Abstract—For the purpose of supply chain management, a crop producer is required to forecast and to forward the information of potential harvest to his customers, such as food processors or wholesale channels. Usually, such forecast relies on the field observation done by experienced producers with domain knowledge for growing crops under different climates. In this paper, we propose a computational approach for cucumber harvest forecasting, which is based on the partial least square (PLS) regression over historical environment and harvest information. The experiment is performed in a 40×8m2 net house consisting 36 tiny farms for collecting the harvest data, and 68 environmental sensors for measuring the illumination, the air temperature, the air humidity, the soil temperature, and the soil moisture. The area of each tiny farm is about 8m2, which is planted with 16 seedlings of organic cucumbers. The harvest and the environmental data are collected with the Smart Agro-management Platform (SAMP) developed by the Institute for Information Industry (III). According to the experimental results, the forecast for the accumulative harvest in the future 4 to 6 days achieves accuracy 70% in split testing, which is close to the averaged forecasting performance of experienced producers. Therefore, the environmental data during the harvest season serves as good factors for building harvest forecasting models.
 

Index Terms—cucumber, data science, environmental factors, harvest forecasting

Cite: Yung-Hsing Peng, Chin-Shun Hsu, and Po-Chuang Huang, "Harvest Forecasting with Environmental Information for Cucumbers Cultivated in Net Houses," Journal of Advanced Agricultural Technologies, Vol. 2, No. 2, pp. 83-87, December 2015. Doi: 10.12720/joaat.2.2.83-87