Premium
Statistical Attribution of the Influence of Urban and Tree Cover Change on Streamflow: A Comparison of Large Sample Statistical Approaches
Author(s) -
Anderson Bailey J.,
Slater Louise J.,
Dadson Simon J.,
Blum Annalise G.,
Prosdocimi Ilaria
Publication year - 2022
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2021wr030742
Subject(s) - streamflow , quantile , statistics , sample (material) , environmental science , land cover , regression , cover (algebra) , statistical model , urbanization , econometrics , mathematics , hydrology (agriculture) , geography , land use , drainage basin , geology , cartography , ecology , chemistry , mechanical engineering , geotechnical engineering , chromatography , engineering , biology
The strengths and weaknesses of different statistical methodologies for attributing changes in streamflow to land cover are still poorly understood. We examine the relationships between high ( Q 99 ), mean ( Q mean ), and low ( Q 01 ) streamflow and urbanization or tree cover change in 729 catchments in the United States between 1992 and 2018. We apply two statistical modeling approaches and compare their performance. Panel regression models estimate the average effect of land cover changes on streamflow across all sites, and show that on average, a 1%‐point increase in catchment urban area results in a small (0.6%–0.7%), but highly significant increase in mean and high flows. Meanwhile, a 1%‐point increase in tree cover does not correspond to strongly significant changes in flow. We also fit a generalized linear model to each individual site, which results in highly varied model coefficients. The medians of the single‐site coefficients show no significant relationships between either urbanization or tree cover change and any streamflow quantile (although at individual sites, the coefficients may be statistically significant and positive or negative). On the other hand, the GLM coefficients may provide greater nuance in catchments with specific attributes. This variation is not well represented through the panel model estimates of average effect, unless moderators are carefully considered. We highlight the value of statistical approaches for large‐sample attribution of hydrological change, while cautioning that considerable variability exists.