Session: DAC-12-01 Engineering for Global Development

Paper Number: 70571

Start Time: August 17, 03:20 PM

70571 - Machine Learning Method for Forecasting Weather Needed For Crop Water Demand Estimations in Low-Resource Settings Using A Case Study in Morocco 

Developing countries have low crop productivities which can be due, in part, to a lack of technology. These countries also often do not have the infrastructure needed to support weather forecasting models, which are computationally expensive and require detailed inputs from local weather stations to be accurate. Local, low-cost weather prediction services are needed to enable optimal irrigation scheduling and increased crop productivity for rural farmers in low-resource settings. This work proposes a machine learning approach to predict the weather inputs needed to calculate crop water demand, namely evapotranspiration and precipitation. The focus of this work is on the accuracy with which Moroccan weather can be predicted with a vector autoregression (VAR) model compared to using typical meteorological year (TMY) weather, and how this accuracy changes as the number of weather parameters is reduced. The results showed that the VAR model outperformed using the TMY data for both the full data sets and the reduced data sets. This indicates that using a machine learning model with few low-cost sensors could be an alternative to using historical weather when predicting weather needed for evapotranspiration calculation. Also, using this machine learning method to increase the accuracy of the weather prediction could lead to an increase in crop productivity.

Presenting Author: Carolyn Sheline MIT

Authors:

Carolyn Sheline MIT
Amos V. Winter Massachusetts Institute of Technology