Agrochemicals, machines and biodiversity in Ghana: the ecological cost of land preparation methods of large-scale land acquisition by domestic entities Open Access

Due to its ability to improve soil fertility, pest control and pollination (Mace et al., 2012), biodiversity is essential for maintaining ecosystem services that support agriculture and food security (Bozzola and Smale, 2020; Di Falco et al., 2010; Di Falco and Chavas, 2006). However, a major component of contemporary agricultural development, large-scale land acquisitions (LSLAs), are often driven by local and global demands for food, biofuels and economic expansion, and they usually put immediate agricultural gains ahead of biodiversity preservation (Behrman et al., 2014).

According to several research studies on developing countries, LSLAs are commended for their beneficial effects and capacity to increase agricultural output, create jobs and promote economic growth (Ali et al., 2019; Burke et al., 2020; Deininger and Xia, 2016; Khadjavi et al., 2020). However, their detrimental effects are also well-known and include pollution, water scarcity, animal extinction, deforestation, loss of forest access and conservation, and possible threats to biodiversity and the ecosystem. For instance, studies by Chiarelli et al. (2020, 2022) looked at the connection between LSLAs and water scarcity in Asia and Ethiopia and found that LSLA causes both water scarcity and consumption. According to Davis et al. (2023), who also looked at the effects of LSLAs in 40 Global South nations, LSLAs reduced deforestation by 17% between 2000 and 2018. However, while much of the existing literature has concentrated on the impacts of LSLAs led by foreigners—often involving holdings of 1,000 hectares or more (Cotula et al., 2009)—another form of LSLA is widespread across Africa: those undertaken by domestic entities (LSLADE). These typically encompass medium-scale or large = scale land acquisitions carried out by local citizens, traditional authorities, or government agencies (Abdallah et al., 2023a, b, 2024; Ayamga et al., 2023; Jayne et al., 2016, 2019). According to Jayne et al. (2014), LSLADEs account for substantial areas, including an estimated 1.12 million hectares in Ghana, 0.36 million hectares in Kenya and 0.30 million hectares in Zambia (Jayne et al., 2014), and thus have important potential implications for biodiversity (Wineman et al., 2022). Yet, despite their scale and prevalence, the ecological effects of LSLADEs remain largely unexamined in empirical literature. To our knowledge, the only related study is by Ango (2018), which examined local livelihoods and forest conservation in southwestern Ethiopia, finding that such domestic investments reduced community access to forests and negatively affected forest conservation outcomes. Moreover, Ango's (2018) study did not explore the mechanisms or pathways through which LSLADEs affect biodiversity. In practice, the intensive land preparation methods commonly employed under LSLADE—such as large-scale clearing, mechanized plowing and chemical application—are likely to contribute to deforestation, soil degradation and habitat loss, thereby driving declines in plant and animal species (FoodPrint, 2024). In particular, the heavy use of agrochemicals, which is a widespread feature of many LSLA contexts (Behrman et al., 2014; Shi-le and Xin-ye, 2020), can pollute water bodies, disrupt soil microbial communities and harm non-target organisms (Behrman et al., 2014). Similarly, reliance on mechanized plows and bulldozers for land expansion can further alter ecosystems, fragment habitats and accelerate biodiversity loss (FoodPrint, 2024). Despite these plausible pathways, there is still no study that explicitly examines how specific land preparation practices—including agrochemical use, mechanized plowing and bulldozing—shape biodiversity outcomes. Addressing this gap is critical for informing policies that aim to balance agricultural transformation with ecological sustainability in Africa.

By way of filling the gaps raised, this study answered questions on the effects of LSLADE's land preparation methods on biodiversity, using spatial indicators like the Normalized Difference Vegetation Index (NDVI) and the Green Normalized Difference Vegetation Index (GNDVI) and Ghana as a case study. Specifically, it highlights LSLADE's investment in land preparation methods—including tractors, plows, other equipment and agrochemical inputs—and smallholder use of such methods as the primary mechanisms driving small farms' biodiversity. As mentioned previously, LSLADE may be initiated by private investors, community/traditional authorities, or government agencies. In the empirical analysis, however, LSLADE's investments in land preparation methods and smallholder use of such methods are operationalized as the community-level weighted share of the total value of tractors, plows, other equipment and agrochemical inputs and whether household use LSLADE-provided tractors, plows and agrochemical inputs during land preparation, respectively. These aggregations are regardless of ownership type, thereby capturing the overall pressure exerted by LSLADE activities on local ecosystems. Further, this aggregation captures the overall pressure exerted by LSLADE activities on local ecosystems but does not separately estimate effects by category.

Ghana serves as an ideal case study for examining these dynamics due to two main reasons. First, it has an extensive history of land acquisitions, established regulatory frameworks and a growing prevalence of LSLADE. Although regulations such as those by the Lands Commission (2016) are intended to manage land acquisitions, evidence suggests that LSLADE is on the rise (Jayne et al., 2016) and can have significant implications for ecosystems and biodiversity. Thus, the results could be relevant for informing stakeholders about the consequences of LSLADE. Second, Ghana's unique context, where local actors such as community elites, chiefs' families and non-farm sector investors play pivotal roles in land deals (Jayne et al., 2014, 2016), further underscores its relevance.

The rest of the paper includes literature review, conceptual framework LSLADE-biodiversity nexus, methods, results, discussions and conclusions with policy implications.

Large-Scale Land Acquisitions by Domestic Entities (LSLADEs) have become a defining feature of Ghana's agricultural transformation. Unlike transnational land deals that are often driven by foreign agribusinesses or state-to-state arrangements (Borras and Franco, 2012), LSLADEs are initiated by domestic actors—medium-scale farmers, local entrepreneurs, traditional authorities, and in some cases, state-backed schemes (Jayne et al., 2016, 2019). Compared to transnational LSLAs that frequently adopt capital-intensive and standardized land preparation methods, including large-scale bulldozer clearing, mechanized leveling and extensive use of herbicides and fertilizers, sometimes complemented by irrigation infrastructure, LSLADE's operations are embedded within local institutional structures and are influenced by customary land tenure arrangements and community-level negotiations. Consequently, LSLADEs often employ locally accessible methods of land preparation such as tractor ploughing, bulldozer clearing on a contract basis and herbicide application. The key distinctions between LSLADE and transnational land acquisitions—ranging from scale and financing to spillover effects and environmental oversight—are summarized in Table 1. These domestic investments reflect a growing class of commercial farmers who acquire significant tracts of land, usually through customary tenure systems, to expand production and integrate into local markets (Abdallah et al., 2024; Jayne et al., 2016).

In operational terms, LSLADEs tend to rely heavily on mechanization for land preparation. Survey evidence shows that tractors are the most widely used equipment, with about two-thirds of medium-scale farms employing them for plowing and land leveling (Chapoto et al., 2014). Implements such as disc plows and harrows are often attached to tractors, enabling faster and more uniform seedbed preparation compared to manual methods (Chapoto et al., 2014; FAO, 2018). Bulldozers, though less frequently used, play a crucial role in clearing heavily vegetated or uneven terrain, allowing investors to rapidly bring new or fallow lands into cultivation (Caliskan, 2013; Global Health NOW, 2025; Quansah et al., 1997). These mechanized practices set LSLADEs apart from many transnational acquisitions, where large-scale mechanization is often bundled with advanced technologies, irrigation schemes, or vertically integrated value chains that reduce direct dependence on local service markets (Abdallah, 2025; Ali et al., 2019).

Chemical methods also form a critical component of LSLADE land preparation strategies (Abdallah and Jayne, 2023; Abdallah, 2024). Herbicides are particularly prevalent, with over half of domestic medium-scale farms adopting them to manage weeds more efficiently and reduce labor requirements (Chapoto et al., 2014). The widespread use of agrochemicals underscores a shift towards input-intensive production systems. However, unlike transnational projects—which may have stronger regulatory oversight or access to international standards—LSLADEs often operate under looser enforcement environments (Ayamga and Laube, 2020). This raises concerns about inappropriate chemical application, runoff, and contamination of soil and water resources.

Institutionally, LSLADEs exhibit a strong reliance on local mechanization service providers. Since tractor ownership is limited, many domestic investors rent machinery, with roughly one-third of farms depending on external service markets (Chapoto et al., 2014). This reliance differentiates LSLADEs from transnational land deals, where investors typically internalize machinery ownership and management. While the outsourcing of tractor services by LSLADEs can stimulate rural service markets and generate local employment, it may also create competition for machinery between medium-scale investors and surrounding smallholders, potentially crowding out the latter from timely access to land preparation services.

The environmental and social implications of LSLADEs are far-reaching. Mechanized land clearing has been linked to deforestation, soil compaction and the destruction of natural habitats (Global Health NOW, 2025), while agrochemical use contributes to soil nutrient depletion, biodiversity loss and water pollution (Shi-le and Xin-ye, 2020; Sun et al., 2019). Unlike transnational land acquisitions, where environmental safeguards may be externally mandated by investors' home countries or international institutions, LSLADEs operate largely within domestic regulatory frameworks that often lack stringent enforcement (Ayamga and Laube, 2020). Consequently, the ecological costs of LSLADE practices can spill over into neighboring smallholder systems, affecting ecosystem services, reducing soil fertility and altering water availability.

Summarily, LSLADEs embody a hybrid model of agricultural transformation. They mirror transnational land acquisitions in their reliance on mechanization and chemical inputs but differ institutionally by being embedded within domestic economies and service markets. This embeddedness shapes not only their operational strategies but also the scale, pathways and intensity of their environmental spillovers, making it critical to examine LSLADEs as a distinct phenomenon rather than subsuming them under the broader literature on transnational land deals.

Remote sensing has become a cornerstone for biodiversity monitoring because spectral vegetation indices provide consistent, repeatable measures of vegetation amount, phenology and (indirectly) habitat heterogeneity. NDVI and GNDVI are widely used proxies for vegetation abundance and greenness and have been related to plant biomass and, in some contexts, to species richness or community composition (e.g., NDVI as a proxy for vegetation cover and GNDVI for chlorophyll/greenness) (Bonthoux et al., 2018; Ghaderpour et al., 2020; Kogan, 2019; Nieto et al., 2015). However, the strength and interpretation of NDVI–biodiversity relationships depend on ecological context, spatial scale and sensor characteristics. Recent reviews and empirical tests show that spectral proxies can track aspects of biodiversity but that the relationship is idiosyncratic: spectral diversity often correlates with species richness in structurally heterogeneous landscapes but performs less well where diversity is cryptic or at scales coarser than organismal patchiness (Pinon et al., 2024). To improve inference, recent studies (e.g., Anees et al., 2024; David et al., 2022; Gao et al., 2023; Jia et al., 2016; Mutanga et al., 2023) supplement NDVI/GNDVI with alternatives such as the Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI) and Normalized Difference Water Index (NDWI) to improve inference about vegetation condition, canopy structure and moisture—attributes that mediate biodiversity responses. EVI, for example, reduces saturation in high-biomass areas and can better follow temporal dynamics in dense canopies (Gao et al., 2023; Shi et al., 2017); SAVI controls soil background in sparse vegetation (Kareem et al., 2023; Ren et al., 2018); NDWI helps isolate moisture dynamics relevant to wetland or riparian habitat quality (Chu and Zhang, 2025; EOS Data Analytics, 2023). Empirical work (e.g., Gao et al., 2023; Shi et al., 2017) in African drylands and croplands demonstrates that combining indices (NDVI/EVI/SAVI/NDWI) and leveraging seasonal time-series often strengthens links to plant diversity and habitat condition relative to single-index approaches.

Spectral diversity and the spectral variation hypothesis (SVH) offer a complementary route: instead of a single index, spectral heterogeneity across bands or pixels is used as a proxy for species heterogeneity. Recent critical reviews and empirical tests indicate that SVH can work well in open vegetation and heterogeneous mosaics but is sensitive to spatial resolution, shadow/cloud removal and the spectral bands used; in short, SVH can be powerful but requires context-specific validation (e.g., calibrating spectral diversity against plot-level species data). These recent syntheses underscore why combining field biodiversity measures (richness, evenness and Shannon diversity) with remote indices—and being explicit about scale and sensor limits—is essential for robust inference.

LSLAs are often driven by global demand for agricultural land, government policies promoting foreign investment and the availability of arable land (Anseeuw et al., 2012; Jayne et al., 2016; Land Matrix Africa Regional Focal Point, 2020; Zoomers, 2010). While these investments can boost agricultural productivity and infrastructure (Deininger et al., 2011), they often result in the displacement of local communities, particularly smallholder farmers, leading to loss of livelihood, increased tensions over land use and interruption of their environment (Ahmed et al., 2018; Borras et al., 2022; Li, 2011). A key component of the smallholder environment whose interruption effect is direct on livelihood is smallholder farm biodiversity (Abdallah, 2024). Like all LSLAs, the establishment of LSLA by domestic entities (LSLADE) has been noted to strain local water resources, contribute to environmental degradation and reduce biodiversity of nearby small farms (Abdallah and Jayne, 2023). We highlight and present an in-depth discussion of the pathways through which LSLADE affects small farms' biodiversity drawing on existing theoretical and empirical literature.

Like LSLA by foreign entities, the discussion around the effect of LSLADE on small farms’ biodiversity follows two contrasting narratives: the development optimist perspective, which argues that LSLA can be effectively managed to provide shared benefits while reducing negative consequences and the pessimist perspective, which emphasizes the potential for significant harm to livelihoods, the environment and biodiversity (Boamah, 2014). These differing viewpoints reflect the larger debate between agroecology and industrial agriculture. Agroecology emphasizes ecological sustainability, the preservation of biodiversity and the resilience of local communities (Altieri and Nicholls, 2005), while industrial agriculture prioritizes large-scale production and mechanization, often at the cost of ecological and social systems (Collier, 2017). This study adopts a critical viewpoint, linking LSLADE to the critique of industrial agriculture due to its indirect yet substantial impacts on biodiversity within smallholder farming systems.

Figure 1 outlines the indirect ways in which LSLADE influences the biodiversity of nearby smallholder farms, emphasizing its effects on land use, agricultural practices and socio-economic factors. Initially, LSLADE—as shown in box 1 of Figure 1—contributes to a decline in biodiversity of smallholder farms by investing in land preparation methods, including the use of agrochemicals and machines—tractors, bulldozers and ploughs—which alters the surrounding ecological landscape (Abdallah, 2024). Like any other business, the growth of LSLADE largely depends on profit, which frequently requires investments in the clearing of forests using heavy machinery like bulldozers and plows, along with a significant use of agrochemicals to make the land productive as quickly as possible to maximize profit (Abdallah and Jayne, 2023; FoodPrint, 2018). Investments in such land preparation methods lead to habitat loss, fragmentation of ecosystems and the displacement of species that are vital for essential ecological functions, such as pollination and natural pest control (FoodPrint, 2024; Wineman et al., 2022). These disruptions may ultimately diminish the biodiversity—species richness, evenness, diversity and vegetation health—of neighbouring smallholder farms. Research, including studies by Carlson et al. (2012) and Davis et al. (2020), shows notable declines in biodiversity in areas adjacent to large-scale agricultural developments.

Second, LSLADE indirectly affects biodiversity through socio-economic pressures imposed on smallholder farmers. As land prices rise due to LSLADE, some smallholders are displaced, while others are forced to intensify their farming practices on smaller parcels to sustain productivity (Abdallah, 2021; Wineman et al., 2022). This often leads to a greater dependence on agrochemicals use supplied by LSLADE, which can harm soil quality, contaminate water sources and diminish biodiversity on these farms (Wineman et al., 2022). Chiarelli et al. (2022) and Oberlack et al. (2021) highlight how these economic challenges push farmers towards unsustainable practices, jeopardizing the ecological stability of smallholder systems.

Lastly, LSLADE can contribute to biodiversity loss of these farms through technological spillovers (Blomström and Persson, 1983; Dessy et al., 2012; Globerman, 1979). The renting or transfer of knowledge on the use of preparation methods—including tractors, ploughs and other machinery—by LSLADE to nearby smallholders facilitates agricultural expansion into previously uncultivated or forested areas. While this expansion may temporarily increase farm productivity, it fragments habitats and accelerates biodiversity declines in surrounding ecosystems. Wineman et al. (2022) and Magliocca et al. (2020) underscore the role of mechanization in exacerbating land-use change and ecological degradation.

This framework identifies LSLADE as a key driver of indirect biodiversity loss on nearby smallholder farms through its ecological, socio-economic and technological effects. By connecting these pathways to observable results—like decreases in species richness, evenness and diversity, along with lower vegetation indices (NDVI, GNDVI)—the study lays the groundwork for grasping the wider impacts of industrial agricultural growth. The findings may also highlight the necessity of integrating agroecological principles into land-use policies to reduce biodiversity loss, improve ecosystem services and bolster the resilience of smallholder farming systems.

The study is situated in northern Ghana, encompassing the Northern, Northeast and Savannah Regions, which together make up a substantial share of the country's landmass. With a land area of roughly 70,384 square kilometres, this zone is the largest in Ghana (MoFA, 2013, 2019) but remains sparsely populated, ranging between 18.8 and 87.1 persons per square kilometre and hosting a total population of 3,623,151 in 2021 (GSS, 2021). The region's ecological conditions are well-suited for diverse crops such as maize, rice and shea nuts, and relatively low land costs combined with expansive tracts of arable land enhance its attractiveness for mechanized and commercial agriculture. However, land governance is shaped by both traditional and statutory systems, which often intersect in complex ways (Lands Commission, 2016). Under customary tenure, local chiefs and family heads control access and rights to land, typically without formal documentation, making the system vulnerable to contestation and elite capture. At the same time, statutory management—overseen by the Lands Commission and related agencies—seeks to regulate land acquisition and use but frequently overlaps with customary authority, compounding administrative ambiguities. These dual governance structures, coupled with the availability of large tracts of under-documented land, create opportunities for large-scale land acquisitions by both domestic and foreign investors (Abdallah et al., 2023a, b, 2024; Ayamga et al., 2023), rendering the region a particularly relevant setting for the present study.

For the analysis, we relied on a two-round household panel survey comprising 400 agricultural households, conducted in 2019 and 2023. The sample was designed to capture communities most affected by LSLADE, ensuring that the data were representative of the study area. A multi-stage sampling procedure was employed. In the first stage, ten districts were purposively selected based on the predominance of LSLADE, drawing on documented evidence from the Northern Regional Lands Commission and the Land Matrix database (Land Matrix, 2021). Selection criteria included the extent of land area acquired, the number of transactions and the breadth of affected communities. These districts were prioritized because of their high concentration of LSLADE activity and heterogeneity in exposure levels. In the second stage, between two and six communities were randomly chosen from each district in proportion to the number of affected communities, yielding a total of 20 communities. In the third stage, 20 households were randomly drawn from a listing of 4,124 affected households across the 10 districts, producing a total sample of 400 households (see Table 2). The listing exercise was conducted in collaboration with local headmen to ensure accuracy.

The first survey round was carried out during the 2019 cropping season using a semi-structured questionnaire programmed in Kobo Toolbox ^[1]. The instrument captured household demographics, LSLADE-related information, land preparation methods, biodiversity indicators and management practices. A second round of data collection was undertaken in 2023, during which the same households were revisited. Tracking was facilitated by contact details, household identifiers and GPS coordinates recorded during the baseline. In addition, local leaders assisted in locating relocated households and repeat visits were made to ensure high coverage. As a result, 398 households were successfully re-interviewed, yielding 796 observations and a re-interview rate of 99.5%. Attrition was minimal, limited to two households that had migrated, and attrition analysis revealed no significant baseline differences between households retained and those lost, confirming the representativeness of the panel.

The panel structure allows for longitudinal analysis of LSLADE's effects, with the 2019–2023 interval chosen to capture medium-term dynamics while addressing unobserved heterogeneity. Data collected in Kobo Toolbox were cleaned and processed in Stata 18, where key outcome variables, including biodiversity measures such as species richness, evenness and diversity, were constructed. To enrich the analysis, we complemented the survey data with the Ghana Living Standards Survey Round 7 (GLSS 7), conducted in 2016–2017 by the Ghana Statistical Service. The GLSS 7 used a two-stage design, selecting enumeration areas (EAs) from the 2010 Population and Housing Census with probability proportional to size, followed by systematic household sampling within each EA. This dataset provided additional information on agricultural inputs, land preparation and household conditions, which was harmonized with the panel data at the community level.

The land preparation methods of LSLADEs are captured using two key variables. The first variable “LSLADE_investment” is LSLADE investment in land preparation methods measured as the community-level weighted share of the total value of tractors, plows, other equipment and agrochemical inputs under LSLADE, while the second variable “LSLADE_method_use” captures farmer use of LSLADE-provided tractors, plows and agrochemicals during land preparation. The LSLADE investment in land preparation methods quantifies the proportion of financial resources allocated by LSLADE to these mechanized equipment and chemical inputs in each community. This variable reflects the intensity of LSLADE's land preparation efforts and its potential influence on biodiversity through mechanization and agrochemical use. Thus, higher investment levels may suggest more extensive land clearing and mechanization, which may correlate with greater impacts on biodiversity (e.g., reduction in species richness, evenness, diversity and vegetation). To construct LSLADE_investment variable, we followed three steps. First, we used the Ghana Living Standards Survey Round 7 (GLSS7, 2016–2017) to compute the total value of mechanized land preparation equipment (tractors, plows, harrows, bulldozers, etc.) and agrochemicals (herbicides, pesticides and fertilizers used specifically for land preparation). For each household in the GLSS7 sample, we summed the reported expenditures on these categories to generate household-level investment values. Second, to identify which part of these investments could be reasonably attributed to LSLADE-type farms, we classified farms into small-scale (0–5 ha), medium-scale (5–50 ha) and large-scale (>50 ha) using the thresholds defined by Abdallah and Jayne (2023). Because LSLADEs are most consistent with the medium- and large-scale categories, we aggregated household-level investments within each farm-size category and computed their shares at the community level. In other words, the community-level LSLADE investment share reflects the weighted proportion of mechanized and chemical input value accounted for by medium- and large-scale farms relative to the total community investment in land preparation inputs. Third, since GLSS7 data lack georeferenced coordinates at the household level, we aligned them to our survey sample by aggregating at the community level. This was done by matching GLSS7 enumeration areas to the corresponding communities in our two-round panel dataset, using district identifiers, community names and administrative boundaries as linking keys. Where direct matches were ambiguous, we verified consistency using district-level agricultural profiles from the Ministry of Food and Agriculture. It is important to note that while our household panel data were collected in 2019 and 2023, the GLSS7 data were collected in 2016–2017, which raises potential concerns regarding temporal comparability. Thus, to ensure temporal comparability, we treated the GLSS7-based LSLADE investment variable as a baseline proxy for structural differences in community-level investment intensity. Given the absence of more recent nationally representative data with similar farm-size breakdowns, we assumed that relative patterns in investment shares (e.g., the dominance of medium-scale farms in mechanization access) remained stable between 2016 and 2023. To test this assumption, we conducted a sensitivity check by adjusting the GLSS7 values for inflation and interpolating investment growth rates using FAO mechanization trend reports and Ministry of Agriculture input-use surveys (2017–2022). Specifically, we experimented with two alternative specifications: (1) weighting LSLADE investments by the share of cultivated land area instead of value and (2) computing the LSLADE share using only the medium-scale category (5–50 ha) and only large-scale category (>50 ha) as opposed to medium + large farms combined. The results of the estimates are available on request but Appendix (Table A1) shows a summary of sensitivity analysis. Across these alternative weighting schemes, the direction and significance of the estimated biodiversity impacts remained consistent, though the effect sizes varied slightly. This procedure ensures that the LSLADE_investment variable credibly reflects the intensity of mechanized and chemical land preparation inputs attributable to domestic large-scale investors, while acknowledging and testing sensitivity to timing and weighting assumptions. The second variable, farmer use of LSLADE-provided resources, indicates whether farmers in a community utilized LSLADE equipment and chemical inputs for their land preparation activities. This variable captures the extent of LSLADE's interaction with local farming practices and the diffusion of its methods within the community. Together, these variables provide complementary insights into the scale and reach of LSLADE's land preparation methods, enabling an analysis of their direct and indirect impacts on biodiversity. It is, however, important to note that this approach of aggregating LSLADE investments at the community level and whether households use any of the land preparation methods can have limitations. This is because LSLADE actors in the sampled districts can vary across private domestic investors, communal/traditional authority-led acquisitions and government-backed projects. This heterogeneity can influence biodiversity impacts through differences in resource access, operational scale and land management practices. For example, government-backed schemes may intensify biodiversity pressures through mechanization subsidies, while some traditional authority-led initiatives might retain patches of vegetation for cultural or ecological reasons. These differences are not explicitly separated in the statistical models but can be relevant when interpreting results.

Biodiversity is a multidimensional concept (Gotelli and Colwell, 2001), consisting of species richness, diversity and evenness. These indicators are often constructed from survey data based on literature (e.g., Bozzola and Smale, 2020). However, because of the susceptibility of recall data to misreporting errors (Abay et al., 2022), we employed spatial indicators of vegetation, including the Normalized Difference Vegetation Index (NDVI) and Green Normalized Difference Vegetation Index (GNDVI) for estimations. The use of these vegetation indices for measuring biodiversity is based on the idea that alterations in vegetation cover and structure may reflect shifts in biodiversity. The NDVI and GNDVI were both constructed from Landsat 8 images from NASA/USGS. Specifically, bands 3, 4 and 5 were employed for the construction of NDVI and GNDVI. These bands were downloaded, mosaic and or clipped to our area shapefiles using ArcGIS 10.8.2. NDVI and GNDVI were then calculated with a raster calculator in ArcGIS 10.8.2 using the expressions: $NDVI=(NIR−Red)(NIR+Red)$ and $GNDVI=(Green−Red)(Gree+Red)$⁠. Where NIR represents the reflectance value of near-infrared light, Green and Red represent the reflectance values of green and red spectral bands (Avdan and Jovanovska, 2016). The cell values of a raster were finally extracted to the geographic information systems (GIS) point location of the sampled households. Both NDVI and GNDVI typically ranged from −1 to +1. Higher positive values indicate dense and healthy vegetation, while a lower or negative value indicates less vegetation or non-vegetated areas. Values higher than zero generally indicate the presence of vegetation, while values near 0 indicate little or no vegetation. These indicators correlate strongly with biodiversity (e.g., McFarland et al., 2013; Nieto et al., 2015). The GNDVI is like NDVI but specifically focused on the green spectral band and therefore useful when analyzing vegetation that might have distinct spectral characteristics in the green band. It is therefore employed to capture distinct features of vegetation in northern Ghana. Moreover, some vegetation types, such as grasslands or crops with a high chlorophyll concentration, exhibit stronger reflectance in the green band compared to the red band. By focusing on the green band, GNDVI can provide additional information about vegetation dynamics and health, especially in areas where the green band is a more reliable indicator of vegetation vigour.

We also captured household/plot-level covariates and institutional variables from the field survey and village/community-level variables or spatial controls using GIS point location of the sampled households. The village/community-level variables or spatial controls include population density and built-up area data with a spatial resolution of 250 -meter grids from the sixth version of Global Rural-Urban Mapping (GRUMPv1); precipitation and temperature data of 2.5 min resolution from WorldClim; elevation data from the Shuttle Radar Topography Mission and vegetation indices with a spatial resolution of 30 meters from Landsat 8 of NASA/USGS6. The population density was generated using estimates of population from GRUMP with the estimates of arable land from Global Agro-Ecological Zones v4 (Fischer et al., 2021). The details of these variables are presented in Table 3.

As the conceptual framework shows, biodiversity indicators are determined by $LSLADE$ through its land preparation methods (i.e., LSLADE's weighted share of investment in land preparation methods in a community and whether farmers used LSLADE-provided tractors, plows and agrochemicals during land preparation). Thus, our main empirical specification for testing this proposition is generally expressed as:

where $Yit$ represents the outcome variables (e.g., biodiversity indicators) for household's farm $i$ at time $t$⁠. Further, $Xi$ is a vector of household/plot- and community-level variables or spatial controls farm $i$ at time $t$⁠; parameter $β1$ is the coefficient; and $ηit$ is time-invariant unobserved heterogeneity in farm $i$ at time $t$⁠. However, there are important challenges in estimating the effect of some of the indicators of LSLADE on the biodiversity indicators. For the LSLADE's investments in land preparation methods in a community, we assumed exogeneity in the sense that LSLA investment in a community is beyond farmers' control and not correlated with household/plot-level time-varying unobservable factors. Nonetheless, unobserved heterogeneities—including soil quality, climate or agroecological conditions—affecting biodiversity may also correlate with LSLADE's investments. To consistently identify LSLADE's investments effect $β2$​, we explore the panel nature of the data. We initially employed the random effects (RE) model based on model diagnostics, including the ad Breusch-Pagan Lagrange Multiplier (LM) test and the Hausman test. However, while the random effect model allows inclusion of key time-invariant spatial controls, such as population density and climatic variables (e.g., averages of precipitation and minimum and maximum temperatures), which are significant predictors of biodiversity outcomes, it assumes that unobserved household-level factors are uncorrelated with the independent variables—a potentially strong assumption that may lead to biased estimates if such an assumption does not hold. On the other hand, these time-invariant variables would be excluded in the fixed effects (FE) model, which focuses solely on within-household variation over time but may provide more accurate estimates of the impact of LSLADE's investment on biodiversity by controlling for all time-invariant household-specific factors that could bias the estimates. Consequently, we were confronted with a dilemma: either focus on the results of the RE that allow inclusion of key time-invariant spatial controls but risk accuracy of results or use FE that best identifies LSLADE's investment effect but excludes key time-invariant spatial controls. Since our main interest is in the effect of LSLADE's investment, which would be much better identified with the FE model, we employed the FE model in the estimations but also present the RE model as the basis for comparison. To complement the FE model, we also employed the Mundlak-adjusted RE model (Mundlak, 1978), which retains the time-invariant variables while addressing potential bias from unobserved household-specific factors. The Mundlak-adjusted RE does this by including the means of time-varying covariates (e.g., household averages of precipitation and temperature), thereby controlling correlations between time-invariant effects and independent variables. This allows us to estimate the impact of LSLADE's investment while retaining critical spatial and environmental controls, which would otherwise be dropped in the FE model.

Regarding analysis of the effect of farmers’ use of LSLADE's tractors, plows and agrochemicals on biodiversity, we also use the RE, FE and Mundlak-adjusted RE models to allow inclusion of population density and climatic factors and to deal with unobserved heterogeneities. We also combined the Mundlak device (Mundlak, 1978) with the control function approach (CFA) (i.e., Mundlak-adjusted CFA models) to control for potential endogeneity arising from the fact that unobservable factors that affect our outcomes (i.e., biodiversity indicators) may also influence households' decision to use land preparation methods of LSLADE. For example, while land preparation methods from investors may influence biodiversity, they may also be high in uncultivated areas where the establishment of new farms requires the clearing of unwanted vegetation. Such reverse causality may also lead to the endogeneity of the choice of land preparation methods from LSLADE.

The Mundlak-adjusted CFA proceeds in two stages. In the first stage, generalized residuals are generated from the first-stage model with pooled probit for use of land preparation methods. Here, the household/plot-level covariates, institutional variables from the field survey, and village/community-level variables or spatial factors and mean of all time-varying variables are included in the probit model. In the second stage of the Mundlak-adjusted CFA, the residuals, household/plot-level covariates, institutional variables from the field survey, village/community-level variables or spatial factors, and mean of all time-varying variables are included in a pooled OLS (POLS) estimation of biodiversity. Bootstrapping with 100 replications was applied to each of the second-stage estimations to obtain accurate results. The inclusion of the generalized residuals breaks the endogeneity link between land preparation methods and the error terms in the outcome equations (Wooldridge, 2015). The first- and second-stage equations are expressed as:

where $LGDEi*$ is the latent indicator for whether $ith$ households use land preparation methods from LSLADE; $Yit$ is the index of the outcomes of interest including biodiversity indicators of households' farm $i$ at time $t$⁠; $Xi$ indicate household/plot-level, institutional and village/community-level variables or spatial controls; $X̅$ is their means to control time-varying unobserved heterogeneities; $Rˆi$ is the generalized residuals from the first stage pooled probit estimates; $βi$⁠, $αi$⁠, $δi$⁠, $ρi$⁠, $λi$⁠, $ωi$ and $ϑi$ are the respective coefficients; $ηi$ and $εi$ are the normally distributed error term. Further $ρi$ can be taken as the impact of use of land preparation method from LSLADE on each of the biodiversity indicators while $λ$ a test for exogeneity of land preparation method from LSLADE in equation (3). If $ρi$ is significant, exogeneity is rejected and thus implies the inclusion of the residuals corrects for the endogeneity. Otherwise, the inclusion of the residuals is not necessary. For proper identification of equation (3), the variables in the first-stage probit or logit estimations need to contain at least one instrument. In this study, $Ii$ the instrumental variable includes the proportion of farmer's peers in the community who interact with domestic investors. The proportion of farmers' peers in the community who interact with domestic investors is defined as n/(m_1), where n represents the number of farmer's peers using land preparation methods from LSLADE in a sampled community, and m represents the total number of peers surveyed in that community. The number of other farmers residing in the same community using land preparation methods from LSLADE would very likely inform the decision of a particular farmer to also use such methods. We argued that there is no reason to believe that this variable would directly influence households' outcomes or correlate with error terms in the outcome equations, except through this instrumental variable, thus satisfying the exogeneity assumption. Through a simple falsification test, we confirmed that the selected instrument is valid.

However, even after applying RE, FE, Mundlak-adjusted RE and CRE-CFA, we still have concerns about potential misreporting. In Ghana, government and other wealthy private investors also render tractor, plows and bulldozer services, or sell agrochemicals to farmers during land preparation. For instance, government subsidies on chemical fertilizer can blur the lines between public and private interventions. Thus, farmers may not clearly distinguish between services provided by LSLADE, government programs or those of the wealthy private investors, leading to overreporting or underreporting of LSLADE method use. Moreover, if LSLADE practices are associated with benefits or recognition, farmers may falsely claim usage to appear compliant. Such potential misreporting can yield biased results even with the application of standard techniques like RE, FE, Mundlak-adjusted RE and CRE-CFA. To account for this issue, we adopt the partial identification and bounding framework proposed by Lin et al. (2024). This approach provides a robust estimation method that accommodates uncertainty in treatment classification by allowing for misclassification of the treatment variable. Unlike RE, FE, Mundlak-adjusted RE and CRE-CFA which assumed perfect reporting accuracy, leading to biased and inconsistent estimates when this assumption is violated, the approach by Lin et al. (2024) instead introduces bounds on the extent of misclassification to credible estimation of the effect of farmers' use of LSLADE's land preparation methods even when misreporting occurs. Under this framework, we specify plausible bounds for the rates of false positives (farmers who incorrectly report using LSLADE methods) and false negatives (farmers who fail to report actual use). These bounds were informed by auxiliary data, validation surveys, or external administrative records when available. By incorporating these bounds into the estimation process, we derive partially identified intervals for the treatment effect, which remain valid under a range of misreporting scenarios. Additionally, we enhanced our analysis with an instrumental variable (IV) approach to address potential endogeneity in treatment assignment. An appropriate IV must be a binary variable predicting farmers' use of LSLADE methods but remain uncorrelated with the unobserved factors influencing biodiversity outcomes, except through its effect on the treatment. In our case, we employed whether the farmer has peers in the village interacting with LSLADE as instruments. This combination of partial identification and IV estimation further strengthens the robustness of the causal inference by addressing both misreporting and endogeneity concerns simultaneously. By using this methodology, we mitigate the biases introduced by misreporting while maintaining the validity of the estimated treatment effects. This also enhances the reliability of our findings and ensures that the conclusions drawn about the relationship between LSLADE methods and biodiversity are both valid and defensible. By addressing misreporting explicitly, we contribute to the growing literature on the environmental impacts of agricultural intensification while providing actionable insights for sustainable agricultural policy.

To contextualize the findings or provide readers with a snapshot of the data, we present in Table 4, the descriptive statistics of the variable used in the sample and over the two survey periods. The statistics reveal notable trends in key variables across the study period (2019–2023). Specifically, NDVI and GNDVI showed a slight decrease from 2019 to 2023. The mean NDVI increased from 0.18 (SD: 0.05) in 2019 to 0.14 (SD: 0.05) in 2023, and the mean GNDVI decreased from 0.29 (SD: 0.06) to 0.16 (SD: 0.06). These changes suggest a deterioration in vegetation health and biomass density over the period. The community-level share of LSLADE investment in mechanized equipment and agrochemicals increased from a mean of 40.86 (SD: 29.71) in 2019 to 58.95 (SD: 23.59) in 2023, reflecting greater intensity of LSLADE operations. The use of LSLADE-provided methods for land preparation also increased from 0.33 (SD: 0.17) in 2019 to 0.61 (SD: 0.19) in 2023, showing broader adoption of LSLADE practices among farmers. Household characteristics showed slight changes, with a decline in the age of household heads, education levels and asset ownership, suggesting potential economic challenges. Leadership roles within traditional and government structures increased, reflecting evolving social dynamics within the communities. Meanwhile, environmental risks intensified, with higher incidences of floods and droughts reported in 2023 compared to 2019.

Table 5 further presents the biodiversity outcomes (NDVI and GNDVI) for three LSLADE investment quintiles in two survey years (2019 and 2023). NDVI values across all three quintiles (ranging from 0.49 to 0.50) are relatively stable over the two survey years (2019 and 2023). This stability suggests that the vegetation cover in the study areas has not experienced significant changes during the period in question, despite varying levels of LSLADE investment in land preparation. On the other hand, GNDVI values show some variation compared to NDVI. While Quintile 1 remains stable at 0.47, Quintile 2 showed a slight decline from 0.45 in 2019 to 0.44 in 2023. Quintile 3, however, experienced a noticeable improvement, with GNDVI rising from 0.40 in 2019 to 0.43 in 2023. This suggests that in some areas with higher LSLADE investment in land preparation (as represented by the higher quintiles), the land may have experienced improvements in vegetation cover, while others may have seen slight declines. However, this observed pattern of increase in GNDVI should be interpreted with caution. First, these are unadjusted descriptive values that do not control confounding factors such as climatic variability, differences in crop type, or concurrent land management practices.

Figure 2 further presents the distribution of the NDVI and the GNDVI across quintiles (i.e., Q1-Q5) of LSLADE investment. As mentioned previously, both indices provide remote-sensing measures of vegetation health and ecological quality, with higher values indicating denser, greener and healthier vegetation cover. The results reveal that households located in areas with lower levels of LSLADE investment (i.e., Q1–Q2) exhibit relatively higher NDVI and GNDVI values, suggesting healthier vegetation and more stable biodiversity conditions. In contrast, higher LSLADE quintiles (Q4–Q5) are associated with a marked decline in both NDVI and GNDVI, consistent with substantial ecological degradation in communities exposed to intensive land preparation activities of LSLADE. Notably, the third quintile (Q3) shows a modest increase in GNDVI, while NDVI remains stable, indicating localized improvements in vegetation greenness at moderate levels of LSLADE activity. The modest increase in GNDVI observed at the mid-level quintile (Q3) may reflect short-term greening effects linked to fertilizer application, irrigation, or selective clearing of degraded plots, which can temporarily enhance chlorophyll concentration and vegetation productivity (Herrmann et al., 2005).

It is also important to mention that in some communities within the highest LSLADE investment category, land preparation was followed by the establishment of perennial or long-cycle crops (e.g., mango, cashew) and irrigated fields, which can increase vegetation greenness in satellite indices even as overall biodiversity declines. Thus, a localized increase in GNDVI does not necessarily indicate a net biodiversity gain and may reflect short-term vegetation recovery or changes in vegetation composition rather than improved ecosystem health. It is, therefore, important to note that while these descriptive analyses provide valuable insights into the raw relationships between LSLADE's land preparation methods and the outcome variables, they do not account for potential endogeneity or unobserved confounders. As such, the relationships depicted should be interpreted as descriptive rather than causal. The next section presents multivariate regression analyses that rigorously control confounding factors to estimate the causal effects of land preparation methods on biodiversity.

This section presents the effect of the share of LSLADE's investment in land preparation methods and farmers' use of LSLADE land preparation methods on biodiversity, as measured by NDVI and GNDVI. Rather than presenting coefficient estimates, which are detailed in the appendix (Tables A2–A3), the results here focused on the elasticity estimates for the effect of LSLADE's investment share in land preparation methods and the average partial effects (APEs) of farmers' use of LSLADE land preparation methods. The elasticity estimates capture the percentage change in biodiversity outcomes (i.e., NDVI and GNDVI) resulting from a 1% change in LSLADE's share of investment, while the APEs reflect the average change in these outcomes due to farmers' adoption of LSLADE land preparation methods. These estimates were generated from the coefficient estimates of the RE, FE and Mundlak-adjusted RE models. The results from RE, FE and Mundlak-adjusted RE models align closely, indicating that the effects of LSLADE investments and methods are robust to different model specifications. The Mundlak-adjusted RE models account for unobserved heterogeneity by including household-level means of covariates, and their results are comparable to the FE models, suggesting that omitted variable bias is unlikely to drive the findings. The Mundlak-adjusted CFA models further validate these effects by controlling for potential endogeneity in farmers’ adoption of LSLADE methods. Across all models, the results consistently show that LSLADE investments and methods negatively influence biodiversity.

Table 6 presents the elasticity estimates for the effect of LSLADE's investment share in land preparation methods on biodiversity indicators. For NDVI, the elasticity estimates are negative and statistically significant across all models. Specifically, a 1% increase in LSLADE investment results in a 0.06%–0.07% decrease in NDVI, depending on the estimation model. These results are consistent across the Random Effect (RE), Fixed Effect (FE) and Mundlak-adjusted RE models, underscoring the robustness of the findings. For GNDVI, the elasticity estimates are higher, with a 1% increase in LSLADE investment leading to a 0.10%–0.11% decrease in GNDVI. The results are statistically significant at the 1% level across all models, suggesting a stronger effect of LSLADE investment on vegetation greenness compared to overall vegetation cover (NDVI). These findings highlight that LSLADE investments negatively contribute to enhancing small farms' biodiversity, with more pronounced effects on vegetation greenness (GNDVI). This implies that such investments may promote healthier vegetation, likely due to improved soil preparation and agronomic practices introduced by LSLADE.

Table 7 reports the average partial effects (APEs) of farmers' use of LSLADE land preparation methods on NDVI and GNDVI. For NDVI, the APEs indicate that adopting LSLADE land preparation methods decreases NDVI by 0.01–0.02 percentage points, depending on the estimation model. The effects are statistically significant at the 5% level in most models, reinforcing the positive contribution of LSLADE methods to overall vegetation cover. For GNDVI, the APEs are consistently higher, with the LSLADE method use decreasing GNDVI by 0.02–0.03 percentage points. These effects are statistically significant at the 1% or 5% levels across all models. The stronger effects on GNDVI align with the elasticity results, suggesting that LSLADE methods have a particularly negative impact on vegetation health and greenness.

Figure 3 presents the estimated coefficients and 95% confidence intervals of the effect of farmers' use of LSLADE-provided land preparation methods (tractors, bulldozers and agrochemicals) on vegetation-based biodiversity proxies, namely the NDVI and the GNDVI. The negative coefficients consistently reported across specifications suggest that LSLADE method use is associated with declines in vegetation health and greenness. The CIs are tight in RE and FE models but widen slightly in the CFA specification. This suggests the estimates are reasonably precise, though accounting for endogeneity introduces some uncertainty. But importantly, most confidence intervals do not cross zero, underscoring the robustness of the negative effect.

The findings of this study contribute to the growing literature on the environmental consequences of large-scale agricultural investments, revealing a consistent negative impact of LSLADE investments and methods on biodiversity. Specifically, the results showed that LSLADE's investments in land preparation methods decrease biodiversity indicators, including NDVI and GNDVI. The subsequent analysis of farmers' use of such land preparation methods and its effect on such biodiversity indicators reinforced these findings, with the APEs for NDVI and GNDVI revealing a substantial decline in vegetation health and density due to the use of bulldozers, tractors, plows and agrochemicals. This pattern aligns with the hypothesis that intensive land preparation methods, often associated with mechanization, monoculture, or chemical use, may degrade the quality of vegetation, disrupt soil health and reduce ecological diversity (Behrman et al., 2014; FAO, 1998). Land preparation methods, while efficient for clearing large tracts of land for agricultural production, contribute to the loss of biodiversity by disrupting native ecosystems, destroying habitats and reducing the availability of natural resources that local communities depend on. These results are robust across multiple models, indicating that both the share of LSLADE's investment in land preparation methods and farmers' adoption of these methods reduce biodiversity outcomes, as measured by NDVI and GNDVI. This study extends the debate on agricultural intensification by providing evidence of its ecological costs and raises critical questions about the sustainability of such investments.

This study's results align with previous research documenting the environmental risks associated with intensive agricultural practices. Meyfroidt et al. (2014) emphasize that large-scale investments often prioritize immediate productivity gains, leading to biodiversity losses through habitat conversion, soil degradation and the homogenization of ecosystems. Similarly, Branthomme et al. (2023) highlights that biodiversity loss is a frequent outcome of agricultural investments, particularly in regions where ecosystems are already vulnerable. The consistent negative effects observed in this study underscore these broader trends, suggesting that LSLADE investments may exacerbate environmental degradation in smallholder farming systems. Contrasting with studies that emphasize the potential benefits of large-scale investments, such as Deininger et al. (2011) and von Braun and Meinzen-Dick (2009), the findings here reveal a trade-off between productivity and ecological sustainability. While LSLADE interventions may provide short-term gains in agricultural output, their adverse effects on vegetation health and greenness, as captured by the stronger impacts on GNDVI, indicate potential long-term consequences for ecosystem services. This study also contributes methodologically to the literature by employing elasticity estimates and APEs to quantify the impact of LSLADE investments and methods. By summarizing complex relationships in a clear and interpretable manner, this approach facilitates comparisons with prior research and provides actionable insights for policymakers and practitioners. The robust results across model specifications, including RE, FE and Mundlak-adjusted RE models, further validate the observed relationships and suggest that the negative effects are not artifacts of unobserved heterogeneity or endogeneity.

The findings have important implications for policy and practice. First, they highlight the need to integrate biodiversity-friendly practices into LSLADE interventions to mitigate their ecological costs. Agroecological approaches, such as reduced tillage, crop diversification and organic soil management, could help balance productivity with environmental conservation. Second, the results underscore the importance of incorporating biodiversity monitoring into agricultural investment projects. Using indicators like NDVI and GNDVI can provide early warnings of ecological degradation and guide adaptive management strategies. In the broader context of sustainable development, these findings challenge the assumption that agricultural intensification is an unqualified pathway to rural development. Instead, they point to the need for a more nuanced approach that recognizes and addresses the trade-offs between agricultural productivity and ecological sustainability. Future research should focus on identifying specific practices within LSLADE interventions that contribute to biodiversity loss and exploring alternative strategies that enhance both productivity and environmental resilience.

This study demonstrates that LSLADE investments and methods have significant negative impacts on biodiversity in smallholder farming systems, as evidenced by reductions in NDVI and GNDVI. The findings, consistent across various econometric models, reveal that both the share of LSLADE investment in land preparation methods and farmers' adoption of these methods lead to declines in biodiversity. The results highlight the ecological trade-offs associated with large-scale agricultural intensification, particularly its adverse effects on vegetation health and ecosystem resilience. The robust negative relationship between LSLADE practices and biodiversity underscores the need to re-evaluate the design and implementation of such investments. While they may offer short-term gains in agricultural productivity, their long-term ecological costs raise critical concerns about sustainability. This study emphasizes that achieving balanced development in agricultural systems requires addressing these trade-offs to ensure both productivity and environmental health.

To mitigate the ecological risks associated with LSLADE practices, it is recommended that agricultural policies and investments prioritize the integration of biodiversity-friendly practices. Sustainable land preparation methods, such as agroecological approaches, reduced tillage and diversified cropping systems, should be encouraged to minimize biodiversity loss while maintaining productivity. For example, instead of complete bulldozing, strip clearing that alternates bands of cultivated and uncultivated land can preserve habitat corridors while still allowing agricultural expansion. Reduced tillage methods, including minimum and zero tillage, help conserve soil organic matter, reduce erosion and sustain soil biodiversity. Establishing vegetative buffers, such as grasses or trees along rivers, wetlands, and field boundaries, can minimize agrochemical runoff while providing microhabitats for species. Precision herbicide application through calibrated sprayers or drone-assisted techniques reduces excessive chemical use, prevents drift and protects non-target organisms. Diversified cropping systems, including intercropping and the integration of legumes or cover crops, also serve as biodiversity-friendly strategies by reducing monoculture pressures, improving soil fertility, and sustaining bird and insect populations.

It is also essential to move beyond general calls for biodiversity-friendly management and provide practical specificity in terms of monitoring protocols and feasible land preparation practices. Additionally, monitoring biodiversity indicators, such as NDVI and GNDVI, should become an integral part of agricultural investment frameworks to provide early warnings of ecological degradation and guide adaptive management. Further, a shift toward more sustainable LSLADE practices is essential for safeguarding biodiversity and ensuring the long-term viability of smallholder farming systems. The monitoring can be conducted at least once per agricultural season, both before and after land preparation, to capture immediate ecological changes, and annually to track longer-term trends. Key indices should include NDVI and GNDVI as primary measures of vegetation health, complemented by Enhanced Vegetation Index (EVI) to address canopy saturation and Soil-Adjusted Vegetation Index (SAVI) to reduce soil brightness effects. Monitoring may also be guided by clear thresholds, such that a sustained 5–10% decline in NDVI or GNDVI relative to a community's five-year baseline would trigger ecological alerts requiring corrective measures such as buffer restoration or reduced mechanization intensity. Institutional responsibilities are central to effective monitoring. The Lands Commission should integrate biodiversity monitoring into the approval of land allocations, while the Environmental Protection Agency (EPA) should oversee compliance with ecological thresholds. District-level agricultural extension services should work with farmer-based organizations to support local monitoring using handheld GPS-enabled devices and open-access satellite platforms such as Google Earth Engine.

By adopting approaches that balance agricultural productivity with ecological preservation, LSLADE investments can contribute to a more sustainable and resilient agricultural future.

The supplementary material for this article can be found online.