A novel dynamic stretching solution to eliminate saturation effect in NDVI and its application in drought monitoring

The normalized difference vegetation index (NDVI) is one of the key input variables for developing drought indices. However, the NDVI quickly saturates in high vegetation surfaces, and thus, the generalization of a drought index over different ecosystems becomes a challenge. This paper presents a novel, dynamic stretching algorithm to overcome the saturation effect in NDVI. A scaling transformation function to eliminate saturation effects when the vegetation fraction (VF) is large is proposed. Dynamic range adjustment is conducted using three coefficients, namely, the normalization factor (a), the stretching range controlling factor (m), and the stretching size controlling factor (e). The results show that the stretched NDVI (S-NDVI) is more sensitive to vegetation fraction than NDVI when the VF is large, ranging from 0.75 to 1.00. Moreover, the saturation effect in NDVI is effectively removed by using the S-NDVI. Further analysis suggests that there is a good linear correlation between the S-NDVI and the leaf area index (LAI). At the same time, the proposed S-NDVI significantly reduces or even eliminates the saturation effect over high biomass. A comparative analysis is performed between drought indices derived from NDVI and S-NDVI, respectively. In the experiment, reflectance data from the moderate resolution imaging spectroradiometer (MODIS) products and in-situ observation data from the meteorological sites at a regional scale are used. In this study, the coefficient of determination (R (2)) of the stretched drought index (S-DI) is above 0.5, indicating the reliability of the proposed algorithm with surface soil moisture content. Thus, the S-DI is suggested to be used as a drought index in extended regions, thus regional heterogeneity should be taken into account when applying stretching method.