The temperature lapse rate, the rate of temperature change with elevation, is a common parameter used in all sorts of environmental models and downscaling routines. It’s easy to calculate, right? Just plug some temperature and elevation data into a linear regression and there you go. Wrong!

Motivated by my interest in elevation dependent warming, I tried calculating some lapse rates from station temperature data. I was surprised at some of the numbers I was getting. It turns out that it is really easy to get outrageous lapse rates if you’re not careful. The figure below shows some example temperature data from a group of 30 weather stations in Oregon. Panel b shows some different linear regressions you could create from this data, depending on which stations you used. Panel c shows the distribution of lapse rates you could calculate from samples of 10 stations from this dataset.

(a) Black line shows mean elevation across longitude over the study domain of the Oregon Cascades, United States (42.8–44N, 120.5–123W). Coloured points show mean summer maximum temperature from 30 weather stations with symbols indicating the originating network. Vertical grey dashed line marks the maximum elevation of the transect and divides the stations into west and east groups. (b) Examples of lapse rates calculated from these stations, including using only sites from the US Cooperative Observer Program (COOP), the US Interagency Remote Automatic Weather Stations (RAWS) network, the US Natural Resources Conservation Service Snowpack Telemetry (SNOTEL) network, only sites east or west of the divide, the mean environmental lapse rate (MELR, −6.5 C km−1), or all stations (n = 30). (c) Distribution of lapse rates estimated from 15,000 samples of 10 stations from the full population shown in (a). Colours and symbols in (c) indicate lapse rates shown in (b). Vertical jitter has been added to points in (c).

The lapse rate for this region could be positive (an inversion) or it could be as steep as -15C/km. I found this pretty disturbing. How are we supposed to downscale temperature data or trust our environmental modeling results if this single parameter, the temperature lapse rate, is so uncertain?

This led me to embark on a journey to understand the sources of uncertainty in temperature lapse rates and to develop ways to calculate better lapse rates. All save you the gory statistical detail, and share with you a single figure summary of this work (Lute and Abatzoglou, 2020). The decision tree below condenses all of our findings into best practices for calculating temperature lapse rates!

Decision tree for estimating temperature lapse rates