The Impact of Topography and Rainfall Regime on Temporal and Spatial NDVI-Derived Metric of Herbaceous Vegetation in Arid Linear Erosion Features: A Satellite Imagery and Machine Learning Analysis

Global climate change increases the frequency of synoptic forcing, across the Mediterranean Basin. These systems often trigger convective storms characterized by intense rainfall, resulting in the formation of linear erosion features (LEFs), such as gullies. Previous studies focused on the morphological characteristics of LEFs, whereas the relationships among LEF geomorphology, landscape topography, and rainfall were poorly studied. This knowledge gap is attributable to the multitude of topographic and rainfall variables, which make manual analysis both challenging and time-consuming. Recent advances in high-spatial-resolution satellite imagery, together with machine learning techniques, provide a potential alternative. Accordingly, the following objectives were defined: (1) design a framework for quantifying temporal changes in fresh herbaceous vegetation across different natural and human-induced linear erosion features (LEFs) in arid hilly loess landscapes using satellite imagery data; (2) develop a predictive model for geomorphological patterns of LEFs based on rainfall and topographic parameters. Three sites were studied, all in the northern Negev, with a semiarid climate and loess deposits covered with a dense LEF net. A unique framework was developed, including the use of acquired PlanetScope® imagery datasets, the allocation of elevation segments along the LEFs, and the normalization of the median NDVI value of each segment by that of the reference plot. Using ‘Random Forest’, the importance, and the directionality of the factors affecting the LEF fresh herbaceous over the elevation segments was determined by a linear correlation. The finding indicates that the NFVR of the LEF baseline (Intercept) was affected only by <1% from the aspect, while there was a 26% increase in NFVR downslope. The LEF (or segment) length affects the Intercept of the NFVR correlation, changing by 21% and 37%, respectively. The steepness affects the NFVR Intercept only by (0–8%), while there was a 43% effect on the NFVR change. In practice, the framework enables the prediction of segments vulnerable to land collapse, and adequate management of gullied areas.