Identifying and assessing vegetation behaviour in riparian zones at large scale in the Brazilian Savannah

Abstract

Riparian zones (RZs) have a clear distinct behaviour than the rest of the landscape. Particularly in water-limited regions, such as the Brazilian Savannah (Cerrado biome), where dry season may extend 5 months, the difference between riparian and upland zones is highly pronounced due to vegetation water access to groundwater, and this can have implications on the climatic and hydrological cycles. In order to quantify this difference at large-scale, it was herein proposed to (1) map RZs using topographical information, (2) investigate how land cover is distributed among topographic gradients and (3) investigate vegetation behaviour through remote sensing vegetation measurements and evapotranspiration (ET) estimation. A 140,000 km² upland region inside the Cerrado biome, called the Urucuia aquifer system, was chosen as study site. The region has seen a huge agricultural expansion during the last decades, with mechanized and irrigated crops increasingly using water from its underground reserves, which associated with climate change can have a big impact on the ecosystem, and understanding the role of RZs can be essential to quantify this impact. The height above nearest drainage (HAND) index was used to map RZs, by visually assessing bellow which values the index provided a reasonable RZ buffer comparing with Google Earth imagery. We also used HAND to quantify across its values the historical land cover distribution obtained by the MapBiomas database, and analyse vegetation behaviour in RZs and upland zones (UZs) using remote sensing vegetation measurements of normalized difference vegetation index (NDVI) and normalized difference moisture index (NDMI) and ET estimation from the surface energy balance algorithm for land (SEBAL). A necessary step for HAND computation is a defined stream network, for which the main challenge is identifying channel heads. Herein it was developed an algorithm that produced a varying draining area threshold (vDAT) map for channel initiation, using the topographic position index (TPI) as an auxiliary variable. This algorithm was tested using MERIT-DEM. With the stream network, HAND values bellow 5 m provided the best RZ buffer. As for land cover distribution, we captured that forests naturally occur more densely in the extreme values of HAND (very shallow and very deep) and that farmland historical occupation in the Urucuia region occur more in the upper portions of the terrain, possibly due to soil conditions stablished during landscape formation and evolution. As for vegetation activity, the land cover class seems to have more influence on vegetation behaviour than topographic position, for all indicators computed. Yet, NDMI values in Riparian Forests are greater than in Upland Forests, particularly towards drier conditions, in terms of both seasonality (drier months) and inter-annual variability (drier years). Despite this indication of more water available in RZs than UZs, the ET estimation could not capture these differences, possibly due to difficulties in estimating this variable in natural vegetation with high degree of water stress.