Impact of urbanization on the relationship between antecedent precipitation and landslides: evidence from the San Francisco Bay Area

Abstract:

Given that land use modifications associated with urbanization (e.g., deforestation, hillslope cutting) often increase the likelihood of landslides, we hypothesize that less precipitation is required to trigger slope failure in urbanized areas. To test this hypothesis while controlling for potential collinearities in space and time, we model the relationship between one- to thirty-day cumulative antecedent precipitation and landslides across varying degrees of urbanization using a panel regression with fixed effects. This econometric approach offers a variety of benefits over traditional landslide susceptibility methods, for example by allowing us to flexibly account for the influence of unobserved variables that may impact landslide susceptibility (e.g., lithology, land cover), while controlling for temporal patterns in reporting. We initially piloted this empirical method on the Pacific Coast region of the coterminous United States (i.e., CA, OR, WA) and found the strongest relationship between antecedent precipitation and landslides in urban areas. Here, we test the robustness of this urban/rural comparison to proximity of rural areas to urban development by conducting sub-regional analyses for the San Francisco Bay Area, defined here to include Marin, San Francisco, San Mateo, and Santa Cruz counties. These coastal counties are not only mountainous and, thus, frequently experience orographic precipitation, but they also contain approximately equal fractions of urban and rural area. Focusing on this smaller spatial extent allows us to examine the influence of reporting bias on our regional results, under the assumption that landslides occurring in rural areas are less likely to be unreported within the Bay Area. Initial results of these sub-regional analyses indicate that the relationship between one-day antecedent precipitation and landslides is >2x greater in urban areas compared to rural areas. The magnitude of this urban/rural comparison increases across durations of cumulative antecedent precipitation and is robust when we explicitly control for variables that may covary with urban area classification (e.g., slope, average precipitation). These findings support our regional results and, thus, our hypothesis that less precipitation is required to trigger landslides in urban areas.