Ecological Archives A022-121-A1

Julia L. Moore, Song Liang, Adam Akullian, Justin V. Remais. 2012. Cautioning the use of degree-day models for climate change projections in the presence of parametric uncertainty. Ecological Applications 22:2237–2247. http://dx.doi.org/10.1890/12-0127.1

Appendix A. Detailed methods for developing contemporary and future temperature data sets.

From the station data, an interpolated contemporary temperature data set was generated at a grid of 90 by 90 meter cells across Sichuan Province as follows. First, mean daily temperature values for each station were averaged on each day of the year over all years between 1980 and 2009 for which data were reported. This produced a single year of averaged daily temperature data at each station (366 days of data, as a consequence of leap years). To interpolate these daily temperatures across every grid cell in Sichuan, a multiple linear regression model was constructed that predicts daily temperature from elevation, latitude and longitude variables following methods described elsewhere (Chuanyan et al. 2005). A separate model was fit for each day using data from all stations, with modeling carried out iteratively using the R statistical package (version 2.7.1). Regression coefficients for each of the 366 days were then used to predict a vector of mean daily temperature at each 90 by 90 meter cell across the full spatial domain, yielding a contemporary temperature data set consisting of 366 surfaces, each representing one day of interpolated mean temperature.

To generate a future temperature data set for 2050, the original temperature data for all 68 weather stations were entered into a statistical model that incorporates latitude, elevation, and time variables. The resulting future temperature projection makes the simple assumption that the rate of temperature increase observed between 1980 and 2009 in Sichuan will continue unchanged into the future. First, daily mean temperature data from 1980 to 2009 was averaged at each station to provide monthly means. A mixed-effects model was constructed to fit mean monthly temperature to elevation, latitude, longitude, and seasonal variables (Table A1) using the STATA function for cross-sectional time-series analysis, xtgee, and weather station as the panel variable. An F test was used to compare a reduced model to the full model, and the final model was selected by evaluating the accuracy of predictions in a data-splitting procedure, using the first 10 years of temperature data to predict the last 20 years of data, and using the first 20 years of temperature data to predict the last 10 years of data. The final model was also evaluated by comparing model predictions to the 2050 Intergovernmental Panel on Climate Change (IPCC) estimates of seasonal temperature in western China (Solomon et al. 2007). The variables retained in the final model are given in Table A1, along with model coefficients. When compared to the full model, which significantly and systematically over-predicted observed temperatures (Student's t, H0 :(pred-obs)avg=0, p < 0.00005), the final (reduced) model did not significantly over- or under-predict station temperatures (p = 0.72). The mean difference between the observed mean monthly temperature and the model-predicted monthly temperature was 0.027 °C. What is more, when the final model was used to project the linear trend observed from 1980–2009 into future years, there was broad agreement with the IPCC predicted increase of 2.3–4.9 °C by 2100 for the region (Solomon et al. 2007). Thus, the simple linear projections of the historic warming trend in Sichuan to 2050 were taken as a plausible future scenario for examining the influence of parametric uncertainty in degree-day models. This final model was then applied at every 90 by 90 meter cell of the spatial domain using the associated predictor variables, with the month in series variable projected forward to 2050 to yield a one year time-series of monthly temperature. The resulting monthly temperature data set was linearly interpolated to yield a daily temperature value at each cell for 2050.

TABLE A1. Predictor variables tested for significance in the statistical temperature projection model, resulting p values in the full and final models, and variable coefficients included in the final model.

TableA1

Literature Cited

Chuanyan, Z., N. Zhongren, and C. Guodong. 2005. Methods for modelling of temporal and spatial distribution of air temperature at landscape scale in the southern Qilian mountains, China. Ecological Modeling 189(1–2):209–220.

Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K. Averyt, M. Tignor, and H. Miller. 2007. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, 2007. Tech. rep.


[Back to A022-121]