Deriving drought indices from MODIS vegetation indices (NDVI/EVI) and Land Surface Temperature (LST): Is data reconstruction necessary?

Abstract

Droughts pose significant economic and ecological concerns, and considering climate change projections, timely monitoring and early warning based on satellite observations must be realized at regional to global scales. Nevertheless, whether data reconstruction is necessary to produce high-quality satellite-based time series data for drought monitoring and the data reconstruction approaches to be applied, if necessary, remain unclear. We attempted to fill this knowledge gap by investigating three widely used data reconstruction approaches, i.e., the Savitzky–Golay filter, harmonic analysis of time series (HANTS) and Whittaker Smoother, across the Lancang–Mekong river basin through the Moderate Resolution Imaging Spectroradiometer (MODIS) MOD13Q1 and MOD11A2 products for 2001–2018 and Google Earth Engine cloud platform. Several remote sensing drought indices based on the unreconstructed and reconstructed vegetation indices (VIs) and Land Surface Temperature (LST) were retrieved for basin-wide drought detection. The performance of the examined reconstruction approaches was evaluated using three statistical indices (coefficient of determination (R2), standard deviation of bias (BIAS(std)) and mean squared error (MSE)), spatial consistency with the reference dataset and capability to characterize the drought events. Data reconstruction considerably enhanced the drought index performances for drought detection; however, reconstruction was not necessary in all situations. The reconstructed drought indices exhibited higher R2 values (by 10.6–10.8%), lower BIAS(std) values (by −1.7– −12.5%), and smaller MSE values (by −5.8– −13.4%) compared to those of unreconstructed indices. For most evaluation indicators, HANTS outperformed the other methods, and Vegetation Condition Index (VCI) and Vegetation Health Index (VHI) outperformed the other drought indices. The findings highlight the importance of data reconstruction to detect and characterize drought events and the dependency of the performance of reconstruction methods on drought indices and evaluation metrics when using MODIS time series data.