Citation:

Green, L. AND Ted DeWitt. CAN A MODEL TRANSFERABILITY FRAMEWORK IMPROVE ECOSYSTEM SERVICE ESTIMATES? A CASE STUDY OF SOIL FOREST CARBON SEQUESTRATION IN TILLAMOOK BAY, OR, USA. A Community on Ecosystem Services, Jacksonville, FL, December 05 - 09, 2016.

Impact/Purpose:

Frequently, environmental decision makers, managers, planners, and scientists need estimates of ecological stocks or processes at sites where little or no primary data or models exist, and they use existing data or models (i.e., that were measured at or developed for other locations) to obtain those estimates (e.g., they transfer those estimates or models to the new site). However, no standard methodology currently exists to assess whether those transfers are justifiable. Scientists at NHEERL Western Ecology Division in Newport, OR, have developed a framework to assess the transferability of ecological estimates or ecological models, and demonstrate the application of this methodology to estimate carbon sequestration rates for forests in the Tillamook Bay (Oregon) watershed. The abstract contributes to SHC 2.61.

Description:

Budget constraints and policies that limit primary data collection have fueled a practice of transferring estimates (or models to generate estimates) of ecological endpoints from sites where primary data exists to sites where little to no primary data were collected. Whereas benefit transfer has been well studied; there is no comparable framework for evaluating whether model transfer between sites is justifiable. We developed and applied a transferability assessment framework to a case study involving forest carbon sequestration for soils in Tillamook Bay, Oregon. The carbon sequestration capacity of forested watersheds is an important ecosystem service in the effort to reduce atmospheric greenhouse gas emissions. We used our framework, incorporating three basic steps (model selection, defining context variables, assessing logistical constraints) for evaluating model transferability, to compare estimates of carbon storage capacity derived from two models, COMET-Farm and Yasso. We applied each model to Tillamook Bay and compared results to data extracted from the Soil Survey Geographic Database (SSURGO) using ArcGIS. Context variables considered were: geographic proximity to Tillamook, dominant tree species, climate and soil type. Preliminary analyses showed that estimates from COMET-Farm were more similar to SSURGO data, likely because model context variables (e.g. proximity to Tillamook and dominant tree species) were identical to those in Tillamook. In contrast, estimates from Yasso were an order of magnitude less than estimates extracted from SSURGO or COMET-Farm. This difference may have been due to lower context similarity between Yasso sites and Tillamook Bay. The greatest logistical constraints in assessing model transferability were the identification and vetting of context variables used in models. Though user-friendly models with an attractive interface may be preferred, documentation on model construction and data sources are often sparse and limit transferability. On the other hand, transferability of models well-vetted in peer-reviewed literature can be limited by site-specificity. Our model transferability framework applied to a real world case study involving ecosystem services helps demonstrate the utility of this approach and highlights important considerations when deriving estimates from transferred models.

Record Details: