TURNING FIELD, SOIL, WEATHER AND CROP DATA INTO A PREDICTIVE CULTIVATION PLANNING PIPELINE FOR SMARTER AGRICULTURAL DECISIONS - (FARM2FORECAST)

Abstract

Farmers in India make most of their cultivation decisions, such as which crop to sow, when to sow it, how much fertiliser to apply, and when to irrigate, on the basis of experience and advice passed on informally. These decisions are becoming harder because rainfall has grown less predictable, soil fertility varies considerably between neighbouring fields, and market prices change from season to season. This paper presents Farm2Forecast, a data pipeline and predictive analytics system that combines field records, soil test results, weather observations and forecasts, and historical crop data to support cultivation planning for small and medium farms. The system collects data from several sources. Field details such as location, area, irrigation source, and past crops are registered by farmers or extension workers through a mobile application. Soil Health Card results provide nitrogen, phosphorus, potassium, pH, organic carbon, and micronutrient values. Daily weather observations and medium-range forecasts are fetched from public meteorological services, and district-level crop yield and market price records are loaded from government open data portals. A batch and incremental ETL pipeline cleans these data, aligns them to a common field and season key, and stores them in a partitioned analytical warehouse. Three models are built on the integrated data. A crop suitability model ranks candidate crops for each field and season based on soil, climate, and water availability. A yield prediction model, using gradient boosting on soil, weather, and management features, estimates the expected yield of the recommended crops. A sowing window model combines forecast rainfall and soil moisture estimates to suggest the most favourable dates for sowing. A rule-based nutrient module converts soil test values into fertiliser recommendations, and an economic layer uses expected yield and recent prices to estimate the likely return from each option.