Associating dietary quality and forest cover in India

With the global population expected to reach almost 10 billion by 2050, there is growing concern for how increasing demands for high-quality, sustainable diets will be met. Whilst food production and environmental conservation are often viewed as competing land uses, a growing body of literature supports the importance of forests in providing food and nutritional security. This study used data from India ’ s National Family Health Survey (NFHS-5) on food consumption and nutritional status in infants and young children aged 6 – 59 months and tree cover data to examine the statistical associations between tree coverage and indicators of nutritional health in India – a country home to > 200 million of the world ’ s undernourished, a rising double burden of malnutrition, and with vast forest and tree cover. In analyses conducted, tree cover was positively associated with higher dietary diversity in infants and young children aged 6 – 23 months, and with reduced likelihood of stunting and wasting in children aged under 5 yrs. in areas with the highest tree coverages, after controlling for other possible variables. These relationships do not always hold at regional level; further work is needed to explore the mechanisms underlying the relationships examined within given contexts. Overall, these findings indicate the potential for forests and trees to contribute to quality diets in India. In emphasising the potential complementarity between environmental conservation and food provisioning, this work has important implications for food security and environmental policy.


Introduction
Despite huge increases in food and nutrient production in recent decades, conventional agricultural strategies have fallen short of eliminating global hunger and other forms of malnutrition (Geyik et al., 2020).Dominant strategies to tackle malnutrition have prioritised the production of dietary staples within specialised agricultural landscapes, seeking to increase yields and calorific production (and profits) via agricultural intensification and expansion (Pingali, 2015).Whilst such strategies have arguably attenuated undernourishment (insufficient calories), the focus on staples has often come at the expense of dietary quality and dietary diversity (Ickowitz et al., 2022;Khoury et al., 2014;Pinstrup-Andersen, 2012).Simultaneously, strong evidence also supports that agricultural intensification is amongst the leading causes of biodiversity loss and declining ecosystem services (IPBES, 2019).With global population set to reach almost 10 billion by 2050, there is urgent need to re-evaluate how we produce quality food if we are to meet goals to reduce all forms of malnutrition and meet environmental and climate commitments (Clark et al., 2020;Gergel et al., 2020;Springmann et al., 2018).
In this context, a relatively new field of research has examined how forests and trees can support healthy and sustainable diets (Skov Olesen et al., 2022).Whilst dominant paradigms around agriculture and environmental conservation have tended to view tensions between food production and sustainability goals as insurmountable and mutually exclusive, recent scholarship and policy discussion has highlighted the opportunities that exist in deriving an integrated agenda for sustainable diets and food systems transformations (Swinburn et al., 2019;Willett et al., 2019).Retaining and promoting the maintenance of forests within agricultural landscapes has become a particular point of interest in policy debates around food and sustainable diets, with growing evidence that forests and uncultivated systems can play an important complement to agricultural production, supporting the provision of sustainable and nutritious diets to forest-proximate, rural, and Indigenous Peoples' communities (Powell et al., 2022.;IPBES, 2022.;Vinceti et al., 2013;Vira et al., 2015).
Forests account for almost a third of global land surface area (4.06 billion hectares; FAO, 2022).Whilst precise estimates are lacking, it is estimated that around 4.17 billion people live within 5 km of a forest (termed forest-proximate people, FPP 1 ; FAO, 2022; Newton et al., 2020) and that at least 3.5 billion people in the world make use of or derive income from non-timber forest products (NTFPs) as important food, health, and livelihood resources (Shackleton and de Vos, 2022).Proximity to forests can support nutrition through a number of direct and indirect pathways.Directly, forest and tree-based systems remain a direct and significant source of "wild" foods for millions of people worldwide (FAO and Alliance of Bioversity International, 2021;Rowland et al., 2017), Including bush-meat, fish, fruits, leaves, nuts, and seeds, wild foods are neither cultivated nor purchased but caught or collected by households.In some sites, wild foods may contribute significantly to diets in terms of energy and macronutrients (Broegaard et al., 2017;Nasi et al., 2011); in others, they may provide facilitate dietary diversity (Powell et al., 2013) or supplement important micronutrients (Golden et al., 2019;Tata et al., 2019).Forest-based wild foods may also be more heavily used at particular times of year, during environmental or food crises (e.g.Noromiarilanto et al., 2016;Ntwenya et al., 2017).Forest-based forms of agriculture (e.g., agro-forestry or shifting cultivation) may also help to directly promote more diverse and nutritious diets (Broegaard et al., 2017;Jamnadass et al., 2011).Such forms of agriculture tend to comprise the planting of multiple crops, rather than crop specialisation, and may support forest rejuvenation and the maintenance of biodiversity and ecosystem services (Clough et al., 2011).
Forests may also enhance dietary quality indirectly (Baudron et al., 2019;Gergel et al., 2020).This includes through the wide variety of ecosystem services that benefit agricultural production (including soil fertility, pollination), and healthy environments/bioavailability of nutrients (e.g., water quality; Reed et al., 2017).Where there is good access to nearby markets and favourable terms of transaction, the collection and sale of timber and non-timber forest products (NTFPs) may enable increased incomes for forest-proximate communities, allowing the purchase of more diverse and nutritious foods, or inputs for agricultural production (Dawson et al., 2014).An analysis of approximately 8000 forest-proximate households in 24 low-and middle-income countries reported that forest income (derived from forest products) contributed to over a fifth of total income (Angelsen et al., 2014), with forest income shares greatest for lowest income households.Fuelwood obtained from forests can also support food preparation activities for rural households (Baudron et al., 2017;Powell et al., 2015).
Conversely, a small but growing body of evidence shows that ecosystem alteration, including forest loss, can have negative impacts on the nutrition and health of proximate communities (Assaf et al., 2018;Pienkowski et al., 2018;Whitmee et al., 2015).Deforestation 2 is occurring at rapid pace across the world, with 420 million hectares of forest lost between 1990 and 2020, particularly in Latin America, the Caribbean and sub-Saharan Africa (Curtis et al., 2018;FAO, 2022).Just as proximity and access to forests may support nutritious and sustainable food provisioning, the loss of forest cover may be negatively associated with dietary quality and human health.Forest loss may directly inhibit access to valuable food sources, such as wildmeat, insects and wild fruits and vegetables.Modelling based on a small cohort of preadolescent, wildlife-reliant, Madagascan children suggested that reduced access to wildlife (due to reasons including forest loss) would be associated with a 27% increased incidence of anaemia (Golden et al., 2011).The negative impact on diets is likely to be greatest for those most reliant on wildlife, with poor access or limited means to access nonforest food alternatives (Galway et al., 2018;Ickowitz et al., 2016).Reduced access to wild foods may also erode the "safety net" that forest foods may provide for certain population groups at times of seasonal shortage.Forest loss may also entail declines in agricultural practices such as shifting cultivation and agroforestry, which often support a greater diversity of food crops.Indirectly, forest loss may affect nutrition via declines in the availability of fuelwood and non-timber forest products that support local food practices, livelihoods and provide purchasing power to obtain nutritious foods.Forest loss may also see declines in ecosystem services (e.g., regulating hydrological cycles, supporting pollinators) that support and regulate productive agriculture (Reed et al., 2017).Rasolofoson et al. (2018) also noted that where forest loss increases the travel distance to collect forest products, the time that primary food providers (often women) spend away from food preparation or other caregiving activities may be increased, with negative effects on dietary quality (Mishra and Mishra, 2017).
The advent of publicly available, high-resolution satellite data on land (including forest) cover and change, alongside georeferenced health data, has enabled a flurry of publications examining the linkages between forests and nutrition-related outcomes at the national and regional level (Acharya et al., 2020;Assaf et al., 2018;Galway et al., 2018;C. Hall et al., 2019;Hall et al., 2022.;Ickowitz et al., 2014Ickowitz et al., , 2016;;Johnson et al., 2013;Rasmussen et al., 2020;Rasolofoson et al., 2018).These studies (summarised in Supplementary material) have generated important hypotheses around forests and food -namely, suggesting that proximity to healthy forest (i.e., more expansive, experiencing no net loss) may support higher quality diets and nutritional outcomes.To our knowledge, no previous studies have exploited national-scale datasets to examine forest-nutrition pathways in India.However, with the country home to almost a quarter of the world's undernourished, a rising double burden of malnutrition, and with vast forest and tree cover, this is an important omission from existing literatureand one to which this study contributes.
According to the biennial State of Forest report from the Forest Survey of India, the nation boasts 809,537km 2 of forest and tree cover, representing almost a quarter (24.62%) of the country's total land area (Forest Survey of India, 2021).Whilst total forest and tree cover is reported to have increased nationwide by 1540 km 2 from 2019 to 2021, there have been concomitant declines in moderately dense (40-70% canopy density) forest cover.Scrub (<10% canopy density) forest area increased during the same period, and 11 states reported declines in total forest cover (Forest Survey of India, 2021).In 2012, 5 % of the world's forest proximate people resided in India: an estimated 88 million people in India lived within 5 km of forest, representing an increase of over 10 million since 2000, attributed principally to population increase (Newton et al., 2020).Rights of access to forest vary across the country according to political classification of forest lands, which dictate the extent of human use permitted, as well as territorial rights claims under the Scheduled Tribes and Other Traditional Forest Dwellers (Recognition of Forest Rights) Act of 2006 (herein referred to as the Forest Rights Act [FRA]; (Pratap, 2010).
The Indian experience illustrates that economic growth and increased spending on nutrition alone is insufficient to reduce undernutrition (Subramanyam et al., 2011;Vollmer et al., 2014).India boasts the most extensive and expensive food welfare support programme in the world, with the cost of food subsidies rising 25-fold from the early 1990s to reach over $16 billion in 2012 (Kishore et al., 2013), yet India has made little progress in reducing undernutrition within recent decades.India was ranked 107th out of 121 countries on the 2022 Global Hunger Index, with a level of hunger deemed serious (Von Grebmer et al., 2022).A legacy of India's Green Revolution in the 1960s, mainstream food and agricultural policies in India have historically focused on the production on energy-rich staple grains and improving access to 1 Forest proximity cannot always be conflated with forest dependency.Forest dependency is generally used to define human populations that benefit to some extent from forests.However, there is divergence in meaning and estimates of how many forest dependent people (FDP) live worldwide (Newton et al., 2016).FPP can be considered at least a partial proxy of dependence.
2 Deforestation is defined as the conversion of forest to other land use independently of whether human-induced or not (FAO, 2022).

C. Milbank
these via its extensive systems of food welfare (Banik, 2016).Whilst these dual cornerstones of Indian agricultural policy might successfully support the provision of sufficient dietary calories, the seemingly narrow policy focus on the adequate provision of staplesand lack of attention to more nutritious and diversified food productsdemonstrably continues to fail to address other forms of malnutrition.It also fails to address concerns about the control of access to food (food sovereignty), which can be seen as distinctive to enhancing food supply, the latter of which has tended to be the focus of Green Revolution policies.Against this backdrop, examining the relationships that might exist between forest health and human nutritional health within India constitutes an important empirical exercise.This study assessed whether statistically significant associations between metrics of forest cover and dietary quality and nutrition exist within India.It thus assesses the extent to which the nutritional status of India's population may be related to the state of the wider forested and non-forested environment, which has important policy implications for both food security and natural resource management.

Materials and methods
In line with findings from other low-and middle-income contexts, it was hypothesised that higher forest cover, and reduced forest loss, would be associated with improvements in indicators of nutritional health in Indian infants and young children.In four primary models, the relationship between forest cover and change, and selected indicators of dietary intake and nutritional outcomes in Indian infant and young children aged 6-59 months were examined at national and regional level.The four outcomes of interest were (1) minimum dietary diversity, (2) dietary diversity scores, (3) stunting, and (4) wasting.The focus on infants and young children recognises both the vulnerability of this agegroup to malnutrition (Bhutta et al., 2017) and importance of this critical feeding window for child development and subsequent noncommunicable disease risk in adulthood (Black et al., 2013).Potential confounders of forest-nutrition relationships were identified and adjusted for, and sensitivity analyses also completed.Sub-regional analyses allowed sites where associations differed to aggregate analyses to be identified.This included areas with high % Indigenous representation and where access rights may differ.Explanations were derived for why these regional differences might exist based on prevailing literature documenting socio-and political-ecological conditions on the ground.Whilst these analyses take forest proximity to be a proxy for forest access and use, we acknowledge that myriad factors may also condition access to forest resources (e.g., legal, and physical infrastructure; Newton et al., 2016).This is particularly true in a forest landscape as politically and ecologically diverse as India's.All data preparation and analysis were performed using ArcGIS [ESRI (UK) Ltd., 10.8.1], Google Earth Engine, and STATA version 17 (STATA Corporation, College Station, TX, USA).

Data sources
This study used individual and household-level data from the Demographic and Health Survey (DHS) programme for India, and the most recent remote sensing data provided by (Hansen et al., 2013) and the Global Land Analysis & Discovery (GLAD) lab at the University of Maryland, Google, USGS, and NASA.
Data on infant dietary intake and nutritional outcomes came from the latest round of the Demographic and Health Survey, known within India (and herein referred to) as the National Family Health Survey (NFHS).The NFHS was initiated in India in 1992/3 and uses standardised questionnaires and processing methods to aggregate population health data that are comparable across states, as well as other countries (IIPS/India and ICF, 2022).Surveys are completed at five-year intervals.Used within this study, NFHS-5, completed in 2019/20, is the most recent.NFHS selects households using a two-stage stratified strategy, which supports proportional (to size) representation of populations from each state/union territory, rural and urban households, slum and nonslum households, and scheduled caste (SC) and scheduled tribe (ST) groups (IPBES/India and ICF, 2022).NFHS-5 covered 28 states and eight union territories of India.Almost 636,700 households participated, with an average household response rate of 98% (IIPS/India and ICF, 2022).Data on possible confounding variables (Table 1) were also extracted from the NFHS (DHS, 2022).
Data on forest cover and change came from the publicly-available Global Land Analysis & Discovery (GLAD) lab at the University of Maryland, Google, USGS, and NASA (Hansen et al., 2013) v.1.9).Forest cover data are defined for the year 2000 and defined within the dataset as canopy closure for all vegetation taller than 5 m in height.These data are the most recent available on tree cover and are provided as percentage values (per pixel).Tree cover quintiles were also derived from these data for model inclusion.Several bands of forest cover change data are made available for 2000-2021: "loss" is defined as a change from a forest to non-forest state (stand-replacement disturbance) relative to baseline and is scored as 1 if loss has occurred during any year between 2000 and 2021 and zero in the case of no loss.Within the dataset provided by Hansen et al., "gain" is defined as the inverse of loss (i.e., nonforest to forest change) and is scored as 1 if gain has occurred during any year between 2000 and 2012.All data are available at 30 m resolution.
The NFHS uses a 'cluster' as the geographical sampling unit.Depending on population size, clusters correspond to a village or group of settlements (Dontamsetti et al., 2018).Geographic position system (GPS) data are recorded for all clusters during NFHS data collection or at the listing stage of the national survey.These GPS data points allows forest data and demographic data to be spatially joined.For reasons of confidentiality, the coordinates marking the centre of each cluster are displaced ("geo-masked" or "geo-scrambled") from the actual location by up to 2 km for urban clusters and up to 5 km for rural clusters (DHS Spatial Interpolation Working Group, 2014).To account for displacement, we created 5 km buffers around clusters, with tree cover (%) and cover change values calculated per buffer polygon.In doing so, we take forest proximity to be a proxy for access and the ability to derive benefit from forests, acknowledging that other important factors may also condition access (e.g., legal, and physical infrastructure; Newton et al., 2016).Two main forest loss variables were defined: (1) the area within a defined buffer that registered loss during 2012-2020 (continuous variable), and (2) a binary variable ("Recent Loss") representing the occurrence of any recent loss of forest within the buffer.In defining both variables, any pixels that registered gross forest loss after 2020 or before 2012 were recoded as loss = 0 to ensure that forest cover change (our primary exposure variable), preceded the outcome of interest and recognising that meaningful gain may have occurred in areas reporting gross losses prior to 2012.With gain data beyond 2012 unavailable within the secondary geospatial dataset, we assumed that any regrowth since 2012 would likely not constitute "nutritious" forest regrowth (for example, regrowth may constitute young forests with limited potential to confer nutritional benefits).

Statistical analyses
Four separate regression models were fit to examine the relationships between the two environmental variables (forest cover and forest loss) and child nutritional outcomes (DDS, MDD, stunting, and wasting).
In children aged 6-23 months, models were run to examine the association between forest cover and forest cover change with dietary diversity score and minimum dietary diversity (MDD).NFHS-5 asked female respondents about the diet of their youngest living child aged below 24 months, including questions on whether children had consumed foods from specific food groups in the 24 h preceding the survey.From these responses, infant and young child dietary diversity scores (IYCDDS) and MDD were defined according to the latest WHO Global Nutrition Monitoring Framework (WHO and UNICEF, 2017) based on the consumption of eight food groups (breastmilk; grains, C. Milbank roots, and tubers; legumes, pulses, and nuts; dairy products; flesh foods; eggs; vitamin A rich fruits and vegetables; other fruits and vegetables).Minimum dietary diversity was defined as having received five out of the eight food groups in the preceding 24 h (WHO and UNICEF, 2017).Minimum dietary diversity was computed as a binary variable, where "1" indicated the (met) consumption of minimum dietary diversity.To avoid a loss of information from only using a binary variable, separate models were run with dietary diversity score as a discrete, dependent variable, and MDD as a binary variable.
In children aged under 5 years, associations between forest cover and forest cover change with stunting and wasting were explored.Stunting is a measure of chronic nutritional deficiency -demonstrating the impacts of prolonged poor nutritionbut also reflects the impact of repeated or chronic infections during early childhood (Black et al., 2008).Wasting is a measure of recent, shorter-term nutritional deficiency, reflecting recent weight loss or failure to increase weight (as well as recent acute sickness, including diarrhoeal episodes).Under NFHS-5, all children aged under 5 yrs.that were listed in the household questionnaire (i.e., a registered household member) had their height and weight measured.As data are prepared for public use, "z-scores" are calculated and provided by NFHS according to the latest WHO Growth Standards (World Health Organization, 2006).Z scores represent the deviation of a child's height-for-age or weight-for-age from the median of the international standard population, divided by the SD of the standard population.The WHO defines a child aged under 5 yrs.as stunted when their height-forage z score (HAZ) is more than two standard deviations below the mean on WHO Child Growth Standards.The WHO defines a child aged under 5 yrs.as wasted when their weight-for-age z score (WAZ) is more than two standard deviations below the mean on WHO Growth Standards.Since z-scores are sensitive to age, children with missing age data are not provided z-scores.Further, children where HAZ scores are below or above ±6SDs, or WAZ are below or above -6SDs and + 5SDs, scores are flagged as invalid.For this study, HAZ and WAZ data were extracted on the first listed child with valid and non-missing data that was within the 6-59-month age range.We defined stunting and wasting as binary variables according to the WHO definition (>2SDs).
When binary outcomes (e.g., minimum dietary diversity) were examined, logistic regression models were used, with odds ratios reported.Poisson regression was used when continuous outcome variables were examined (e.g., dietary diversity score) and the mean exceeded the variance.For models that included forest loss as the exposure variable of interest, forest cover in the year 2000 was also included to control for the starting level of cover within the region (following Assaf et al., 2018).Only one child per household was included in each model to avoid clustering at the household level.Given that complementary feeding (after exclusive breastfeeding) is recommended at 6 months, children aged below 6 months were excluded from all analyses (WHO and UNICEF, 2017).Standard errors were clustered at the community level to account for correlations between observations within a cluster.For all analyses, an alpha level of 0.05 was used as the threshold for statistical significance.
Alongside nutrition and forest cover variables, potential confounders of the forest-nutrition relationship (extraneous factors that distort the association between two other variables; (Woodward, 2013), and moderating factors along the causal pathway were identified and their effects examined in modelling.These additional variables include child characteristics, household characteristics and environmental/community factors, and were selected based on previous identification in the literature, and hypothesised factors that may plausibly influence the intake of nutritious diets.Recognising that levels of market integration may be related to forest proximity and influence dietary intake (Ickowitz et al., 2016), travel time to city with a minimum of 50,000 was used as a proxy for market participation.Noting the limitations of previous works, we included data on caste and tribal status, recognising thatparticularly in the Indian context -different populations have different use and access rights to forests, and may be able to derive different nutritional benefits.Variables that may affect child feeding and nutritional status, such as participation within Aanganwadi and other local programmes for child development, are not provided under NFHS-5 and could not be linked with the dataset.
Prior to multivariate modelling, all variables were tested for collinearity using Pearson's r and examined in a correlation matrix to prevent collinearity in the multivariate model.Initial models contained all variables that were significantly (p < 0.05) related with outcome variables in univariate regression analyses, and variables considered important confounders based on existing literatures (Table 1).Variables were subsequently removed one-by-one based on their relevance as confounders and effect on model performance.Confounding was assessed by examining the change in effect size (e.g., ORs) for the association between the outcome and remaining predictors between the model that included the confounder in question and the model with it removed.The final models contained all variables that remained significant throughout adjustment and other variables considered important confounding factors.

Regional sub-analyses
Given India's vastness and high biocultural diversity, it is highly plausible that forest-nutrition relationships vary across the country and may be obscured in national-level analyses (Ickowitz et al., 2016).Forest and land use practices, as well as regulations around access, can vary across regions (Choudhury and Aga, 2019;Lee and Wolf, 2018).The northeast of India, for example, is governed under Sixth Schedule of the Constitution of India and has a degree of autonomy, forest governance and tenure regimes differ quite substantially to the rest of the country.Climate and therefore forest type also varies across the nation, resulting in differences in the possible dietary benefits that can be gained from forest proximity.Cultural, socio-economic, and ecological factors may also drive differences in dietary and nutritional outcomes in different regions.Unmeasured variables correlated with forest and tree cover and/or nutritional indicators may also vary spatially, affecting reported model outputs (Ickowitz et al., 2016).
To explore possible regional differences, data were extracted on the states in which clusters were located and grouped states into distinctive regions.Models were run on seven regional sub-samples, comparing the results between regions and with those of aggregate models.There is much debate about what constitutes the different regions of India: our classification of seven regions (Table 2) was informed by commonlyheld divisions as well as knowledge of the socio-political and ecological strata within India.States within our northern hilly region (Jammu and Kashmir, Ladakh, Himachal Pradesh and Uttarakhand) are not normally distinguished from other states in the north, but they are climatically and geographically distinctive compared to Punjab, Chandigarh, Haryana, NCT of Delhi, Rajasthan, and Uttar Pradesh).Similarly, Jharkhand is included in the Central region with Madhya Pradesh and Chhattisgarh in recognition of its high Scheduled Tribe population, who may utilise forests in different ways to their non-tribal counterparts.Odisha, another state with a high tribal population, is included in the Eastern region on account of its long coastline, and the influence of this marine system on dietary choices.

Forest and tree cover and dietary intake
The first set of analyses examined a sample of 62,332 infants and young children in India aged 6-23 months with complete dietary intake data (Table 3).Of these, over half were female (52%), and the average age was 14 months.Almost 15% were fully weaned, and over two thirds were receiving complementary feeding alongside breastfeeding.Almost 80% came from rural households (79.7%) and almost half of households (47.7%) owned land usable for agriculture.Almost a quarter of these households came from the lowest wealth quintile (23.6%); around one in five belonged to Scheduled Castes (20.88%) and Scheduled Tribes (20.36%).Data were limited on the paternal education of infants and young children (84.9% unspecified).Of their mothers, over half had completed secondary education (53.5%); almost one in five had received no education (19.2%).
The mean number of food groups consumed by infants and young children in the 24 h preceding the survey was 3.27 (SD = 1.92), with grains being the most consumed food group besides breastmilk (N = 42,021; 67.4%).Almost half (48.8%) of infants and young children had consumed fruits or vegetables in the preceding 24 h, including over 43% having consumed vitamin A rich fruits or vegetables.The consumption of flesh foods (13.5%) or eggs (18.2%) was lower.Around a quarter of infants and young children reached more than five food groups, which is considered the "minimum" threshold for dietary diversity for this age group (MDD; WHO, 2017).More than a third of children aged 18-23 months met minimum dietary diversity standards, compared to 14% of children aged 6-11 months, reflecting an expected age trend.The northeast was the region with the greatest proportion of infants achieving minimum dietary diversity (34.35%) and highest average diversity scores.Average tree cover (cluster-level) was 14.6% and varied from 0% to 84.25%.>40% of infants and young children came from households that had experienced proximate tree cover loss since 2012; over 1 in 5 had not experienced loss.The average travel time to a city of >50,000 inhabitants was 34 min; for around 7% of households travel time exceeded 90 min.Average altitude of households was above 360 m, with a maximum altitude of 5700 m.
Table 5 summarises the results of multivariate regressions.After adjustment, tree cover was statistically significantly associated with minimum dietary diversity (≥5 food groups consumed) in infants and young children aged 6-23 months, with higher effect size at greater coverage quintiles.For example, the odds of achieving minimum dietary diversity were 40% higher for infants and young children living in areas with greatest tree coverage compared to lowest (OR MDD = 1.406; 95%CI 1.265, 1.536 at greatest; model 1).With dietary diversity defined as a continuous outcome variable, tree cover was positively significantly associated with dietary diversity score (model 2).Whilst odds were negative, recent proximate tree cover loss was not significantly associated with minimum dietary diversity (p > 0.05).Some demographic variables were statistically and positively associated with dietary diversity in both models.For example, children aged 18-23 months had almost 3.7 times the odds of meeting minimum dietary diversity than children aged 6-11 months.Infants and young children in the wealthiest households had over 1.4 times increased odds of meeting minimum dietary diversity than the poorest (OR MDD = 1.426; 95%CI 1.319, 1.542).Infants and young children from Scheduled Tribe households achieved significantly greater diversity scores than those with no Scheduled background (OR MDD = 1.126; 95%CI 1.045, 1.212), whereas the opposite was true for infants and young children from Other Backward Classes 3 (OR MDD = 0.875; 95% 0.825, 0.927).Maternal but not paternal education was positively associated with infant and child dietary diversity.The odds of achieving MDD for infants and young children living in a rural area were significantly reduced (OR MDD = 0.849; 95%CI 0.800, 0.902), and increased with altitude (e.g., OR MDD = 1.314; 95% CI 1.208, 1.429 for highest altitude compared to lowest).

Forest and tree cover and nutritional outcomes
The second main set of analyses examined relationships between tree cover (static and change) and nutritional outcomes in children under 5 yrs.The stunting sample comprised 146,664 infants and young children aged 6-59 months from different households in India and the wasting sample comprised 147,190 infants and young children, based on the same selection criteria (Table 4).For both samples, less than half of children were female (48%), with an average age of just below 3 years.Almost 80% of children came from rural households (79.1%) and almost half of households (46.8%) owned land usable for agriculture.Almost a quarter of households came from the lowest wealth quintile, and around one in five belonged to Scheduled Castes and Scheduled Tribes.Over half of mothers (of measured children) had completed secondary education (52%); 21% had received no formal education.
Over a third of children could be classed as stunted (35.9%; heightfor-age Z scores less than 2SDs below the mean on WHO Growth Standards), including almost 15% of children severely stunted (more than 3SDs below).These figures are comparable to those presented by the Global Hunger Index for 2022 (35.5%).Stunting prevalence in the sample was highest in the Northeast (39.7%) and Eastern regions defined (39.1%), and in infants aged 3-4 years (38.8%).Over 17% of children could be classed as wasted (weight-for-height z score less than minus 2SDs below the mean on WHO Growth Standards), including 7% severely wasted.Over 5% could be classed as overweight (6.78%).Wasting prevalence was highest in the Central belt region (Madhya Pradesh, Chhattisgarh, Jharkhand), and the prevalence of overweight was highest in the South.
Average tree cover (cluster-level) was over 14% and varied from 0% to 82.1%.For over a fifth of children, no proximate forest loss was recorded at any time since 2000, however over 40% of children came from households that had seen forest loss in their surrounding region since 2012; almost 13% had seen loss prior to 2012.Loss was largely concentrated in the South and North of the country, where over half of households had experienced recent proximate loss.The greatest losses before 2012 were seen in Northeast and Eastern regions.The average travel time to a city of >50,000 inhabitants was under 35 min for both samples, and for around 7% of households travel time exceeded 90 min.Average altitude of households was above 370 m, with a maximum altitude of 5743 m.
In multivariate analysis (Table 5), tree cover was strongly and negatively associated with stunting, with the lowest odds of stunting observed at the greatest tree coverages (OR = 0.835; 95%CI 0.786, 0.886).Loss of tree cover within the last decade saw an increase in odds of stunting, but this increase was non-significant (p > 0.05).After adjustment, the odds of stunting were reduced for maternal education (e.g., OR ST = 0.612; 95% CI 0.583, 0.641 for higher education vs no education), wealth (OR ST = 0.454; 95%CI 0.433, 0.476 for wealthiest vs poorest), ownership of land usable for agriculture (OR ST = 0.952; 95%CI 0.927, 0.976) and travel time to nearest city of >50,000 inhabitants (OR ST = 0.808; 95%CI 0.720, 0.907 at furthest vs nearest).After adjustment, odds ratios for stunting remained significantly higher for all Neither tree cover quintiles nor recent loss were significantly associated with wasting.Maternal education (OR WA = 0.781; 95% 0.724, 0.814 for higher vs none) and wealth (OR WA = 0.733; 95%CI 0.691, 0.777) were significantly associated with reduced odds of wasting.Belonging to a Scheduled Caste or Tribe was associated with significantly increased odds of wasting.Living at the highest altitudes was associated with reduced odds of wasting (OR WA = 0.908; 95% CI 0.852, 0.968 compared to lowest).Regionally, the Central "Tribal" belt saw the highest odds of wasting (OR WA = 1.477; 95%CI 1.366, 1.597) against the North of India as reference.

Regional sub-analyses
Fig. 1 summarises the results of multivariate regressions run for each outcome class within each defined region within India.These analyses show that in the northeast of India, there is a positive significant relationship between tree cover and dietary diversity (OR MDD = 2.28; 95%CI 1.57, 3.31 at greatest vs least cover), and a negative relationship between recent proximate forest loss and dietary diversity (OR MDD = 0.88; 95% 0.78, 0.98) in infants and young children aged 6-23 months.In this region, tree cover was also associated with reduced odds of stunting, and of wasting at upper tree coverage quintiles.In the East and West, tree cover was positively associated with dietary diversity, and was associated with reduced odds of stunting at most cover classes in the West and medium coverage in the East.
In the Central belt, tree cover was only significantly associated with minimum dietary diversity at the lower tree cover class (vs lowest); forest cover loss was associated with an improvement in dietary diversity.The South was the only region not to see any significant associations with any of the forest variables examined.

Discussion
To date, several studies have identified statistically significant associations between measures of forest cover and indicators of infant and young child dietary intake and nutritional outcomes (Ickowitz et al., 2014;Pienkowski et al., 2018;Rasolofoson et al., 2018).Building on this work, this study contributes findings from India, a region with high burdens of child undernutrition and vast forest cover, and of which these empirical questions around forest-nutrition relationships have not previously been investigated.In agreement with earlier work, the results demonstrate statistically significant associations between indicators of forest cover and measures of dietary quality and nutritional status at the national-level in India.At aggregate level, static tree cover was associated with improved dietary diversity in infants and young children aged 6-23 months, and with reduced likelihood of stunting and wasting in children aged under 5 yrs. in areas with the thicker tree coverages.These are plausible relationships, with pathways between forests and nutrition previously conceptualised and studied through primary and secondary research (Powell et al., 2011;Baudron et al., 2019).Whilst the mechanisms underlying such relationships have not been specifically examined here, there are a number of pathways that may be driving the relationships seen.Such pathways might include the direct provision of forest-sourced wild foods, an income pathway (derivation of income from forest products enables purchase of higher quality diets), and an agroecological pathway (whereby ecosystem services provided by

C. Milbank
forests and trees supports agricultural production; Baudron et al., 2019).Of these, the direct food pathway has been shown to be an important and consistent explanatory pathway in previous work (Baudron et al., 2019;Golden et al., 2019;Tata et al., 2019;Powell et al., 2011), with wild foods sourced within and surrounding forest often nutrient-dense and providing important dietary diversity, including in India (Cheek et al., 2022;Narayanan, 2021).Several other variables examined were shown to be significant within aggregate analyses and are worthy of further discussion here.First, Scheduled Tribe (ST) status was associated with improved dietary diversity scores, whereas Other Backward Class (OBC) status was associated with reduced dietary diversity scores.In India, STs are eligible to make claims to forest land under the FRA, whereas OBCs are not (nor are Scheduled Castes).Despite this, higher odds of stunting or wasting were observed for all groups of Scheduled or OBC status, which -in line with existing work (e.g., Kumar et al., 2020;Saxton et al., 2016) -supports that non-food factors may be important drivers of stunting/wasting outcomes in children belonging to these groups.Maternal educational attainment, household wealth and child age were consistently associated with all outcomes, in line with well-established associations (Alderman and Headey, 2017;Black et al., 2008).Seasonality was associated with lower dietary diversity in pre-monsoon (January-March) and post-monsoon (October-December) seasons, and increased odds of wasting during monsoon (April-June) and postmonsoon seasons.Seasonality affects food availability and consumption of both cultivated and wild biodiversity.Previous empirical work has noted that seasonal dependence on wild foods, sourced from within and surrounding forests (Powell et al., 2011;Tata et al., 2019).The results here represent the combined outcome of seasonal changes in cultivated and wild food availability, access, and consumption.Forestsourced foods are not distinguished within NFHS-5 dietary data, precluding insight into the seasonal contributions of forest-sourced foods.
Under favourable conditions, previous studies have shown that market access can support higher quality diets (Jones, 2017).For forestproximate communities, the derivation of income from sale of forest products can enable the purchase of nutritious foods, thus enhancing linkages between forests and nutrition (Sibhatu et al., 2015).However, counter to some previous work (C.Hall et al., 2019;Ickowitz et al., 2014), this analysis found that longer travel times to the nearest city of 50,000 inhabitantswhich was used as proxy for market access -were associated with higher dietary diversity and reduced odds of stunting at aggregate level.Whilst the associations seen here might indicate the lesser importance of an income pathway in India, it is perhaps more plausible that the proxy used does not make visible the importance of more local markets, situated in smaller towns and settlements, in generating forest-based income ( (Shackleton et al., 2011)).With the relationship between market access, forests and foods shown to be highly variable between sites (Baudron et al., 2019;Ickowitz et al., 2018;Skov Olesen et al., 2022), these hypotheses on the importance of the income pathway between forests and nutrition in India warrants further investigation, based on more robust indicators of market participation.
At aggregate level, the occurrence of recent, proximate forest loss was not significantly associated with the diet and nutrition outcomes examined.There are numerous reasons that this may be (not mutually exclusive), including that forest loss may not have been used in ways that support diet and nutritional outcomes (e.g., as wild food source, for agroforestry or NTFP collection).Alternatively, where forest was used in such a way, proximate communities may have found ways to mitigate against losses and supplement diets in other ways.A third and important explanation is that the forest lost may have been inaccessible to communities.Questions of access are particularly pertinent within the Indian context, where rights claims made by tribal communities under the Forest Rights Act (FRA) have been unevenly administered across the country (Lee and Wolf, 2018), and in many places violated (Choudhury and Aga, 2019).In line with a growing global trend towards more stringent conservation policies, the nation's ongoing expansion of its protected area network and increased restrictions on natural resource has been seen to undermine the access rights permitted by the FRA and threatens to destabilise links between forests and nutrition (Fanari, 2022;Skov Olesen et al., 2022;Vasquez and Sunderland, 2023).The remote sensing data used here did not allow accessible forest and inaccessible forest to be distinguished, which is a critical limitation of this work.Individuals may live adjacent to forest but local regulations, tenure systems, or customs may impose restrictions on use, thus depriving local communities of the benefits of their forest proximity.Data that examine whether rights to forest (under India's FRA) have been successfully claimed remain in their infancy and could not be Whilst aggregate-level analyses offer useful broad patterns, comparison of national results with regional sub-analyses suggest that relationships seen at national level do not always hold at the regional level.The same has been shown in previous work (Ickowitz et al., 2016;Rasolofoson et al., 2018).The Central region was the only region to see forest loss significantly associated with minimum dietary diversity (MDD) in infants, and only the low tree cover quintile was associated with MDD.This combination of anomalous results, which on the face of it might appear to undermine forest-food hypotheses, might be explained by expansion of the agricultural frontier into areas of tree cover within the Central belt.The central belt, including Madhya Pradesh, Chhattisgarh, Jharkhand states, is characterised by thin, thorn and dry tropical forest (Forest Survey of India, 2019).Whilst limited evidence shows that dry forests can be nutritionally valuable (Rowland et al., 2015;Sunderland et al., 2015), it may still be true that the loss of thin forest to make way for agricultural production may increase dietary diversity more than the former forest did, either as the foods directly provided are more nutritious, or via enhanced income opportunities from non-forest livelihoods.In short, a non-forested state covered by agriculture, may be more "net nutritious" than the forested state in this region.These transitions need further verification and empirical, longitudinal study in situ, and the exploration of indicators of forests' "nutrition potential" is earmarked for future research (Gergel et al., 2020).
In the South of India, no statistically significant relationships between indicators of tree cover/loss or diet-nutrition outcomes were visible.Despite more than half of included households in the South being situated in areas (5 km buffers) of the thickest forest cover, this lack of visible relationship suggests that households may not be deriving nutritional benefit from proximity to forest.This may indicate a dietary transition away forest dependence, towards alternative forms of nutrition, in a part of India where wealth is high (NITI Aayog, 2021)so whilst dense, moist forest exists, it is not accessed for nutritional purposes, or not to an extent that is evident in aggregate data.Such transitions may also be experienced in other Indian regions, including the Central belt.
Interestingly, in the northeast, tree cover was a strongly associated with improved dietary diversity, and loss was associated with worsened diversity outcomes.Reduced odds of wasting in children under 5 yrs.were seen with increased tree cover, and odds were increased with recent and older loss.In this region, not only is forest cover high (representing almost 65% geographical area; (Forest Survey of India, 2021), but forest proximity can be assumed to represent a far better proxy for access and use -which is a critical limitation of previous analyses.In the northeast, Scheduled Tribes constitute a majority population, and four of the states that constitute the northeast are governed under Sixth Schedule of the Constitution of India, an instrument of "tribal self-rule" (Datta and Sen, 2020).This grants the region a degree of autonomy, and consequently, forest governance and tenure regimes differ quite substantially to the rest of the country.Whilst certain national legislation (such as the Wildlife Protection Act) that restrict use of natural resources apply to the northeast region, it is plausible that self-governance regimes still allow forest-proximate communities to take fuller advantage of forest resources in ways that communities elsewhere in India cannot.Indeed, limited in situ study within the northeast supports the nutritional value of forest-based agriculture in securing diverse diets comprising wild and cultivated plant and animal biodiversity (Pandey et al., 2022).However, despite its high forest and tree cover and strong regional forest-nutrition associations, it is important to note that levels of malnutrition in the northeast remain amongst some of the highest in India (IIPS, 2022).As in other regions of India, staple crop production and the cultivation of non-food cash crops (such as rubber and broom grass) has been promoted under north-eastern state policies.Such shifts away from traditional forest-based cultivation have been associated with reductions in dietary diversity, increased consumption of white rice and reduced consumption of traditional staples (millet, maize, and brown rice) and wild foods (Behera et al., 2016).Given the region's ongoing nutritional challenges and rich forest biodiversity, such political-ecological drivers of poorer nutrition warrant further investigation.
There are several important limitations to this analysis and others of its nature.Chief amongst these is that whilst independent and control variables were carefully selected and defined to precede the outcomes examined, statistically significant associations do not indicate causality.The analysis was conducted for one NFHS survey round only: analysis of successive rounds of the dataset alongside changing forest cover may offer greater insight into the ways that trends in forest cover are associated with dietary and nutritional change.Shortfalls in the geolocated datasets used are also worth highlighting.For example, the data used also do not enable distinction between forest types/classes (for example primary forest vs plantation): different types of forest, their age, structure, fragmentation, disturbance regimes may have different "nutrition potential", with some better able to support improved nutrition (Gergel et al., 2020).Analysis by Rasmussen et al. (2020) for example, supported that the forest fragmentation can affect the food available, the way that people access forest resources or manage forest-centred agriculture (Rasmussen et al., 2020).The parameterisation of forest cover as closed canopy at 5 m (defined by Hansen et al., 2013) does not permit insight into the possible forest structural factors that may support nutrition.Dense non-food monoculture, for example, that have already replaced substantial swathes of tropical forests within southeast Asia (Tropek et al., 2014), would fall under this definition of forest, and are unlikely to host high edible biodiversity.The timeframes of forest loss (between 2000 and 2021) and gain (2000−2012) data, which were pre-defined within the Hansen et al. dataset, also limit the analysis, as any regrowth occurring after 2012 that may contribute significantly to forest-based diet would not be accounted for.In a similar vein, the dataset also presented forest loss with respect to the baseline year, and not as actual reduction in tree cover.As remotely-sensed datasets evolve and enable more precise insight into greater breadth of forest-level variables, future research should seek to explore how actual change in forest cover, type and use are related to dietary shifts and nutritional health status.
Whilst dominant paradigms around food production and forest conservation have tended to view the two as competing land uses, this work emphasises the important complementarity that may exist between the two.In line with earlier studies, this study suggests that conservation of forests and trees alongside agricultural production may play a key role in enhancing dietary quality and child nutrition.In India, where an ongoing fixation on the production and provision of dietary staples has failed to address acute problems of child undernutrition, and where the rights of forest-proximate peoples to access and use forests continue to be curtailed in the name of conservation, these synergies have important implications for policy.Providing evidence of the importance of forests to one facet of infant and child diets and nutrition, this work adds impetus to calls for nutrition-sensitive landscapes, in which questions of food security and nutrition are integrated with the management of landscapes and biodiversity (Powell et al., 2013).The results presented at regional-level suggest the importance of contextual factors, particularly tenure and access regimes that govern access to forests, that may moderate forest-nutrition relationships at the local C. Milbank level.This indicates not only the need for smaller-scale and in situ research into such factors, but also highlight how policy interventions aimed at nutrition (and environmental) improvement must attend to and vary according to local socio-political context.Analyses such as these raise hypotheses for further investigation in situ, however such questions cannot be adequately addressed solely with large, geolocated datasets.This work emphasises the limitations of such datasets in revealing mechanisms and processes that affect forest-nutrition linkages at the local level.In situ, primary investigations over a prolonged period are important to disentangle the mechanisms underlying the relationships examined within given contexts.This includes exploring the role of any unexplored socio-ecological factors that may explain the associations seen, and how restricted access to forests and tree-based systems, market-related and other socio-political factors may enhance or threaten forest-nutrition linkages.

Table 1
Summary of variables examined in analyses.

Table 2
Defined regions within sub-analyses.
3The terms "Scheduled Tribe", "Scheduled Caste" and "Other Backward Class" are legally-recognised terms used by the Government of India and within the NFHS dataset and are not terms chosen by the author.The categories refer to groups that have historically faced social, educational, and economic exclusion, discrimination, and disadvantage.

Table 4
Summary Statistics of Model Variables for stunting and wasting analyses in children aged 6-59 months.

Table 5
Summary of national-level results of multivariate regression models for minimum dietary diversity (MDD), dietary diversity score, stunting and wasting.

Table 5
(continued ) * indicates p < 0.01; * indicates p < 0.05.Ref denotes the referenced category used to derive odds ratios.indicatesthat the variable was not included in the final model due to insignificance. * Summary of results of regional sub-analyses for minimum dietary diversity, stunting and wasting.In the analyses here, we can consider only proximity to forest cover, and proximity to forest loss, which under a wellfunctioning Forests Rights Act (and other permissive forest governance regimes, such as in the northeast) may serve as a decent proxy for rights of use in India but does not allow engagement with other socio-political and ecological realities, which may influence forest-nutrition relationships over time and at ground-level.Besides questions of access, the implications of factors such as infrastructural development, rural-urban migration, and rural nutrition programmes, that may confound or mediate forest-nutrition relationships, cannot be untangled via large, geolocated datasets and warrant longitudinal, place-based investigation.