A SPATIO-TEMPORAL FRAMEWORK FOR MODELING ACTIVE LAYER THICKNESS

• Published in: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
• Main Authors: J. Touyz, D. A. Streletskiy, F. E. Nelson, T. V. Apanasovich
• Format: Article in Journal/Newspaper
• Language: English
• Published: Copernicus Publications 2015
• Subjects: Technology; T; Engineering (General). Civil engineering (General); TA1-2040; Applied optics. Photonics; TA1501-1820; Arctic; Active layer monitoring; Active layer thickness; Circumpolar Active Layer Monitoring Network; Climate change; north slope; permafrost; Alaska
• Online Access: https://doi.org/10.5194/isprsannals-II-4-W2-199-2015; https://doaj.org/article/61acc864fa6d4b45ba9b8b70322c3731

The Arctic is experiencing an unprecedented rate of environmental and climate change. The active layer (the uppermost layer of soil between the atmosphere and permafrost that freezes in winter and thaws in summer) is sensitive to both climatic and environmental changes, and plays an important role in the functioning, planning, and economic activities of Arctic human and natural ecosystems. This study develops a methodology for modeling and estimating spatial-temporal variations in active layer thickness (ALT) using data from several sites of the Circumpolar Active Layer Monitoring network, and demonstrates its use in spatial-temporal interpolation. The simplest model’s stochastic component exhibits no spatial or spatio-temporal dependency and is referred to as the naïve model, against which we evaluate the performance of the other models, which assume that the stochastic component exhibits either spatial or spatio-temporal dependency. The methods used to fit the models are then discussed, along with point forecasting. We compare the predicted fit of the various models at key study sites located in the North Slope of Alaska and demonstrate the advantages of space-time models through a series of error statistics such as mean squared error, mean absolute and percent deviance from observed data. We find the difference in performance between the spatio-temporal and remaining models is significant for all three error statistics. The best stochastic spatio-temporal model increases predictive accuracy, compared to the naïve model, of 33.3%, 36.2% and 32.5% on average across the three error metrics at the key sites for a one-year hold out period.