Tools

Intra-annual land use and land cover classification

Our research group focuses on developing high-resolution (10m) intra-annual Land Use and Land Cover (LULC) classification maps tailored for dynamic landscape monitoring. Our classification scheme includes 12 LULC classes, comprising both static categories such as built-up areas, trees, barren lands, and shrubs & scrubs, as well as dynamic intra-annual classes. Notably, our maps capture water seasonality, which reflects the presence of water across different agricultural seasons, and cropping frequency classification, which determines the number of agricultural cycles a tract of land is used for.

The selection of these classes is driven by the need to track land cover changes with high temporal accuracy, ensuring timely insights for agricultural policy formulation and watershed management. 

Our classification pipeline integrates both supervised and unsupervised techniques, optimized for efficiency and scalability. We enable large-scale pan-India LULC mapping while operating within the free-tier limits of platforms like Google Earth Engine (GEE). This ensures that researchers, policymakers, and stakeholders can execute and customize our classification workflows without requiring high-end computational infrastructure. We are working on the latest paper of this research work and will update the link soon. 

Example: LULC map of Barwani district in Madhya Pradesh for the hydrological years 2021.

• Paper: Bansal, Chahat, et al. “IndiaSat: A Pixel-Level Dataset for Land-Cover Classification on Three Satellite Systems-Landsat-7, Landsat-8, and Sentinel-2.” Proceedings of the 4th ACM SIGCAS Conference on Computing and Sustainable Societies. 2021. https://dl.acm.org/doi/pdf/10.1145/3460112.3471953
• Repo: https://github.com/ICTD-IITD/IndiaSAT_LULC_Deliverables

Tree cover monitoring

Forests and trees are the backbone to maintain our planet’s ecological balance, offering habitats, absorbing carbon dioxide, regulating climate, and sustaining biodiversity. Beyond traditional forests, small-scale plantations also play a vital role in rural livelihoods by enhancing agroforestry productivity and generating carbon credits. While restoration and conservation initiatives are underway, they require granular and timely insights into tree canopy health. However, tracking tree health indicators like canopy density and height over time has remained a challenging task, particularly in a country as ecologically diverse and vast as India. Modern remote sensing technologies allow us to move from static snapshots to near real-time, actionable insights, enabling stakeholders to implement timely interventions. We introduce a novel, scalable, and open-source framework for monitoring tree canopy density and height distribution across India using advanced remote sensing and machine learning techniques at a fine spatial resolution of 25m. 

We have developed Agro-Climatic Zone (ACZ) specific spatiotemporally robust models, which outperform global models by tailoring insights to local conditions, and use openly available data from the Sentinel-1, Sentinel-2 and Global Ecosystem Dynamics Investigation (GEDI) satellite systems The outputs of our model can especially help forest departments, policymakers, and local communities to identify areas in need of restoration and to assess the effectiveness of ongoing conservation activities.

• Research Paper: Dhruvi Goyal, Harsh Singh Chauhan, and Aaditeshwar Seth. 2024. Scalable Monitoring of Tree Canopy Density and Height in India at High Spatial Resolutions. In The 13th International Conference on Information & Communication Technologies and Development (ICTD 2024). https://www.cse.iitd.ernet.in/~aseth/forest-health-ictd2024.pdf 
• Github Repo: https://github.com/Dhruvi-Goyal/Tree-Health-Monitoring/ 
• GEE App: https://ee-ictd-dhruvi.projects.earthengine.app/view/tree-health-monitoring 

Building A Vulnerability And Socioeconomic Index

In the absence of regular censuses, socio-economic factors proxied using a wealth index can be obtained reasonably accurately from classifiers trained on satellite-based data. The figure below shows such a sample output from initial work by our group on poverty mapping. Detailed indicators related to the type of housing, primary source of drinking water, main source of lighting, source of fuel, asset ownership, etc. are aggregated into a composite wealth index at the village level, and tracked over a period of almost two decades from 2003 to 2019.

Research Paper

Tracking Socio-economic Development in Rural India Using Satellite Imagery. A. Gulgulia, A. Gupta, A.P. Sarashetti, A. Sinha, and A. Seth. ACM JCSS 2022. https://www.cse.iitd.ernet.in/~aseth/socio-econ-adi.pdf

GitHub-Repository

https://github.com/badri1995/ADI_complete

Impact assessment of water structures

Rainwater harvesting site assessment, for example, considers factors such as the soil type, slope, precipitation, soil moisture, land cover and land use. Remote-sensing data products of digital elevation maps and precipitation are now freely available. Machine learning-based land-cover classification algorithms that operate on freely available satellite imagery can provide high-resolution accurate maps, rapidly updatable. Soil moisture can similarly be estimated based on land surface temperature and vegetation indexes. We feel it is, therefore, feasible to conceive of a process whereby such datasets are used to identify candidate rainwater harvesting sites, fairness-based prioritizations are made to ensure the rights of vulnerable communities and participatory methods are then used to obtain community feedback and consensus to demand fund allocation for the identified sites.

Our research group conducts impact studies to understand and quantify the impact of Natural Resource Management (NRM) structures such as farm ponds and check dams, on agricultural productivity and drought sensitivity. Previously, we conducted a study on the site-level impact of MGNREGA farm ponds in AEZ 13 of India based on the Difference-in-differences (DiD) method. We found that on average farm ponds have a positive and significant impact on crop yield in the monsoon season. Further, we also found that farm ponds can help reduce sensitivity to droughts for up to three weeks of drought incidence. Building on these findings, we conducted another  study to understand the impact of farm ponds across the AEZs in India. In this study, we aim to understand the average treatment effect of farm ponds from a context-specific lens, based on the regional irrigation  infrastructure in terms of the borewell and canal presence, prevalent farm pond typologies, prevalent crop types, and so on. 

Further, we are implementing a Double ML (DML) based method to quantify the treatment effect of farm ponds across the different AEZs, and to build regional prediction models to predict the impact of prospective farm ponds based on their site-level features. We are also working on impact studies for check dams, wherein we have additionally worked on the identification of the zone of influence, which would give us more precise values for the treatment effect. Going forward, we plan to assess the landscape-level impacts of NRM structures

Research Paper

 Assessing the impact of farm ponds on agricultural productivity in Northern India. 2024. Ramneek Kaur, Kshitiz Bansal, Devang Garg, Ramita Sardana, Saketh Vishnubhatla, Sanjali Agrawal, Shruti Kumari, Parag Singla, and Aaditeshwar Seth. In Proceedings of the 7th ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies (COMPASS ’24) https://www.cse.iitd.ac.in/~aseth/farm-ponds.pdf

GitHub-Repository

https://github.com/SanyaKapoor/Site-Level-Impact-Assessment-of-Farm-Ponds

Farm boundaries

ince agriculture in most parts of India is rain-fed, global products like Dynamic World struggle to differentiate between agricultural land and scrubland. The error is evident in the areas of Surgana where a hilly but scrubby area, comprising a lot of fields in it, is classified mostly as scrubland. These classification errors hinder downstream tasks such as predicting cropping intensity on agricultural land. To address this, we developed a local compute pipeline that downloads high-resolution images at 1.19 m pixel resolution and applies computer vision models to segment the images into farmland, plantations, and scrubland.

Next, we designed an algorithm to classify these segments based on the insight that, unlike the irregularly shaped scrublands, agricultural fields exhibit either lower randomness—quantified by entropy measures—or structured planting patterns, which can be detected using techniques like the Hough transform. The resulting farmland, plantation, and scrubland boundaries are then converted into vector assets and uploaded to Google Earth Engine. Within our LULC pipeline, these vectors are used to refine downstream tasks, such as predicting cropping patterns exclusively within farm areas.

Research Paper

Coming soon.

GitHub-Repository

https://github.com/ramank1137/Scrubland-Field-Delineation

Detection of farm ponds and wells

Water resource management is critical for sustaining agriculture, particularly in regions prone to erratic rainfall and water scarcity. Traditional methods of monitoring water structures, such as farm ponds and wells, are often manual, labor-intensive, and inefficient for large-scale mapping. While satellite imagery has been used to study land use and hydrological changes, existing approaches struggle to accurately identify smaller water bodies across diverse landscapes. Addressing these challenges, we have built a deep learning-based framework to systematically detect, map, and analyze farm ponds and wells using high-resolution satellite imagery.

Our work integrates the YOLOv11 deep learning model with geospatial processing to develop an automated pipeline for identifying and geotagging water structures using high resolution satellite imagery and our model achieves high accuracy in detecting farm ponds and wells across different agroecological zones in India. The framework includes entropy-based filtering to reduce false positives, ensuring reliable detection across varied terrains. The geospatially processed results are converted into GeoJSON formats for seamless integration with GIS platforms, facilitating large-scale impact assessment. This automated approach has the potential to enable better resource planning, improve irrigation infrastructure monitoring, and support data-driven decision-making for sustainable water management in rural areas.

We are now using this output to build an irrigation profile of villages. 

Example: Visualization of the outputs: farm ponds and wells

Research Paper

Coming soon.

GitHub-Repository

https://github.com/aatifnisar01/Ponds_and_wells_detection.git

Fairness In Funds Allocation

The objective of this research is to conduct an analysis of fairness of expenditure across different locations in the Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGS). We build a framework and conduct an (almost) pan-India analysis using publicly available MGNREGS data to determine disparate treatment and disparate impact of different types, towards SC/ST groups, and uncover the prevalence of different processes through which this discrimination may be operating. The results can be used to define precise policy actions at the state, district, and panchayat levels, to counter this inequity. Further, even the identification of disparity in welfare allocation can be used to trigger corrective action.

We introduce disparate treatment and impact metrics for MGNREGS related to the efficiency of scheme coverage, intensity of scheme uptake, and scheme demand. We also build a unique characterization of panchayats (local administrative units) in terms of whether SC/ST population is dominant in the panchayat, and distributed across the villages of the panchayat or concentrated in a few villages or most villages. We have developed an inequity index as an indicator of inequity in fund allocation at panchayat, district, and state levels. The district-level inequity indices for scheduled caste and scheduled tribe are presented on the map where a value 1 indicates less inequity and 5 indicates a high level of inequity. Analysis of treatment and impact metrics in these different categories of panchayats helps point towards different processes through which discrimination could be operating, such as through local elite capture by dominant castes or through administrative apathy at higher bureaucratic levels. 

Ultimately, the research aims to develop a model to estimate an optimal allocation of funds to promote equitable development among the panchayats. The model will take into account various factors such as the socio-economic and demographic characteristics of each panchayat, its developmental needs, and historical funding patterns to ensure that the allocation of funds is fair and just.

Research Paper

Ashima Mittal, Chintan Sheth, and Aaditeshwar Seth. 2024. An In-depth Analysis of Caste-based Inequity in Welfare Allocation in the National Rural Employment Guarantee Scheme in India. The 13th International Conference on Information & Communication Technologies and Development (ICTD 2024), December 09–11, 2024, Nairobi, Kenya. https://www.cse.iitd.ernet.in/~aseth/caste-based-ictd2024.pdf

GitHub-Repository

https://drive.google.com/drive/folders/1Ouhxz_JELRiboXDXzlc2ZG1Q935AN7uB?usp=drive_link (for welfare metrics at Panchayat, District, and State-level)

https://drive.google.com/drive/folders/1tm8EHY3mClWOsuV5obGBXtr5kqBhYxCA?usp=drive_link (for MIS parameters)

Site assessment for water structures

The Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA) was enacted in 2005 to provide 100 days of wage employment per year to rural households willing to perform unskilled manual labor. The labour is specifically meant to be towards the creation of durable assets such as farm ponds, check dams, percolation tanks, and irrigation channels to strengthen water security in the rural landscapes. 

Site selection for the construction of water assets usually takes into account the proximity of the site to farmlands, but may neglect geological factors such as the presence of lineaments or the infiltration rates and storage capacity of the underlying lithology. This can lead to the selection of improper sites. Such misidentification of suitable sites indeed results in many non-operational structures. To improve the site selection with scientific evidence, a tool called Composite Land Assessment and Restoration Tool (CLART) was developed by the CSO, Foundation for Ecological Security (FES), which provides an assessment of areas suitable for good, moderate, or low groundwater recharge. CLART primarily uses geological layers based on lithology, lineaments, drainage density, and slope, to assess this site suitability. Demands keyed into CLART are associated with a recharge suitability score and conveyed to MGNREGA or other schemes for allocations.

Experienced CSO field staff and volunteers also indicated that in addition to seeing the recharge potential at a particular site, as indicated by the CLART layer, the exact slope at that point and a comparison with the surrounding slope, would be useful to judge runoff accumulation capacity for the site. We plan to provide site-level assessment in our Commons Connect planning tool that incorporates the CLART layer and augments site assessment further by using site-level indicators to corroborate community demands. Site-level indicators that could be useful include an estimate of the catchment area draining into the point, identify natural depressions for constructing potential waterbodies,  distance to the nearest drainage line, areas under the influence of various stream orders and the sensitivity of runoff accumulation to drought and non-drought years. Feedback on our ongoing work on determining the impact of new water bodies on downstream water bodies  using the simulation method was also found to be quite relevant during field testing of Commons Connect tool.

Research Paper

• Shivani Mehta, et al. Initial Observations from Field Testing of a Digital Participatory Tool to Improve Water Security in Rural India. ICTD 2024, Nairobi, Kenya. https://www.cse.iitd.ernet.in/%7Easeth/commons-connect-ictd2024.pdf

GitHub-Repository

Coming soon.

Hosting:

Site-level layer	Spatial resolution	Temporal resolution	Google drive folder	Github repository	Spatialunit ofavailability /computation
CLART	30m	static	–	Programmable CLART	Computed at pixel level
Upstream-downstream water bodies	–	annual	–	Upstream-downstream water bodies	Computed at block level
Stream order raster	30m	static	Stream_Order_Raster_India.tif	–	Pan India
Distance to upstream drainage line	30m	static	xxx	–	Pan India
Catchment area	30m	static	xxx	–	Pan India
Naturaldepressions	30m	static	xxx	–	Pan India
Runoff accumulation	30m	static	–	xxx	Computed at pixel level

Site assessment for tree plantations

This research work focuses on building a participatory tool for assessing the land suitability for tree plantations and further recommend the best suitable tree species for the selected site that caters to the community’s environmental & economical needs. It leverages multiple soil, terrain, ecology, climate and socio-economic landscape characteristics of the region to generate the plantation feasibility map. The data for all these characteristics are derived by processing high resolution open-source satellite imageries using remote sensing and machine learning techniques. We currently take weights of different features as input from the users/collaborators in their area of interest and do a weighted aggregation of the considered features to propose the site suitability for agroforestry plantation, based on the methodology described in (Nath et al. 2021). 

We are also building a model to assess the plantation’s health over time, by estimating the vegetation density, growth in canopy height and carbon sequestered with time. This would facilitate the farmers, NGOs, government officials and other organizational bodies to analyze the impact of the undertaken tree plantation projects in different sites. 

Additionally, we are making a similar attempt to identify sites requiring restoration support, and creating Species Distribution Models (SDMs) that can help us assess the survival rate of species at these sites based on the local (and projected) climatic and ecological characteristics.

Research Paper

Coming Soon.

GitHub-Repository

https://github.com/Dhruvi-Goyal/Plantation-Site-Suitability