Development of tobacco (Nicotiana Tabaccum) yield estimation models using agronomic and remote sensing techniques

Abstract

Farmers need to monitor crop growth and development and obtain early estimates of final yield. The unavailability of a comprehensive method for estimating crop yield leads to contradicting estimates, subjective national statistics and general planning inefficiency by stakeholders. In this study, experiments were conducted to select a suitable index for assessing varietal, planting date and fertilizer management influence on tobacco canopy reflectance. A hand-held multispectral radiometer was used to take canopy reflectance measurements. This was followed by an investigation into the relationship among canopy reflectance, in-season dry mass and final crop yield. The experiments were conducted at Kutsaga Research Station, near Harare in Zimbabwe. The MODIS satellite derived NDVI was used to assess tobacco growth, estimate crop area and final yield. The relationship between reflectance measurements from the multispectral radiometer and those from the MODIS satellite were used in up-scaling the multispectral radiometer derived yield estimation models for application on the sampled tobacco fields within a radius of 150 km from Harare. In this study, it is demonstrated that although simple ratio index (SRI) had a stronger relationship with biophysical parameters such as above-ground dry mass, plant population and plant height than NDVI, the latter was selected for use because of its stronger relationship with total nitrogen. Varieties, planting dates, and fertilizer application levels could be separated using spectral data between 9-12 weeks after planting. Thedifferent planting times could be separated from 0 to 9, 10 to 12,13 to 18 and 18 to 22 weeks after planting, thus demonstrating these as the optimum period for collecting spectral data for tobacco yield estimation. The mass at untying-NDVI regression coefficient of determination decreased with later planting from September (R2 = 0.79), October (R2 = 0.64), November (R2 = 0.695) and finally December (R2 = 0.515). The yield-NDVI regression models for the September and the October-planted crops were statistically similar (p = 0.424), and so were those for the November and December planted crops (p = 0.541). There were no significant differences (p = 0.220) among the mass at untying - NDVI regression curves for K RK 26, T 66 and K E1 and for the fertilizer application levels (p = 0.167). Since the relationships among tobacco in-season dry mass and yields with NDVI were not affected by tobacco variety and fertilizer application levels, a combined model for estimating tobacco yield using NDVI was developed. v Using remote sensing based on the MODIS satellite derived NDVI data, the third to fourth week of November and the third to fourth week of February were the optimal times for discriminating the September-October from the November-December planted tobacco. The tobacco crop areas for the 2010/ 2011, 2011/ 2012 and 2012/2013 cropping seasons were estimated, and yield estimates were calculated from the long-term cropped yield- area regression model. An up-scaling factor from the multispectral radiometer derived model to the MODIS derived model was developed, and a model for estimating tobacco yield using NDVI was derived. A regression analysis of the observed versus predicted yield was significant (p<0.05). The results show that tobacco yield can be estimated from the MODIS satellite derived NDVI using the model: Ytot = A(48.28 *av NDVIMOD2 – 37.51*av NDVIMOD+ 8.003). It is recommended that the model be used by tobacco industry to complement existing methods.