Spatial rangeland variability: using summary statistics and multifractal analysis to classify and monitor rangelands.

Rangelands ecosystem comprises more than a third of the global land surface, sustaining key ecosystem services and livelihoods. Unfortunately, they suffer from severe degradation on a global scale. Tailored-monitoring of rangeland will allow us to improve their management and maintain their social-ecological systems.

MODIS data are commonly used to calculate Normalized Differenced Vegetation Index (NDVI) and NDVI anomaly (NDVIa) to monitor rangelands. In this study, we compare summary statistics and multifractal analysis to see if using complexity based tools improves our ability to differentiate land uses and types using remote sensing.

We collected time series using satellite data of MODIS (MOD09Q1.006) from 2000 to 2019. An area from southeastern Spain (Murcia province) of 6.25 Km² was selected. This area comprised 132 pixels with a spatial resolution of 250 x 250 m² and a temporal resolution of 8 days. This area includes irrigated and rainfed crops, shrubs and forested patches.

Multifractal detrended fluctuation analysis (MF-DFA) focuses on measuring variations of the moments of the absolute difference of their values at different scales. This allows us to use different multifractal exponent such as generalized Hurst exponent (H(q)), and its range (ΔH) to characterize the area. Here, we have selected H(1), H(2) and ΔH, to reflect variance, persistence and multifractality, respectively. Then, we compare them to the average, standard deviation and kurtosis of our NDVI and NDVIa series.

Our results indicate that MF-DFA, allow us to see more clearly the differences among the pixels than the summary statistics. Particularly H(1) and H(2) of NDVI reflects more precisely the vegetation profile and land uses of the selected area. On the other hand, NDVIa allows us to highlight those pixels where several uses occur, or some feature such as roads interact with NDVI. MF-DFA appears as a promising tool to classify and monitor rangelands.

Acknowledgements: The authors acknowledge the support of Project No. PGC2018-093854-B-I00 of the Ministerio de Ciencia, Innovación y Universidades of Spain, “Garantía Juvenil” scholarship from Comunidad de Madrid, and the financial support from Boosting Agricultural Insurance based on Earth Observation data - BEACON project under agreement Nº 821964, funded under H2020EU, DT-SPACE-01-EO-2018-2020.