Two-Way Feedbacks Between Soil Moisture and NDVI in Tropical South America

Abstract:

We study relations between monthly series of vegetation dynamics (Normalized Difference Vegetation Index—NDVI) and soil moisture in regions of tropical South America including Colombia (CO), Venezuela (VE), and Central (CA), South (SA), East (EA) and West (WA) Amazonia, from 2010 to 2018. NDVI data comes from NOAA’s Advanced High-Resolution Radiometer (AVHRR), while soil moisture data come from the Soil Moisture/Ocean Salinity satellite mission (SMOS). We use lagged cross-correlations (Pearson test), cross wavelet analysis (wavelet cross spectrum and wavelet coherence), and a nonlinear causality method, PCMCI, that allows identifying common causal drivers and links among high dimensional simultaneous and time-lagged variables. Results for cross-correlations show that when SMOS leads NDVI (negative lags), greater positive correlations are observed in SA, EA and for an area on the right side of the Andes between the CO and VE regions, for negative lags (3-4 months). For lags between 0 and -2 months, cross-correlation in the CO and VE regions are accentuated. Similarly, when NDVI leads SMOS (positive lags), the greatest negative correlations are observed in the SA, EA and the zone between the CO and VE regions. Results of cross wavelet analysis show very different behavior across regions: in CO, VE and CA both variables exhibit a strong positive locking (in phase) at the 12-month frequency band (annual cycle), whereas in WA, EA and SA both variables exhibit a strong negative locking (out phase) (Fig 1). Results of nonlinear causality metrics show significant differences among regions. For instance, from NDVI to SMOS, negative nonlinear causalities (NC) are evident in CO, VE, and CA (at different time lags), whereas positive ones appear in WA, SA, and EA. On the other hand, from SMOS to NDVI, positive NCs appear in CO, VE, WA, and EA, whereas negative NCs appear in CA and SA (Fig 2). Our results confirm the existence of strong two-way feedbacks between SMOS and NDVI in tropical South America, which evidences the fundamental role of the land surface-vegetation interaction in the continent.

Figure 1. Relations between soil moisture and NDVI in Central Amazonia. (Left) Cross wavelet transform. (Right) Nonlinear causality diagram.