*4.3. Precipitation as a Driver for Erosion*

Prior research at this site using 14 months of data found that Duration and TotAcc were the drivers for erosion, most strongly in channels. With six years of data, the present study confirmed the earlier result when erosion and precipitation data were lumped without regard for season. OLS regression models of annual erosion for the nine erosion variables, using the set of lagged precipitation parameters as independent variables, and overwhelmingly retained Duration parameters most frequently (24 times) (Table 6). This means that overall nine OLS models of erosion outlined in Table 4, Duration and lagged Duration independent variables had significant coefficients 24 times. Despite the high correlation between TotAcc and erosion variables (Table 3), TotAcc was retained less frequently in the models (7 times) due to the high correlation between Duration and TotAcc (r = 0.903, *p* = 0.001) (Table 3), indicating multicollinearity. Lagged intensity parameters were likewise retained fewer times; AvgInt parameters were retained 14 times, while MaxInt parameters were retained only 5 times. Therefore, using lumped annual data, Duration was the most important predictor of erosion, indicating that over the long term, prolonged precipitation is key.


**Table 6.** Retention frequency of lagged precipitation parameters (Duration, Total Accumulation (TotAcc), and Average and Maximum Intensity (AvgInt and MaxInt, respectively) in OLS regression models of erosion annually and seasonally for the full study area and for each geomorphic area: Channels, Interfluves, and Sidewalls.

When erosion data were partitioned by geomorphic areas (Table 6), channel models overwhelmingly retained Duration most often. In contrast, sidewall and interfluve models retained Duration and AvgInt at approximately the same frequency (retained in 6 and 5 interfluve models and 8 sidewall models, respectively). This shows the importance of precipitation intensity as a driver for erosion in these two geomorphic areas. This may occur because interfluves and sidewalls may be more exposed to rain splash erosion, which is associated with higher intensity precipitation. Channels are not as steeply sloped as sidewalls and gully channel erosion is associated with the flow within the channel, which occurs after long-duration events that result in saturation-related runoff.

When erosion data were partitioned by season, the influence of precipitation intensity became apparent, especially during summer and to a lesser degree winter. This may be observed in Table 6, where MaxInt lagged parameters were retained 12 and 6 times in summer and winter erosion models, respectively, but only 0 and 3 times in spring and autumn models, respectively. This indicates that, while over the long term, Duration was the most important driver, during certain individual seasons intensity became important. This emphasizes the importance of the mechanics of convectional storms (summer) and frontal storms (winter) as an additional factor in seasonal erosion patterns. These patterns are also apparent when model results are partitioned by both season and geomorphic area (Table 6).

Partitioning the data by season, therefore, produces additional knowledge that was not previously captured. We conclude that different drivers may be more effective agents of erosion in different seasons and, therefore, we recommend that studies of precipitation driven erosion should, wherever possible, partition data by season.
