Natural Capital Project, Woods Institute for the Environment, 371 Serra Mall, Stanford, CA 94305-502, USA.
Natural Resource and Environmental Management, University of Hawaii at Manoa, 1910 East West Rd, Honolulu, HI, 96822, USA.
Sci Total Environ. 2017 Feb 15;580:1381-1388. doi: 10.1016/j.scitotenv.2016.12.103. Epub 2016 Dec 28.
Geospatial models are commonly used to quantify sediment contributions at the watershed scale. However, the sensitivity of these models to variation in hydrological and geomorphological features, in particular to land use and topography data, remains uncertain. Here, we assessed the performance of one such model, the InVEST sediment delivery model, for six sites comprising a total of 28 watersheds varying in area (6-13,500km), climate (tropical, subtropical, mediterranean), topography, and land use/land cover. For each site, we compared uncalibrated and calibrated model predictions with observations and alternative models. We then performed correlation analyses between model outputs and watershed characteristics, followed by sensitivity analyses on the digital elevation model (DEM) resolution. Model performance varied across sites (overall r=0.47), but estimates of the magnitude of specific sediment export were as or more accurate than global models. We found significant correlations between metrics of sediment delivery and watershed characteristics, including erosivity, suggesting that empirical relationships may ultimately be developed for ungauged watersheds. Model sensitivity to DEM resolution varied across and within sites, but did not correlate with other observed watershed variables. These results were corroborated by sensitivity analyses performed on synthetic watersheds ranging in mean slope and DEM resolution. Our study provides modelers using InVEST or similar geospatial sediment models with practical insights into model behavior and structural uncertainty: first, comparison of model predictions across regions is possible when environmental conditions differ significantly; second, local knowledge on the sediment budget is needed for calibration; and third, model outputs often show significant sensitivity to DEM resolution.
地理空间模型常用于量化流域尺度的泥沙贡献。然而，这些模型对水文和地貌特征变化的敏感性，特别是对土地利用和地形数据的敏感性，仍然不确定。在这里，我们评估了一种这样的模型，即投资泥沙输送模型，对包括总面积为 6-13500km 的 28 个流域在内的六个地点的性能进行了评估，这些流域的气候（热带、亚热带、地中海）、地形和土地利用/土地覆盖情况各不相同。对于每个地点，我们将未经校准和校准的模型预测与观测值和替代模型进行了比较。然后，我们对模型输出与流域特征之间进行了相关性分析，随后对数字高程模型（DEM）分辨率进行了敏感性分析。模型性能在不同地点之间存在差异（总体 r=0.47），但特定泥沙输出量的估计与全球模型一样或更准确。我们发现泥沙输送指标与流域特征之间存在显著相关性，包括侵蚀性，这表明最终可能会为未测流域开发经验关系。模型对 DEM 分辨率的敏感性在不同地点和地点内都有所不同，但与其他观测到的流域变量没有相关性。这些结果通过对平均坡度和 DEM 分辨率范围不同的合成流域进行敏感性分析得到了证实。我们的研究为使用投资或类似地理空间泥沙模型的建模人员提供了有关模型行为和结构不确定性的实用见解：首先，当环境条件有很大差异时，可以对模型预测进行跨区域比较；其次，需要对泥沙预算有本地知识进行校准；第三，模型输出通常对 DEM 分辨率有很大的敏感性。
