Impact of Look-Back Period on Soil Temperature Estimation Using Machine Learning Models
Tomislav Kovačević, Lovre Mrčela, Andro Merćep, Zvonko Kostanjčar · 2020
Temperature is one of the most important properties of soil. It affects plant growth, germination, nitrification, and suitable planting and harvesting dates in an agricultural production process. Installing weather stations on every micro-location of interest significantly increases production costs. A cheaper alternative is to estimate soil temperature from existing weather stations in the same broader area and weather data available from application programming interfaces. Although different machine learning models have been developed for the purpose of soil temperature estimation using weather data, there is lack of knowledge about impact of lagged weather data, so-called look-back period, on model performance. In this paper, we quantify the impact of look-back period on the estimation of soil temperature using different machine learning models. As it turns out, the root mean squared error for all tested models drops significantly with p-value less than 0.01 as the look-back period increases.