Publications

Cammalleri, C; Vogt, JV (2019). Non-stationarity in MODIS fAPAR time-series and its impact on operational drought detection. INTERNATIONAL JOURNAL OF REMOTE SENSING, 40(4), 1428-1444.

Abstract
A time-series of fAPAR (fraction of Absorbed Photosynthetically Active Radiation) data collected by the MODIS (MODerate Resolution Imaging Spectroradiometer) Terra satellite between 2001 and 2016 has been analysed at global scale in order to identify possible areas interested by a linear temporal trend. The main scope of this study is to quantify how the possible trend may affect estimates of drought events based on 10-day fAPAR standardized anomalies (z values). The trend analysis, performed according to the Theil-Sen approach, highlighted that about 10% of the globe shows a statistically significant trend (following the Mann-Kendall test at p = 0.05), mostly being positive (increasing fAPAR). The impact of such a trend on standardized anomalies has been formally expressed as a linear function of two quantities: 1) the z value itself, and 2) the year under investigation. This approach allows summarising the variations in z into just two parameters. In the case of negative anomalies (of greater interest for drought analyses) and the most recent years on the time-series (of greater interest for operational monitoring systems), the use of a stationary (no trend) reference tends to underestimate both severity and extension of the areas interested by drought compared to the non-stationary (or trend stationary) reference. The areas mostly interested by significant differences in the outputs of stationary and non-stationary analyses are the Indian subcontinent, eastern China and the Mediterranean countries. As an example of the impacts of the choice of one of the two frameworks on drought monitoring, the results for the recent summer drought in Italy in 2017 have been reported. This example demonstrates how the non-stationary approach tends to be conservative compared to the stationary one, with the former detecting larger affected areas and higher average severity compared to the latter.

DOI:
10.1080/01431161.2018.1524603

ISSN:
0143-1161