Around the estuary of the Zhuo-Shui River in Taiwan, the waters recede during the winter, causing an increase in bare land area and exposing a large amount of fine earth and sand particles that were deposited on the riverbed. Observations at the site revealed that when northeastern monsoons blow over bare land without vegetation or water cover, the fine particles are readily lifted by the wind, forming river dust, which greatly endangers the health of nearby residents. Therefore, determining which factors affect river dust and constructing a model to predict river dust concentration are extremely important in the research and development of a prototype warning system for areas at risk of river dust emissions. In this study, the region around the estuary of the Zhuo-Shui River (from the Zi-Qiang Bridge to the Xi-Bin Bridge) was selected as the research area. Data from a nearby air quality monitoring station were used to screen for days with river dust episodes. The relationships between PM₁₀ concentration and meteorological factors or bare land area were analyzed at different temporal scales to explore the factors that affect river dust emissions. Study results showed that no single factor alone had adequate power to explain daily average or daily maximum PM₁₀ concentration. Stepwise regression analysis of multiple factors showed that the model could not effectively predict daily average PM₁₀ concentration, but daily maximum PM₁₀ concentration could be predicted by a combination of wind velocity, temperature, and bare land area; the coefficient of determination for this model was 0.67. It was inferred that river dust episodes are caused by the combined effect of multiple factors. In addition, research data also showed a time lag effect between meteorological factors and hourly PM₁₀ concentration. This characteristic was applied to the construction of a prediction model, and can be used in an early warning system for local residents.

Several related pre-event environment indices coupled with a 200-year event of Typhoon Morakot were used in this study to analyse landslide risk/scale for mapping and verifying the danger grade of watershed landslides. Risk analysis was modelled using maximum daily rainfall of the tested event as the hazard factor, and a complex indicator derived from indices of return period, road development and green deterioration as the vulnerability factor. The vulnerability factor was then modified using the slope to adjust the weight for excluding the flat areas and increasing impact of steep areas. A comprehensive indicator integrated from indices of river concave, headward erosion and dip slope was provided as the hotspot of potential large-scale landslides which were modified by vegetation index to correct sediment yield due to antecedent effects. Finally, sites with high potential large-scale landslide could be interpreted by the danger grade calculated from collapse risk/scale analysis. The results show that the higher the danger grade, the greater the collapse ratio; and the coefficient of determination for the relationship is greater than 0.9. The model developed in this study for accurately estimating the risk/scale of rainfall-induced landslide can be fulfilled.

Lands reserved for indigenous people in Taiwan are mostly located at junctions between forestry compartments and hillside lands. After heavy rains, sediment disasters can easily occur in such locations. Therefore, identifying primary and derived disaster landslide areas of watersheds in lands reserved for indigenous people is imperative. This study used landslide-prone slopes to correct the vulnerability variable in the model proposed by Lin et al. ( 2017 ), and recalculated the landslide risks of watersheds. In addition, the study used protected objects as study subjects, and considered the likelihood of landslide disaster occurrence and its scale. Risk models for assessing the derived disasters caused by watershed landslides were constructed according to the spatial distribution of sediment delivery ratio. Comparisons between the present model and the original model regarding the relationship between landslide risks and landslide ratios in watershed subdivisions reveal that the corrected model shows significant positive correlation (R² increased from 0.59 to 0.91). Risk models for assessing the derived disasters caused by watershed landslides indicate that the risk and the disaster ratio of the watershed subdivisions show significant and positive correlation (R²=0.84), indicating that the model developed in this study can effectively estimate the likelihood of protected objects encountering sediment disasters.