Soil-vegetation dynamics in arid and semi-arid alluvial fans: a machine learning and geomorphological approach

Abstract Alluvial fans, prominent landforms in arid and semi-arid regions, serve as sites for urban and rural development, agriculture, and infrastructure. Despite extensive research, gaps remain in understanding how geomorphological factors influence soil properties and vegetation cover on these landforms. This study employs machine learning algorithms and geomorphological analysis, combined with soil sampling data including texture, sand, silt, clay percentages, pH, electrical conductivity, total organic carbon, available phosphorus, available potassium, total nitrogen, total neutralizing value, potassium, and remote sensing-derived NDVI to examine the effects of geomorphology on soil characteristics and vegetation cover across different alluvial fan surfaces (young, old, and relict). Among the various machine learning algorithms evaluated, the decision tree demonstrated the highest efficacy in this study, as evidenced by its superior performance in validation tests relative to the other methods considered. Results reveal that NDVI values are generally higher in gullies compared to interfluves and are elevated at the apex of relict fan and the toe of old fan. This distribution is likely influenced by the geomorphology and soil texture, affecting soil moisture availability. Additionally, the higher NDVI at the toe of the old fan may relate to a higher water table. Machine learning analysis identifies available phosphorus as the most influential soil characteristic on NDVI, followed by total neutralizing value, electrical conductivity, and silt content, possibly reflecting plant species’ requirements for specific nutrient profiles.