Estimation of leaf water content from mid- and thermal-infrared spectra by coupling genetic algorithm and partial least squares regression

Abstract

Leaf water content (LWC) is an essential constituent of plant leaves that determines vegetation health and its productivity. An accurate and on-time measurement of water content is crucial for planning irrigation, forecasting drought, and predicting woodland fire. The retrieval of LWC from visible to shortwave infrared (VSWIR: 0.4 to 2.5 μm) has been extensively investigated but little has been done in the mid- and thermal-infrared (MIR and TIR: 2.50 to 14.0 μm) windows of electromagnetic spectrum. This study is mainly focused on retrieval of LWC from MIR and TIR, using genetic algorithm (GA) integrated with partial least square regression (PLSR). GA fused with PLSR selects spectral wavebands with high predictive performance, i.e., yields high adjusted-R2 and low root-mean-square error (RMSE). In our case, GA-PLSR selected eight variables (bands) and yielded highly accurate models with adjusted-R2 of 0.93 and RMSE cross validation equal to 7.1%. This study also demonstrated that MIR is more sensitive to the variation in LWC as compared to TIR. However, the combined use of MIR and TIR spectra enhances the predictive performance in retrieval of LWC. The integration of GA and PLSR not only increases the estimation precision by selecting the most sensitive spectral bands but also helps in identifying the important spectral regions for quantifying water stresses in vegetation. The findings of this study will allow the future space missions (like HyspIRI) to position wavebands at sensitive regions for characterizing vegetation stresses.