Comparison of Different Spectral Vegetation Indices for the Remote Assessment of Winter Wheat Leaf AreaIndex in Mashhad

The role of leaf area index (LAI) in terrestrial ecosystems is undeniable. LAI affects the amount of carbon, water and energy metabolism. Also, many agronomic, environmental and meteorological applications require information on the status of LAI. The time series of the spectral indices obtained from the remote sensing indicates its usefulness in detecting regional-scale LAI changes. So, the desire for the development of models for estimating LAI was increased with using satellite images. Vegetation Indices (VIs), especially the Normalized Difference Vegetation Index (NDVI) and Ratio Vegetation Index (RVI), are most widely used. According to the different sensitivity of VIs to the value of LAI and vegetation characteristics, in this study, we tried to determine an algorithm with a higher accuracy to estimate the LAI of wheat using more variables (VIs). In this study, regarding the wheat growth period in Astan Quds Razavi (AQR) farms, the Landsat 8 satellite images were used from November 22, 2014 to June 20, 2015. LAI was measured simultaneously with passing of Landsat 8 (16-day intervals) from AQR Fields of Mashhad (in five dates from 17 farms) during wheat growing season in 2014-2015.
After pre-processing of satellite images, VIs including the Difference Vegetation Index (DVI), NDVI, RVI, Transformed Difference Vegetation Index (TDVI), Soil Adjusted Vegetation Index (SAVI), Infrared Percentage Vegetation Index (IPVI), Greenness Index (G1 and G2) and Enhanced Vegetation Index (EVI1 and EVI2) were calculated. To select the best variables and the equation for estimating LAI, simple regression (linear, quadratic and exponential) and multiple linear regression (Backward and Forward) methods were used. Finally, to validate and assess the accuracy of the presented models, the mean square error (RMSE), Mean Absolute Error (MAE), Point accuracy based on a percentage of actual value (E%) and correlation coefficient (r) was used. The results of this study showed that simulation of LAI based on the existing equations in the references using the NDVI, EVI1 and EVI2 indices extracted from Landsat 8 satellite images has low accuracy (RMSE:2.71, 3.65 and 3.65). This confirms the necessity of examining and calibrating equations. The results indicate that the accuracy of the wheat LAI estimation by using the NDVI and SAVI index was increased by exponential functions (RMSE:1.18 and 1, respectively) compared to the linear model (RMSE:1.46 and 1.26, respectively). This increase was due to a more accurate estimation LAI lower than 4 and the fixed value of LAI simulated in a range of actual LAI higher than 6. The accuracy of LAI estimation was increased with combination of two VIs (NDVI and SAVI) compared to the linear model of each index separately. Also, the highest accuracy of LAI estimation from the combination of G2 with SAVI and EVI1 (RMSE: 1.03, 1.03, respectively) was observed due to the higher sensitivity of G2 to medium and high LAI compared to NDVI. In addition, the backward and forward regression model was improved the accuracy of wheat LAI estimation compared to other models, due to the greater sensitivity of this model to LAI higher than 6 (RMSE: 0.87 and 0.95, respectively). Although the accuracy of wheat LAI estimation by the forward regression model was higher than all models, but its calculation requires the use of many parameters. Since LAI is an important biophysical parameter in ecological modeling. Accurate and fast estimation of this parameter in large scale for ecological models such as yield and evapotranspiration, and carbon exchange is very important. Considering the results of this research and the opinions of other researchers, it can be stated that the accuracy of the exponential functions and multiple linear regression (Forward regression model) for estimating LAI was higher than simple linear regression.