Time-series cloud noise mapping and reduction algorithm for improved vegetation and drought monitoring

Moderate Resolution Imaging Spectro-radiometer (MODIS) time-series Normalized Differential Vegetation Index (NDVI) products are regularly used for vegetation monitoring missions and climate change analysis. However, satellite observation is affected by the atmospheric condition, cloud state and shadows introducing noise in the data. MODIS state flag helps in understanding pixel quality but overestimates the noise and hence its usability requires further scrutiny. This study has analyzed MODIS MOD09A1 annual data set over Sri Lanka. The study presents a simple and effective noise mapping method which integrates four state flag parameters (i.e. cloud state, cloud shadow, cirrus detected, and internal cloud algorithm flag) to estimate Cloud Possibility Index (CPI). Usability of CPI is analyzed along with NDVI for noise elimination. Then the gaps generated due to noise elimination are reconstructed and performance of the reconstruction model is assessed over simulated data with five different levels of random gaps (10–50%) and four different statistical measures (i.e. Root mean square error, mean absolute error, mean bias error, and mean absolute percentage error). The sample-based analysis over homogeneous and heterogeneous pixels have revealed that CPI-based noise elimination has increased the detection accuracy of number of growing cycle from 45–60% to 85–95% in vegetated regions. The study cautions that usage of time-series NDVI data without proper cloud correction mechanism would result in wrong estimation about spatial distribution and intensity of drought, and in our study 50% of area is wrongly reported to be under drought though there was no major drought in 2014.