Sediment discharge from an agricultural watershed: investigating the influence of land-cover fragmentation using the GSSHA model

摘要

Land-cover patterns may serve as significant predictors of sediment discharge. The GSSHA model, a distributed physically-based watershed model, was used for the Todoroki watershed (Ishigaki Island, Okinawa, Japan) to estimate sediment discharge for different land-cover conditions as inferred from Landsat images (1984-2003). Results indicate that over the past 20 years, the changing land cover of the Todoroki watershed, in the direction of increasing sugarcane areas, had enhanced sediment discharge from the watershed. However, landscape patterns exhibited a comparable or even stronger relationship with sediment discharge compared to land-cover percentages. Based on regression analyses, four pattern metrics (number of patches, radius of gyration, splitting, and cohesion) accounted for 99% of the variability of discharged sediment volume (RMSE = 9.6) compared to 89.5% represented by land-cover percentages (RMSE = 31.6). The pattern metrics plus percentages of forest and sugarcane jointly represented almost 100% of the variability (RMSE = 0.0015). Pattern metrics have more predictive capability than land-cover percentages and can improve understanding of watershed sediment discharge process.