Measuring Cotton Water Status using Water-Related Vegetation Indices at Leaf and Canopy Levels

Drought is one of the major environmental threats in the world. In recent years, the damage from droughts to the environment and economies of some countries has been extensive, and drought monitoring has caused widespread concerns. Remote sensing has a proven ability to provide spatial and temporal measurements of surface properties, and offers an opportunity for quantitative assessment of vegetation properties at different levels. Together with other parameters, vegetation water content is an important indicator of drought that can be investigated by using remotely sensed data. The most widely used water-related parameters in remote sensing are the leaf equivalent water thickness (EWT_leaf), the canopy equivalent water thickness (EWT_canopy), the fuel moisture content (FMC) and vegetation water content (VWC).

A number of different indices have been developed for the estimation of vegetation water content. Spectral indices are the most widely used technique among the above mentioned methods. Besides, numerous methods have been developed to estimate water content from reflectance data. Most previous approaches for vegetation water content estimation use the linear regression technique (LR). In recent years, quantitative remote sensing of vegetation biochemical has been greatly improved by the use of multivariate statistical methods, particularly Artificial Neural Network (ANN). Some comparison studies between regression statistical technique and neural networks have been conducted by many researchers using various datasets. However, the predictive ability of ANN for water content estimation has not been well demonstrated.

Based on the above-mentioned scientific questions, researchers from Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, used the leaf and canopy hyperspectral measurements to analyze the relationship between remote sensing indices and EWT, FMC, VWC, and then discussed the method for improving the accuracy in retrieving water parameters at leaf and canopy levels. The objective of this research is to explore further potential of NIR (near-infrared reflectance), SWIR (shortwave-infrared reflectance) wavelengths to estimate FMC, EWT_leaf, VWC and EWT_canopy using leaf and canopy hyperspectral reflectance, and compare the performance of LR method and ANN technique for vegetation water content deriving based on remotely sensed water-related vegetation indices. In this study, sites of cotton field in Shihezi, Xinjiang were sampled. Four classical water content parameters, EWT_leaf, FMC, EWT_canopy and VWC were evaluated against seven widely-used water-related vegetation indices, namely the NDII (normalized difference infrared index), NDWI₂₁₃₀ (normalized difference water index), NDVI (normalized difference vegetation index), MSI (moisture stress index), SRWI (simple ratio water index), NDWI₁₂₄₀ (normalized difference water index) and WI (water index), respectively.

The results proved that the relationships between the water-related vegetation indices and EWT_leaf were much better than that with FMC, and the relationships between vegetation indices and EWT_canopy were better than that with VWC. Furthermore, comparing the significance of all seven water-related vegetation indices, WI and NDII proved to be the best candidates for EWT detecting at leaf and canopy levels, with R² of 0.262 and 0.306 for EWT_leaf-WI and EWT_canopy-NDII linear models, respectively. Besides, the prediction power of LR and ANN were compared using calibration and validation dataset, respectively. The results indicated that the performance of ANN as a predictive tool for water status measuring was as good as LR. The study should further our understanding of the relationships between water-related vegetation indices and water parameters.

The result has been published on Journal of Arid Land, 2012, 4(3): 310-319. The paper is also archived at http://jal.xjegi.com/EN/abstract/abstract152.shtml.