Random forests as a tool for analyzing partial drought stress based on CO2 concentrations in the rootzone of longan trees.

Abstract

This study aims at establishing a relationship between water supply and CO 2 concentration in the rootzone, and to identify disturbing factors using data-driven modelling. In our previous study, 10 longan trees were planted in split-root technique and kept under controlled conditions. During six months, 5 trees were partially irrigated on one side of the root system, while the other side was kept non-irrigated. The sides were switched in a two-week interval. Five control trees received full irrigation on both sides. Monitoring results on CO 2 concentration in the rootzone, soil moisture and stomatal con-ductance indicated a weak correlation between the CO 2 concentration in the rootzone and the soil moisture, but without a statistically significant correlation, partially because air temperature was a main disturbing factor. In this study, Random Forests was applied to establish a CO 2 -water stress relationship based on air temperature, relative humidity, vapour pressure deficit and soil moisture. It was shown that the most important factor on CO 2 concentration in the rootzone was soil moisture, followed by air temperature. Together with the information retrieved, the results suggest a potential of CO 2 monitoring in the rootzone for assessing plant water status continuously and with a minimum level of invasion.