Forecasting yields of processing potatoes in different South African environments using the LINTUL model

Abstract

The potato processing industry relies on yield forecasts for production planning and accurate yield forecasting is a challenge. Expected yield forecasts need to be produced as early as eight weeks before harvest and delivery of potatoes to the factory. A simple crop growth model, LINTUL-Potato, was tested to see if it can accurately forecast yields of potato using long term historical and current weather data as well as crop and management input data. This study looked at calibration and validation of the model in different production areas in South Africa. Three different potato varieties were used in the study. The LINTUL-Potato model uses the linear relationship between biomass production and light use efficiency of solar radiation intercepted by the crop’s canopy. The model was developed for use in potato production systems and this study focused on adapting the model for use within a yield forecasting system for a potato processing company. The aim of this study was to explore how the LINTUL-Potato model can be used in a yield forecasting system. The hypotheses tested were that a simple model can be used to accurately forecast yields of potato using long term historical and actual weather data, the tuber count and size (mass) of potatoes can be accurately forecasted using a simple model, and that the dry matter content of potato tubers can be forecasted using a simple model. The other objectives of the study were to test the accuracy of the yield forecasts produced, monitor evolution of tuber size distribution for use in forecasting the size composition of the yield and investigate potential use of the model for determining and forecasting dry matter content of tubers. The LINTUL model used in this study is a simplified version, hence its parameter values needed to be verified through calibration. To find out if the parameterized model worked well, it had to be validated by obtaining and using independent crop, management and weather data. After calibration, the model was then used for yield forecasting using real-time weather data until the sampling date and then using long-term historical weather data until the expected harvest date. The hypotheses that a simple model can be used to accurately forecast yields of potato using long term historical and actual weather data, and that the tuber count and size (mass) of potatoes can be accurately forecasted using a simple model were accepted and the forecasts produced were reasonably accurate. Results of the model validation showed that accurate forecasts can be produced early in the growing season using current actual weather data up to 65 days after planting and then long term weather data for the remainder of the season up until harvest. LINTUL model performance was evaluated using the r, rs, R2, MAE, RMSE and D-index statistical parameters. Validation of the model for the variety Innovator had a r value of 0.80 (R2 = 0.64) and a strong positive association was observed between the forecasted and observed final yields with rs = 0.56. Ratios based on the actual and attainable yields were calculated for three different varieties and these were adapted to be used in the forecasting system. The adapted ratios were 0.61 for Innovator, 0.70 for Markies and 0.78 for Pentland Dell. The varietal ratios were used to calculate the actual expected yield from the LINTUL forecasted yield at final harvest. For tuber size distribution, the average tuber mass 10 kg-1 sample at final harvest correlated well (r = 0.87) and a strong positive association was observed (rs = 0.60). The average tuber No 10 kg-1 sample was forecasted with lower accuracy (R2 of 0.56) but correlated well with r = 0.75. A strong positive association was also observed (rs = 0.64). A T-test was also used to compare the forecasted and observed fresh tuber yield, tuber No 10 kg-1 and average tuber mass 10 kg-1 at final harvest. The results showed non-significant differences for fresh tuber yield, average tuber count and average tuber mass per 10 kg at final harvest at the 5% significance level. The hypothesis that the dry matter content of potato tubers can be forecasted using a simple model was rejected. The dry matter composition of tubers was influenced by a number of external factors and forecasting was not possible in this study and requires a further study.