An Assessment of Methods and Remote-Sensing Derived Covariates for Regional Predictions of 1 km Daily Maximum Air Temperature

Remote Sensing

Benoit Parmentier⁴  Brian McGill³  Adam M. Wilson¹  James Regetz⁴  Walter Jetz¹  Robert P. Guralnick²  Mao-Ning Tuanmu¹  Natalie Robinson² 

[1] Department of Ecology & Evolutionary Biology, Yale University, 165 Prospect Street, New Haven, CT 06520-8106, USA; E-Mails:;Department of Ecology and Evolutionary Biology, University of Colorado at Boulder, Ramaley N122, Campus Box 334, CO 80309-034, USA; E-Mails:;Sustainability Solutions Initiative, University of Maine, Deering Hall Room 302, Orono, ME 04469, USA; E-Mail:;National Center for Ecological Analysis and Synthesis, University of California Santa Barbara, 735 State Street, Suite 300, Santa Barbara, CA 93101, USA; E-Mails:

关键词: accuracy;    spline;    weather interpolation;    satellite imagery;    meteorological station;    generalized additive model;    kriging;    geographically weighted regression;   

DOI  :  10.3390/rs6098639

来源: mdpi

【 摘 要 】

The monitoring and prediction of biodiversity and environmental changes is constrained by the availability of accurate and spatially contiguous climatic variables at fine temporal and spatial grains. In this study, we evaluate best practices for generating gridded, one-kilometer resolution, daily maximum air temperature surfaces in a regional context, the state of Oregon, USA. Covariates used in the interpolation include remote sensing derived elevation, aspect, canopy height, percent forest cover and MODIS Land Surface Temperature (LST). Because of missing values, we aggregated daily LST values as long term (2000–2010) monthly climatologies to leverage its spatial detail in the interpolation. We predicted temperature with three methods—Universal Kriging, Geographically Weighted Regression (GWR) and Generalized Additive Models (GAM)—and assessed predictions using meteorological stations over 365 days in 2010. We find that GAM is least sensitive to overtraining (overfitting) and results in lowest errors in term of distance to closest training stations. Mean elevation, LST, and distance to ocean are flagged most frequently as significant covariates among all daily predictions. Results indicate that GAM with latitude, longitude and elevation is the top model but that LST has potential in providing additional fine-grained spatial structure related to land cover effects. The study also highlights the need for more rigorous methods and data to evaluate the spatial structure and fine grained accuracy of predicted surfaces.

【 授权许可】

CC BY   
© 2014 by the authors; licensee MDPI, Basel, Switzerland