Presentation Type

Oral Presentation

Category

STEM (science, technology, engineering, mathematics)

Abstract/Artist Statement

There is a need for location-specific, fine-grain, climate impact information to inform regional and local-level climate adaptation. However, such climate information is hard to obtain from the existing models either due to the lack of sufficient data or too coarse of a model output to be useful. Analog impact models (AIMs), provide an alternative approach. AIMs rely on climate analogs (locations in space and time that have similar climate) to forecast future climate impacts at a spatial grain as fine as a four kilometer pixel. Analog impact models assume that locations with equivalent climate (climate analogs) also share other important characteristics, such as vegetation type, primary productivity, disturbance regimes, etc. AIMs provide spatially resolved, fine grain predictions of climate impacts, which have the potential to inform location-specific climate adaptation. The use of AIMs is growing, yet there is a lack of information on the quality of AIM predictions. Validating AIMs is a challenge, as actual climate impacts can not be observed in the present and compared to AIM predictions. We evaluate AIMs by testing their performance on climate analogs in space in the reference climate period. We identify spatial climate analogs in the western US for the 1961-1990 period using 30 year normals of four climate variables (mean maximum temperature of the warmest month, minimum temperature of the coldest month, actual evapotranspiration, and climatic water deficit). We evaluate AIM performance by comparing remotely sensed Landsat tree-canopy data at each pixel of interest (i.e. the observed value) to the tree cover at its candidate analog pixels (i.e. the predicted value) at increasing climatic dissimilarity levels. We find that the AIM predicts tree cover well: the slope of the linear fit of predicted vs actual cover is 0.78 (R2 = 0.78) for climatically closest analogs. Model bias increases and precision decreases with increasing climate dissimilarity between the focal and the analog pixels. Tree cover is often overpredicted for pixels with low tree cover, suggesting that recent disturbance may drive the error at the low cover end. Our study provides support for the utility of climate analogs as a climate impact assessment tool and provides details on the effects of climatic dissimilarity, the number of climate analogs considered, and spatial distribution of spatial analogs on the quality of prediction.

Mentor Name

Solomon Dobrowski

Personal Statement

At its root, my project addresses the challenge of climate change. Specifically, I am studying ways to forecast how natural (e.g., forests) and human systems (e.g., energy needs for cooling and heating) will respond to the coming changes in climate. Forecasting these changes is a challenging task, natural and human systems are complex, and it’s not obvious how these systems may change with climate. That is what analog impact models are intended to do - forecast system changes, based on climate. Indeed, analog impact models (AIMs) are being broadly used by scientists in different disciplines to forecast climate impacts, and are suggested to be a good communication and decision-making tool for managers. But we don’t know how well AIMs work, because we do not have a direct way to evaluate how accurate their predictions are. We would need to wait for climate impacts to happen, before we can validate AIMs, which is not an option. Validating AIMs is the specific question of my project. I am approximating how well we can expect AIMs to predict future climate impacts by using contemporary climate analogs and observed tree cover data. The contemporary validation of analog impact models adds a small brick of knowledge to the climate analog knowledge stack. This brick bolsters our confidence in using these models to predict climate change impacts, and moves us that much closer to being able to plan for climate change at the local level. I came to graduate school at the University of Montana after working for almost seven years as a land-manager for California State Parks. My motivation for going back to school was to gain expertise in the science of climate change and climate impacts, so that I could use it in my land-management practice. I think there is an urgent need to act now to adapt to climate change. Both for the sake of avoiding the worst impacts of climate change, and for the sake of using limited resources that we have wisely, in a way that provides a lasting impact that can withstand climate change. My work as a graduate student either directly (with the submitted project) advances the knowledge about climate impacts forecasting, or does so indirectly, by addressing the science that will ultimately provide that climate impacts knowledge. After finishing with my degree, I am hoping to work at the intersection of science and application, applying new science into practice. My current research is building my competency and fluency in the research world.

GMT20210216-140905_Svetlana-Y_1920x1200.mp4 (22502 kB)
Presentation_video&audio

GMT20210216-140905_Svetlana-Y.transcript.vtt (15 kB)
Transcript

Share

COinS
 

Should we use climate analogs to predict climate impacts? A contemporary validation.

There is a need for location-specific, fine-grain, climate impact information to inform regional and local-level climate adaptation. However, such climate information is hard to obtain from the existing models either due to the lack of sufficient data or too coarse of a model output to be useful. Analog impact models (AIMs), provide an alternative approach. AIMs rely on climate analogs (locations in space and time that have similar climate) to forecast future climate impacts at a spatial grain as fine as a four kilometer pixel. Analog impact models assume that locations with equivalent climate (climate analogs) also share other important characteristics, such as vegetation type, primary productivity, disturbance regimes, etc. AIMs provide spatially resolved, fine grain predictions of climate impacts, which have the potential to inform location-specific climate adaptation. The use of AIMs is growing, yet there is a lack of information on the quality of AIM predictions. Validating AIMs is a challenge, as actual climate impacts can not be observed in the present and compared to AIM predictions. We evaluate AIMs by testing their performance on climate analogs in space in the reference climate period. We identify spatial climate analogs in the western US for the 1961-1990 period using 30 year normals of four climate variables (mean maximum temperature of the warmest month, minimum temperature of the coldest month, actual evapotranspiration, and climatic water deficit). We evaluate AIM performance by comparing remotely sensed Landsat tree-canopy data at each pixel of interest (i.e. the observed value) to the tree cover at its candidate analog pixels (i.e. the predicted value) at increasing climatic dissimilarity levels. We find that the AIM predicts tree cover well: the slope of the linear fit of predicted vs actual cover is 0.78 (R2 = 0.78) for climatically closest analogs. Model bias increases and precision decreases with increasing climate dissimilarity between the focal and the analog pixels. Tree cover is often overpredicted for pixels with low tree cover, suggesting that recent disturbance may drive the error at the low cover end. Our study provides support for the utility of climate analogs as a climate impact assessment tool and provides details on the effects of climatic dissimilarity, the number of climate analogs considered, and spatial distribution of spatial analogs on the quality of prediction.