ANNOUNCEMENTS
The agrarian economy in India, which consumes 78% of the nation’s freshwater, faces significant challenges due to acute water stress. This situation is compounded by the low efficiency of irrigation systems, which currently operate at just 38%. Additionally, approximately 54% of the agricultural area is under water stress, highlighting an urgent need for improved water management. Addressing these issues is critical, particularly for the 89% of farmers who are small and marginal, as they are most affected by these inefficiencies. Climatesmart agriculture offers substantial promise, with the potential to save up to 300 billion cubic meters of water globally each year. In India, adopting such technologies could help achieve the Bureau of Water Efficiency’s goal of a 20% reduction in water use, underscoring the need for innovative solutions.
This research proposes a climate-smart digital agriculture solution designed to tackle these irrigation management challenges integrating regional machine learning models with multi-stakeholder platform based agromet advisory services. At the macro level, the research constructs two indices: Digital Agriculture Potential Index (DAPI) and Digital Irrigation Potential Index (DIPI). These indices are designed to identify states with high potential for digital agriculture and irrigation investments, guiding the strategic allocation of resources to optimize interventions in these areas. While DAPI and DIPI facilitate the optimization of digital agriculture and irrigation investments across states, climate regionalization plays a critical role in enhancing state-level initiatives. By classifying states into climatic zones using advanced spatiotemporal models, region-specific challenges such as evapotranspiration (ET) demands can be effectively addressed, allowing for tailored strategies that align with local climatic conditions.
The research introduces a novel climate regionalisation framework that utilizes spatio-temporal clustering techniques using 21 years (2000–2020) of daily gridded hydroclimatic data across cropping seasons Kharif, Rabi and Zaid. This framework applies advanced spatio-temporal consensus clustering methods, which offer greater robustness than single-algorithm clustering techniques. It also uses a 3-stage climate region validation process with internal validation, evaluative comparison and domain based external validation. The evaluative comparison is performed against spatio-temporal consensus clusters derived from Penman-Monteith-based ET0 monthly statistical characteristics to ensure accurate regional climate assessments.
The delineated climatic regions, defined through spatio-temporal consensus clustering, provide a robust framework for the development of accurate and scalable regional machine learning models for accurate estimation of reference evapotranspiration (ET0). These machine learning models for ET0 estimation can leverage the spatial and temporal patterns captured in the clustered climatic zones to enhance prediction precision and applicability across diverse agro-climatic contexts. Regional reference evapotranspiration (ET0) models for each cropping season and spatio-temporal cluster within are developed using machine learning models, applied on limited daily gridded hydroclimatic variables including features such as month of the year and day of the year. A robust machine learning framework has been implemented, involving five models, two single algorithms and three hybrid algorithms for ET0 estimation. Extreme Gradient Boosting (XGBoost) serves as a key model, utilizing K-Fold Cross-validation (K=5 and K=10) for validation, with its hyperparameters fine-tuned via OPTUNA, which employs the Bayesian hyperparameter optimization (BHO) technique. The 2-Model Stacked Ensemble Meta-Learning architecture integrates XGBoost and Random Forest (RF) as base models, with Multiple Linear Regression (MLR) and MLPRegressor (Deep Learning model) , as meta-models. For one of these hybrid models, MLR is used as the meta-model, adopting both K=5 and K=10 configurations to analyze the influence of K-fold cross-validation in a manner like XGBoost. Hyperparameter optimization for hybrid models is also carried out using Bayesian Optimization technique with OPTUNA framework. Furthermore, SHAP (SHapley Additive exPlanations) feature importance analysis and BEESWARM plots are employed to explain the results of the machine learning models.
Different models provided the best accuracy across various cropping seasons and climatic regions, with distinct rankings of predictor variables in each case. This indicates that the performance of the models and the importance of specific predictor variables vary depending on the temporal and geographical context, highlighting the need for tailored approaches in digital agriculture and irrigation that account for seasonal and regional differences. For the 5 champion models out of the 25 models built across cropping season and spatio-temporal clusters, the accuracy, measured with Kling- Gupta Efficiency (KGE) ranged between 0.83 - 0.97.
At the micro level, the solution addresses challenges of digital innovation adoption in agriculture by proposing a multi-stakeholder platform framework underpinned by an adoption survey of 150 farmers in two districts of Haryana State across multiple blocks. The objective is to ensure that digital solutions are farmer-centric and foster sustainable technology engagement. A novel Digital Innovation Adoption Assessment Framework (DIAAF) was developed to design the survey questionnaire and guide the evaluation process. The survey covers several analyses, including Need Assessment, Perceived Usefulness (PU), Willingness to Pay (WTP), adoption category analysis, homogeneity, farmers’ affinity towards agromet advisory services and factors influencing adoption. These analyses provide critical insights into the essential components of a microlevel strategy for successful digital adoption and the design of digital applications. This comprehensive approach bridges the gap between macro-level potential and micro-level needs, offering a robust, cost-effective model and agromet advisory strategy to promote climate-smart agriculture in India.
Keywords: Climate Smart Agriculture, Digital Agriculture Potential, Digital Irrigation Potential, Spatio-temporal clustering, PCA – DEA, Meta-clustering, Consensus Clustering, Machine Learning and Deep Learning, Multi-stakeholder platform, Adoption Survey.
