Irrigated lands generate crop yields that are more than double those of rain-fed lands. Unfortunately, these systems are often heavily reliant on water supplies, which are diminishing globally. Alternative use of impaired quality waters for irrigation can reduce soil quality through secondary salinization, affecting plant health and yields. With salinization of agricultural lands increasing around the world, further understanding the impacts of this on crop production are required. The aim of this research is to assess the influence of soil salinity and nutrient stress on leaf photosynthetic pigments, gas exchange and biochemical photosynthetic parameters in wheat plants. The feasibility of estimating key photosynthetic pigments from in-situ leaf hyperspectral data is examined using vegetation indices, linear regression models and a random forest machine learning technique.
Results showed that salinity stress presented a significant increase in the chlorophyll and carotenoid contents per leaf area, although the total pigment contents per plant was reduced as a consequence of lower production of leaf matter. While nutrient application enhanced the photosynthetic pigment content per leaf area, its interaction with salinity stress was found to be significant and varied with salinity level. A strong positive relationship was found between SPAD-502 measurements and leaf chlorophyll content and confirmed that SPAD-based retrieval of photosynthetic pigments can be undertaken with confidence irrespective of any prevailing stress in wheat plants. Photosynthetic parameters directly related to biomass accumulation (such as Vcmax, Jmax and gs) varied considerably with stress levels and growth stages, with high values of these parameters observed at low stress and in periods of more vigorous growth. Employing a random forest machine learning approach with all hyperspectral data as input features significantly improved the predictability and accuracy relative to the univariate linear regression model. However, using vegetation indices as direct predictors further improved the estimation accuracy and robustness of the random forest model.
Overall, the findings from this research have implications for large scale estimation of vegetation photosynthetic traits from remotely sensed data, and offer a mechanism by which early detection of stress may be monitored, providing a means for enacting a timely crop management response.
|Date of Award||Feb 2019|
|Original language||English (US)|
- Biological, Environmental Science and Engineering
|Supervisor||Matthew McCabe (Supervisor)|
- Machine Learning
- Plant Stress
- Soil Salinity