result of soil moisture often being high only during the breeding stage of the desert locust lifecycle. After rainfall, soil moisture is high, and vegetation becomes green, allowing locusts to mature sexually and oviposit. But in desert areas, often, little further rain falls at the oviposition site, and hence, by the time nymphs emerge, the soil moisture is much less, leading to a low correlation of soil moisture with the actual presence of desert locusts. However, by using slightly different correlations, several other authors have demonstrated the critical importance of soil moisture.

Gómez et al. [86] utilised six machine learning models (logistic regression model, eXtreme gradient boosting, weighted k-nearest neighbours, feed-forward neural networks and multinomial log-linear models, radial support vector machine and random forest) to predict the presence of desert locust nymphs in the recession region. Through a forward selection procedure, the results revealed that SMOS soil moisture data obtained 12–95 days prior to the observation of nymphs in the field provided sufficient information for forecasting desert locust risks. The model showed that spatiotemporal constraints in data sampling conditioned the predictive capacity of the machine learning algorithms. Boultif et al. [67] used NDVI vegetation cover analysis, the Soil Moisture for Desert Locust Early Survey (SMELLS) based on Sentinel-1 SAR data with thermal disaggregated SMOS, distance from water bodies, topographical elevation derived from the digital elevation model (DEM), the spatial interpolation (IDW) of wind direction and speed in computing EO indices that helped categorise desert locust suitable habitats in Algeria. The authors observed that suitable habitats decreased with an increase in altitude and distance from water resources, as well as the presence of mountains, the scarcity of vegetation and reduced soil humidity, while wind only contributed to movement and gregarisation.

An SVM-based model with a temporal sliding window procedure was proposed by Sun et al. [40]. The model was developed by coupling multi-source time-series imagery with historical desert locust ground survey observations from 2000 to 2020. The results revealed that desert locust nymphs were observed 41–64 days after increased rainfall when soil moisture increased by approximately 0.05 m³/m³, while a subsequent decrease enhanced the chances of observing them after 73–80 days. The authors also noted that areas with sparse vegetation resulting in an NDVI value of 0.18 to 0.25 had nymphs observed after 17–40 days. Rhodes and Sagan [87] modelled desert-locust presence data from Niger and Sudan, multiple pseudo-absence geo-locations, seasonal and annual rainfall, LST, soil moisture, NDVI and land cover data using k-nearest neighbour (kNN), decision tree, support vector machine learning algorithm, random forest, maximum entropy (MaxEnt) and deep neural network (DNN) models. The results indicated the kNN model returned an average accuracy score of 88% and F-1 score of 89% for the present (1) class, while the DNN model showed an average accuracy score of 88% and F-1 score of 89% in Niger. The kNN model returned an average accuracy score of 88% and F-1 score of 88% for the present (1) class, while the DNN model exhibited an average accuracy score of 88% and the F-1 score of 89% in Sudan. The authors concluded that, in both regions, kNN and DNN were the best-performing models. There was, however, no published evidence that these models would yield similar results if they were used to compare data from recession and invasion areas during upsurges or plagues.

Landman et al. [33] revealed that nymphs could be monitored and the probability of their time of hatching and location in remote and inaccessible areas over extensive landmass predicted. The authors used a fuzzy rule set by integrating desert locust survey data with rainfall, soil moisture, temperature regimes and vegetation greenness. The findings indicated that high nymph occurrence could be associated with preceding rainfall and vegetation greenness computed from EO data. Although training the model with Sudan data showed positive results for Turkana County in Kenya, it may not be extrapolatable to different agro-ecological zones where breeding occurred in the second wave of desert locust invasion when swarms migrated following different routes. Mongare et al. [88] used the NDVI from 2018 to 2020 Sentinel-2 imagery to assess potential vegetation damage by desert locusts in Turkana County in Kenya and modelled future climatic scenarios using the MaxEnt model. The results revealed that the highest vegetation damage based on NDVI changes occurred between May and July 2020. The results showed that the most critical factors for desert locust survival were temperature and precipitation. The findings also revealed that 27% of Turkana County was highly suitable for desert locusts, but this could fall to 20% due to climate change.

Using an ensemble modelling approach, Tang et al. [26] evaluated the influence of climate, land use and topography on the distribution of desert locusts. The authors reported that the most important factors that influenced desert locust distribution were temperature and precipitation. The authors suggested that some areas that had reported little or no desert locust incidents could become suitable for desert locusts in the future, while suitable habitats could become just migratory paths. In addition, the findings revealed a static niche selection that could be used to explain the reason behind desert locusts retreating back to recession areas after upsurges and plagues, rather than settling in invasion countries. Mitra et al. [42] used weight-of-evidence (WoE) and frequency ratio (FR) models to evaluate nine critical climatic factors when assessing habitat suitability in India. The authors analysed the correlation of occurrence of desert locust hopper bands and swarms with temperature, precipitation, relative humidity, wind direction and speed, cloud cover and visibility. The study used random forest (RF) and principal component analysis (PCA) machine learning algorithms. The habitat suitability maps revealed that western and central India had a high to very high risk of desert locust invasion. The findings also showed that the most important variables were cloud cover, temperature and wind direction, but precipitation, visibility and wind speed had a minor influence on the distribution of desert locusts.

Khan et al. [89] used the LocustLens machine learning model that was derived from kNN to integrate Terra climate environmental data comprising soil moisture, maximum temperature and precipitation, with reverse geocoded desert locust swarm data from 42 countries. The results obtained from LocustLens were compared with baseline kNN, decision trees (DTs), logistic regression (LR), the AdaBoost classifier, the bagging classifier and the support vector classifier (SVC). According to the authors, LocustLens returned the best results (AUC = 0.98) versus the baseline kNN (AUC = 0.96), SVC (AUC = 0.91), DT (AUC = 0.97), AdaBoost (AUC = 0.91), bagging classifier (AUC = 0.94) and LR (AUC = 0.83). These models’ accuracy levels were, however, too high to be relied upon for real-world applications in assessing a natural phenomenon such as desert locust occurrence. Huang et al. [34] analysed spatiotemporal variations of suitable desert locust habitats in Yemen by integrating the MaxEnt model with space–time cube analysis to identify areas with favourable vegetation, soil types, precipitation and temperature for desert locust survival and breeding. The MaxEnt model was proven valid after revealing kappa coefficients > 0.46 and an AUC > 0.75.

A comprehensive modelling framework for predicting population dynamics from EO data by integrating concepts derived from epidemiological, meteorological, environmental and atmospheric transport disciplines was developed by Retkute et al. [43]. The authors fused temperature, precipitation, NDVI, land cover, clay content, sand content, soil moisture, elevation, wind trajectories and atmospheric dispersion. The model identified suitable breeding sites and explained nymphal development, and it accounted for short- and long-term swarm movements, which were reported to be influenced by turbulent wind flow as a product of atmospheric dispersion. The model also explained the feeding behaviour of the pest; the duration of time spent by swarms at a landing site was determined by the availability of vegetative biomass at the site. While the framework offered a promising guide, the use of long-term and global-scale climatic data may be unreliable during the gregarious phase and within the local heterogenous meteorological dynamics of invasion countries. In addition, the framework did not consider some of the parameters recognised by other studies, such as wind speed, humidity, cloud cover, visibility and topography/geomorphological features [42].

Yusuf et al. [90] modelled EO data using a three-dimensional model, LSTM-based recurrent convolutional networks and the Prithvi geospatial foundational model, achieving AUC values of 0.83, 0.82 and 0.88, respectively. The results showed that multi-spectral EO images from harmonised Landsat and Sentinel-2 products at 2- to 3-day temporal and 30 m spatial resolution were effective in forecasting desert locust breeding. Additionally, Chang et al. [91] used MaxEnt ecological modelling to assess potential changes in desert locust habitat suitability due to climate change. The findings showed that suitable desert locust habitats remained in northern Africa and southwestern Asia, and they were greatly influenced by temperature and precipitation. Under climate change simulations, the authors reported that potential desert locust habitats could increase from the year 2030 onwards, and by 2090, highly suitable areas for infestation by the pest would probably expand.

4. Discussion

Most of the major issues that were identified in each of the reviewed publications were highlighted under the results section. However, to emphasise how the objectives of this review were met, the discussion section presents an abridged version of the key findings. The section also outlines the emerging gaps from which a few opportunities for further research were identified.

4.1. Summary of the Findings

Most studies that incorporated data fusion to model desert locust infestations revealed that precipitation, vegetation, temperature and soil moisture were the most considered environmental variables, as depicted in

Table 1. Studies that have integrated several parameters in desert locust risk modelling.

Author (s)	Study Area	Modelling Method	Precipitation	Vegetation	Temperature	Soil Moisture	Altitude	Wind Direction	Land Cover	Soil Texture	Wind Speed	Humidity	Cloud Cover	Visibility
Piou et al. [68]	Mauritania	Logistic regression	√	√			√
Piou et al. [85]	Recession area	Statistical analysis and Random forest model	√	√	√	√
Kimathi et al. [32]	East and Horn of Africa	MaxEnt model	√	√	√	√				√
Wang et al. [12]	East Africa and West Asia	Multi-modelling approach	√	√	√	√	√
Gómez et al. [86]	Recession area	Multi-modelling approach				√
Boultif et al. [67]	Algeria	Multi-criteria analysis		√		√	√	√			√
Sun et al. [40]	Recession area	SVM-based model	√	√		√
Rhodes and Sagan [87]	Niger and Sudan	Multi-modelling approach	√	√	√	√			√
Landman et al. [33]	Sudan and Kenya (Turkana County)	MaxEnt model	√	√	√	√
Mongare et al. [88]	Kenya (Turkana County)	MaxEnt model and True skill statistic	√	√	√
Tang et al. [26]	Recession and invasion areas	Multi-modelling approach	√		√				√
Mitra et al. [42]	India	Multi-modelling approach	√		√			√			√	√	√	√
Khan et al. [89]	Recession and invasion areas	Multi-modelling approach	√		√	√
Huang et al. [34]	Yemen	MaxEnt and space–time cube analysis	√	√	√	√
Yusuf et al. [90]	Recession area	Multi-modelling approach	√	√		√			√
Retkute et al. [43]	East and Horn of Africa	Integrated modelling framework	√	√	√	√	√	√	√	√
Chang et al. [91]	Recession and invasion areas	MaxEnt model	√		√

.

Although MaxEnt was the dominant model, most studies used the ensemble technique, where a multi-modelling approach was applied to rule out errors of omission and commission or compare the accuracy of various machine learning algorithms. Landsat data were the most used EO resource, although a few studies used Sentinel datasets [59,67,80,88,90]. Passive EO data were the most commonly used data with only a few studies exploring use of active EO data [18,67]. Although many studies on the biology and behaviour of desert locust recognised solar radiation as an important factor that influenced desert locust movement and flight [1,27], there was hardly any evidence of the inclusion of this parameter in the studies that simulated habitat suitability. However, Mitra et al. [42] included cloud cover, visibility and humidity in their study. In addition, only a few studies considered soil texture, wind speed (winds determine which habitats migrating desert locusts invade), land cover/use, elevation/altitude and geomorphological features/topography [42,50,68,69].

The review revealed that most studies on the application of EO technology in desert locust management focused on vegetation change detection to either identify suitable habitats or assess infestation risk, albeit without a quantification of the losses financially. The NDVI was the most commonly used vegetation change detection index [12,31,40,43,47,48,51,56,61,68,70,81,84,85,87,88]. However, the use of NDVI was faulted for being unreliable in ASALs, which constitute the majority of the desert locust recession area [18,47]. In addition, most of the reviewed studies concentrated on the entire or part of the recession area, with only a few country-specific studies [33,67,68,87,88]. There was hardly any evidence of comparative studies between different plagues, upsurges or even two or more waves of invasions. Although desert locusts migrate in search of food and suitable breeding sites [1,8,16], most studies that modelled breeding did not account for the adult stage of the life cycle [33,40,86,87].

4.2. Research Gaps and Opportunities for Further Research

This review acknowledges the immense contributions of scholars in studying and publishing results on the application of EO data in desert locust management. Although a lot of literature has been published, the review identified the following potential areas which could be explored for further research:

(1)

Several studies recognised the role of climate change in altering the distribution of desert locust habitats [26,28,91,92]. It is likely that climate change could increase the frequency and persistence of desert locust outbreaks, expand both the recession and invasion areas, and worsen the intensity of impact on the environment and people. As such, there should be research efforts to identify potentially new recession areas, the expansion of invasion areas, emerging migratory trends and local breeding zones in invasion countries using EO tools, products and services.

(2)

While the limitation of using NDVI in ASALs to effectively identify suitable desert locust habitats due to spatiotemporal coarseness is acknowledged [18,47], the large-scale categorisation of vegetation using EO data could be unreliable in invasion countries. This is because the agroecological zones in some invasion countries, especially in eastern Africa, are diverse with evergreen vegetation in some locations, while in other areas, antidesertification programmes have introduced evergreen invasive species such as Prosopis juliflora. As such, there is a need to experiment with other vegetation indices such as LAI, the fractional vegetation cover (FVC) index, the atmospherically resistant vegetation index (ARVI), the soil-adjusted vegetation index (SAVI) and the perpendicular vegetation index (PVI) to assess their effectiveness in addressing the limitations of NDVI.

(3)

Most studies focusing on the use of EO products to analyse the spatiotemporal distribution of desert locust infestations were conducted by researchers outside the affected areas and acquired survey data from FAO desert locust information service (DLIS) archives [12,26,33,40,64,85,86,89,90,91]. However, there was hardly published evidence of community participation and input from local agricultural officers. The involvement of these key stakeholders in the research process could help researchers understand the quality of survey data and obtain practical field insights to comprehensively understand desert locust population dynamics and thus corroborate the research findings. Moreover, the majority of studies were carried out during years of or soon after upsurges and plagues. As such, there should be efforts to continue research activities during recession and remission periods to sustain the technical and academic capacity which is necessary for early preparedness and anticipatory action.

(4)

The majority of the studies that were reviewed used positive field reports of desert locusts’ presence and assumed the remaining areas were free of infestations. While this could have been true to some extent, studies have reported that desert locusts inhabit remote and sometimes inaccessible areas [1,16,18]. Assuming that all areas without positive presence reports were unaffected by desert locusts could lead to false negative inputs that could affect the findings of studies that model habitat suitability, pest distribution and impact assessment. Studies should, therefore, strive to incorporate both presence and absence field survey reports as part of the model training and testing datasets to reduce potential errors which could produce invalid results and cause unreliable inference. In addition, most of the studies used either swarm or nymph stage datasets separately. It would be interesting to model the different life cycle stages of the desert locust together or even categorise the data into hopper instars and different adult stages for comparative analysis.

(5)

Based on the reviewed studies, the key factors that affect desert locust distribution are precipitation, vegetation, temperature, soil moisture, soil texture, wind direction and wind speed. However, other important factors such as elevation/altitude and geomorphological features such as mountains and water bodies whose characteristics can be identified through EO technology have received little attention in studies on the application of EO data in desert locust management. In addition, although Latchininsky [16] and the authors of several other studies confirmed that desert locusts need to bask shortly after sunrise to activate their flight, there was hardly any evidence of studies incorporating the intensity of solar radiation in determining its influence on the pest’s persistence or early departure from an infestation site. Moreover, data on human-based parameters such as desert locust control activities and conservancies/protected areas could be incorporated as potential habitats in the modelling environment to complement field surveys. Rhodes and Sagan [87] also recommended the inclusion of agricultural and desertification parameters in modelling potential desert locust habitats. Future studies could, therefore, include these additional parameters in the modelling environment to determine their contributions to desert locust infestations.

(6)

Although there was adequate research on the assessment of habitat suitability to explain desert locust geographic distribution, studies that compare the development of different upsurges and plagues or track swarms from outbreak epicentres to invasion zones during these events were rare. In addition, studies that assessed whether migrating swarms follow regular migration routes to or in invasion countries and local breeding patterns thereafter were also rare. There is also limited evidence of studies that compared either spatial and/or temporal distribution of infestations or the impacts between different invasions or different waves of invasions, for similarities or differences, and factors that could facilitate such tendencies. Moreover, there were limited studies that focused on factors that could explain flight height, sudden landing, early roosting, overstay or early departure from one infestation site to another, especially in invasion areas where agrometeorological factors are diverse.

(7)

Klein et al. [18] suggested that VHR EO data from drones and VHR satellite sensors such WorldView-3, GeoEye and SuperView could detect the gregarisation of desert locusts. Nonetheless, there was little evidence of the application of these EO resources in the tracing and real-time tracking of hopper bands or swarms on static roosts or in motion. In addition, despite RADAR being the earliest EO tool to be considered for use in desert locust management [16], there were only a few publications [37,38]) that had used this technology. Moreover, hyperspectral remotely sensed data from PRISMA and EnMaP have been freely available at 30 m spatial resolution since 2019 [82,83]. However, there were limited studies that used the technology to assess damage to vegetation by desert locusts. More studies that utilise these resources would be beneficial to provide evidence of their usefulness and relative advantage in desert locust management activities over the Landsat and Sentinel EO data.

(8)

Apart from modelling by scholars, Qayyum et al. [36] reported that FAO had continuously incorporated similar technological advancements in the desert locust early warning system through initiatives such as the Schistocerca Warning Management System (SWARMS), Africa Real Time Environmental Monitoring Information System (ARTEMIS) and Reconnaissance and Management System for the Environment of Schistocerca (RAMSES) at the global, continental and country levels, respectively. However, local short-term early warning systems still relied mostly on the knowledge, skills and intuition of national agricultural officers. There was limited evidence of publications on the effectiveness of national-level offices, especially in invaded countries, in integrating EO data into early warning systems for situation analysis, forecasting and early warning.

(9)

Many studies on the application of EO data in desert locust management used products from passive sensors, especially the Landsat constellation, and most recently, the Sentinel constellation, which are usually limited by physical barriers, especially cloud cover. Klein et al. [18] reported that the use of active EO sensors could overcome these obstacles and address the challenges of data gaps. There was, however, limited evidence of studies that use active EO sensors to support desert locust management, and hence, future studies could explore the use of data from SAR, Lidar, and Sonar EO technology.

5. Conclusions

This review has revealed that the application of EO technology is inclined in favour of desert locust habitat suitability assessment. However, there have been few applications of EO tools, products and services in the quantification of socioeconomic loss analysis based on damage to vegetation after desert locust infestations. The review also found a progressive advancement in EO data fusion and the use of diverse modelling techniques to address identified data gaps and reduce computational errors. The study noted the existence of great potential in the application of EO technology to anticipatory action in order to ensure resource-use efficiency and environmental conservation. Through EO and field-survey data fusion, it is possible to carry out real-time situation analyses and guide targeted early responses, forecast future outbreaks or invasions and ensure early warnings for preparedness. As a result, this pre-emptive approach can trigger the mobilisation of requisite resources, inform capacity development programmes and guide the targeted deployment of resources. Moreover, desert locust modelling approaches could be replicated in the management of other migratory pests to safeguard food security. Finally, precise, small-scale interventions for migratory pests can allow integrated pest management, reduce the use of synthetic pesticides and hence preserve ecosystem diversity.