Agroforestry, the integration of trees into agricultural landscapes, offers significant socioecological benefits and serves as a natural climate solution. In India, where farmland constitutes 56% of the land cover, these trees are crucial. However, a lack of robust monitoring has hampered understanding of their distribution and vulnerability. This study addresses this gap by mapping and tracking individual farmland trees across India over a decade, focusing on large trees which contribute most significantly to ecosystem services. The research aims to quantify the loss of these trees and investigate the underlying causes, contributing critical information to the ongoing discussion about agroforestry's role in climate change mitigation and adaptation, as well as its importance for agricultural livelihoods and biodiversity. Understanding the dynamics of these trees is essential to inform effective conservation and management strategies, particularly given the alarming rate of tree die-offs observed in other ecosystems due to climate change, diseases, and human activity. The study will quantify the loss of mature farmland trees, providing critical baseline data for future research and policy interventions.

Existing literature highlights significant tree mortality in forests due to wildfires, fungi, insects, and droughts, exacerbated by climate change and human activities. However, monitoring of trees outside forests, particularly in farmland agroforestry systems, remains limited. While studies have emphasized the potential of tree planting in agroforestry for climate mitigation, the focus on protecting existing mature trees and understanding threats to these established systems is lacking. The scarcity of data on tree populations in India's vast agricultural landscapes underscores the importance of this research. Previous research on mapping trees has often relied on satellite imagery with resolutions too coarse to accurately identify individual trees on farms. This study uses higher-resolution imagery and advanced image processing techniques to overcome these limitations and provide a comprehensive assessment of India's farmland tree resources.

This study utilized high-resolution satellite imagery from RapidEye (2010-2011, 5m resolution) and PlanetScope (2018-2022, 3-5m resolution) to map and track individual trees across India. A heatmap-based deep-learning approach was developed for tree detection, leveraging a large number of manually labeled tree crown centers to train the models. Image processing steps included creating country-wide mosaics, addressing variations in image quality through sharpness filtering, and using a phenological window to optimize tree detection. The tree crown centers from the initial mapping (2010/2011) were tracked using a 15m buffer to account for potential spatial shifts, classifying trees as ‘disappeared’ if not detected in subsequent years (2018-2022). A change confidence metric was employed to quantify uncertainty in tree detection and loss, focusing on reporting only high-confidence losses. Qualitative interviews with villagers across several affected regions were also conducted to explore potential drivers of tree loss. These interviews gathered insights into changes in agricultural practices and local perspectives on tree removal.

The study revealed a significant decline in large farmland trees across India. Between 2010/2011 and 2018, approximately 11 ± 2% of high-confidence trees (crown size >96 m²) disappeared. Moreover, between 2018 and 2022, over 5.3 million trees (crown size >67 m²) vanished. These losses were not uniformly distributed, with hotspots of substantial tree loss (up to 50%) observed in central India (Telangana and Maharashtra). Analysis of high-resolution images showed the disappearance of mature trees even in areas with above-average rainfall in the past decade, suggesting that climate change is not the sole driver of these losses. Qualitative interviews strongly implicated changes in agricultural practices as the main cause of tree removal. The expansion of paddy rice cultivation, enabled by increased water availability from new boreholes, led to the removal of trees perceived to negatively impact crop yields. While the study focused on tree losses, it's important to note that it did not consider gains and masked out block plantations; therefore, it could not provide information on net tree changes.

The findings highlight a significant and concerning decline in India's large farmland trees, particularly mature specimens that provide substantial ecosystem services and contribute to local livelihoods. This loss contradicts reports of increasing tree cover in India, emphasizing the importance of differentiating between planted trees and established agroforestry systems. While the removal of trees is part of agroforestry management, the observed scale of loss suggests a shift toward simplified agricultural landscapes, potentially with negative consequences for biodiversity, carbon storage, and the sustainability of rural communities. The observed losses challenge the narrative of increased tree cover in India, as it focuses on large established trees and not newly planted ones, highlighting the need to balance tree planting initiatives with the protection of existing mature trees in agroforestry systems. This necessitates a shift towards more holistic agroforestry practices that value the ecosystem services and socio-economic benefits of mature trees. Future research should focus on the long-term impacts of the observed tree loss and develop strategies to encourage tree retention in agricultural systems.

This study provides a critical assessment of the decline of large farmland trees in India using high-resolution satellite imagery and innovative deep-learning techniques. The results demonstrate a substantial loss of mature trees, largely driven by changes in agricultural practices, rather than solely climate change. These findings emphasize the need for more sustainable land management practices that integrate the preservation of existing trees with agricultural intensification to support both food production and environmental sustainability. Further research is needed to analyze the long-term impacts of this tree loss on ecosystem services and livelihoods, and to develop strategies that promote tree retention within agricultural landscapes.

The study primarily focused on detecting losses of large trees; hence the gains of smaller trees or net changes are not fully quantified. The reliability of tree detection varies depending on image quality and the size of the trees, introducing some uncertainty into the results. The use of a 2020 land-cover map to mask out non-farmland areas could also introduce some error in characterizing historical land cover changes. While qualitative interviews provided valuable insights, a larger-scale quantitative survey could provide more robust evidence.