JIMS8I International Journal of Information Communication and Computing Technology

Agriculture is one of the main sources of livelihood for the rural people of India. Due to the lack of irrigation facilities in most parts of the country, farmers still practice a rice-based farming system that is monsoon dependent. Aman Paddy is majorly grown as a Kharif crop all over West Bengal during the months July to November and is a monsoon dependent crop. This report aims to perform a comparative study for estimating paddy yield using two different approaches for 2017, 2018, and 2019 in the South 24 Parganas district of West Bengal. NDVI based crop classification and yield estimation have been considered the most effective way for crop yield analysis. Two approaches used to estimate the Aman paddy yield using NDVI threshold values and remote sensing techniques for the South 24 Parganas district were: (i) the time series method (ii) a single timestamp approach. A strong positive correlation was established using both the approach with R² = 0.63-0.79. The paddy yield estimated using both the methods showed a maximum deviation of 200kg/ha, 250kg/ha, and 300 kg/ha for 2017, 2018, and 2019, respectively in the said district. After running the linear regression, the paddy yield thus estimated has been validated for accuracy using statistical technique. k-fold cross-validation technique has also been followed using R, for calculating the RMSE, RSquared, and MAE metrics. It is known that the lower the values for statistical metrics, the better is the model. The RSquared values for both the models are found to be 0.99 which proves that both the models are a good fit. However, the RMSE and MAE for the time series approach are lower than the single timestamp approach thereby validating that the time series approach is a better way to estimate the paddy yield. In other words, NDVI based time series analysis seems effective for paddy yield estimation.