Characterizing the topographic index as a tool to represent spatial soil moisture: The effect of classification approach, digital elevation model type, and resolution

Grouping topographic indices (TIs) values, a topographic metric incorporating upslope contributing area and slope gradient, into topographic index classes (TIC) to estimate soil moisture variability has been shown to improve hydrologic model characterization of complex landscapes. However, the performance of TIC may be enhanced by understanding the inputs in its derivation. In this study, in situ soil moisture measurements across 25 locations in a 4.2 ha field and the k ‐means clustering machine learning algorithm were used to identify the optimum number of classes for characterizing field soil moisture. The effect of using equal‐interval, equal‐area, k ‐means, and Fisher classification approaches; U.S. Geological Survey (USGS) (1 and 9 m); and drone‐captured light detecting and ranging (LiDAR) digital elevation models (DEMs) re‐gridded to 0.25, 0.5, 1, 2, 3, 4, 5, and 10 m resolution was evaluated. The k ‐means clustering, along with elbow analysis and Gap* statistics, identified the spatial pattern of the soil moisture in the field into three groups. Spearman's correlation between a three‐class TIC and the identified soil moisture patterns showed wide variability due to DEM resolution for TIC from drone‐captured LiDAR DEMs—from r = −0.41 to −0.17 for high resolution (0.25–4 m) to r = 0.52 to 0.58 for low resolution (10 m), compared to TIC derived from 1 and ∼9 m resolution USGS DEMs ( r = 0.53–0.64). TIC generated using the equal‐interval approach best correlated with identified soil moisture patterns. Selecting the right classification approach, DEM type, and resolution are important considerations when using TI to estimate soil moisture patterns.