Estimating rice yield from paddy rice planting areas in Bukedi sub region using Sentinel 2

Abstract

Rice is a very important crop in the diet of Uganda and the world at large yet its production is still limited and its growth in paddy fields is a threat to the environment and biodiversity. This research project investigated the use of the phenology-based approach of rice growth to extract and map these paddy rice fields using Sentinel 2 satellite imagery in addition to estimating the paddy rice yield at Kibimba Rice scheme using the Monteith model. The terrain mask was applied on the mosaiced images of different months after which other permanent water, urban areas, forest and natural wetland masks were also executed. The mosaiced all masked image was used to mask all the mosaiced processed sentinel 2 images after which the results were used to carry out the classification. Paddy rice maps were produced with classification accuracy varying from 79.69% to 87.5% with a mean of 83.985% and kappa statistic(K^) varying from 0.7679 to 0.8527 with mean 0.8114. high accuracy in remote sensing estimated area of 97% was observed at Kibimba on 16 June 2018 a minimum of 73% on 15 August 2018 at Doho rice scheme which was attributed to water masking that caused gaps in some parts of the scheme. The maps were then be used to extract the NDVI of paddy rice areas during the heading stage. An NDVI of 0.7, a harvest index of 0.25, LUE of 1.8gMJ-1 and an average PAR coefficient of 1353WM-2 was used in the Monteith model to determine the estimated rice yield at Kibimba Rice Scheme. The yield results were tested and validated using yield statistics from Tilda Uganda Limited/ Kibimba rice scheme and it was found out that the model yielded results which had a high correlation with the actual Kibimba statistics with a correlation determinant R2 =0.9882 and correlation coefficient R=0. 994.