The sensitivity of ecosystem service models to choices of input data and spatial resolution. (April 2018)

• Record Type: Journal Article
• Authors: Bagstad, Kenneth J.; Cohen, Erika; Ancona, Zachary H.; McNulty, Steven G.; Sun, Ge
• Is Part Of: Applied geography. Volume 93(2018)
• Page Start: 25
• Page End: 36
• Publication Date: 2018-04
• Subjects: Ecosystem accounting -- InVEST -- Model comparison -- Spatial scale -- Uncertainty -- WaSSI; Geography -- Periodicals; Human geography -- Periodicals; Human ecology -- Periodicals; 910
• DOI: 10.1016/j.apgeog.2018.02.005
• Languages: English
• ISSNs: 0143-6228

Abstract

Abstract: Although ecosystem service (ES) modeling has progressed rapidly in the last 10–15 years, comparative studies on data and model selection effects have become more common only recently. Such studies have drawn mixed conclusions about whether different data and model choices yield divergent results. In this study, we compared the results of different models to address these questions at national, provincial, and subwatershed scales in Rwanda. We compared results for carbon, water, and sediment as modeled using InVEST and WaSSI using (1) land cover data at 30 and 300 m resolution and (2) three different input land cover datasets. WaSSI and simpler InVEST models (carbon storage and annual water yield) were relatively insensitive to the choice of spatial resolution, but more complex InVEST models (seasonal water yield and sediment regulation) produced large differences when applied at differing resolution. Six out of nine ES metrics (InVEST annual and seasonal water yield and WaSSI) gave similar predictions for at least two different input land cover datasets. Despite differences in mean values when using different data sources and resolution, we found significant and highly correlated results when using Spearman's rank correlation, indicating consistent spatial patterns of high and low values. Our results confirm and extend conclusions of past studies, showing that in certain cases (e.g., simpler models and national-scale analyses), results can be robust to data and modeling choices. For more complex models, those with different output metrics, and subnational to site-based analyses in heterogeneous environments, data and model choices may strongly influence study findings. Highlights: Comparative studies of ecosystem service (ES) assessments have been increasing. We synthesize studies comparing effects of ES data and model choices on assessments. We compare InVEST and WaSSI results in Rwanda with varying data and resolution. We provide recommendations for future modelers based on these and others' findings.