1 of 15

The Role of Long-Term Trend and Internal Variability in Altering Fire Weather Conditions in the western United States

Jiale Lou ¹, Youngji Joh ¹, Thomas L. Delworth ²

¹Atmospheric and Oceanic Sciences Program, Princeton University, Princeton, NJ, USA.

² Geophysical Fluid Dynamics Laboratory, NOAA, Princeton, NJ, USA.

Marshall Fire, CO, Dec 2021

Credit: NOAA/NWS Boulder

Session 9A Explaining Extreme Events from a Climate Perspective I

2 of 15

Why vapor pressure deficit (VPD)?

Williams et al. (2022, PNAS)

VPD is closely connected to fire activity in the western US. However, models…

Simpson et al. (2024, PNAS)

3 of 15

Defining extremes: fixed baseline vs. evolving baseline

Sen Gupta et al. (2023) advocated for the use of a fixed baseline. They posit that a fixed-baseline definition accurately captures the increasingly frequent and devastating impacts of marine heatwaves.

Amaya et al. (2023) argue that an evolving baseline, by accounting for long-term warming separately and effectively subtracting it from the total, might provide more practical values for adaptation and mitigation strategies in policy-making and resource allocation.

Figure from Amaya et al. (2023)

4 of 15

Increasing VPD trends in the western US are mainly caused by global warming

Temperature controlled

Humidity controlled

Water demand > water supply,

Leads to drying VPD trends in the western US.

Decomposing total trends into a climatological and a residual component, increases in actual vapor pressure in Baja California and Northern Mexico are mainly cased by background mean states.

Lou, Joh, and Delworth

(under internal review)

ERA5: 1940-2022

Residual

5 of 15

Decadal VPD variability in the southwest US is mainly driven by humidity

A center-of-action in the southwest US

Decadal variability

1-10yrs

PC time series

Spectrum

EOF pattern

Area-averaged VPD

Area-averaged saturation VP

Area-averaged actual VP

1-10yrs

Verified over ERA5

Also valid in JRA55!

6 of 15

Applying a model-analog technique to understand fire-weather precursors

Forecast spread

Advertisement: Jiale Lou’s talk tomorrow @8:50 am

Backward model-analog technique: instead of tracing forward in time to identify the forecast ensembles, backward model-analogs trace backward in time to identify the common precursors that might lead to the occurrence of the pre-described target conditions.

Target state:

7 of 15

Favorable conditions for fire activity in the southwest US

Analogous VPD

Dry VPD condition

Analogous SST

La Nina-like PDV

Analogous SLP/Z500

Positive Southern Oscillation;

High pressure system

Vertical structure of analogous GPH, and zonal wind

High pressure system; Clockwise zonal winds.

Observations vs. SPEAR-LE (30 mem)

996 months 🡪 32,400 months

Backward model-analog (at lags tau):

ERA5-based analog (at lag=0):

VPD evolution:

VPD signal persists a few months prior to the target month.

Soil moisture evolution:

in opposite sign;

slightly longer memory scale; ‘reemergence’ mechanism (Kumar et al. 2019).

ENSO evolution:

La Nina signal;

a few seasons prior to the target condition.

8 of 15

Pinpoint the discrepancies in models

Area-averaged VPD = saturation VP minus actual VP

While both the PDO and NINO3.4 can be accurately simulated by SPEAR-LE, decadal component of VPD is absent, resulting in the weak relationships with the PDO.

Decomposing the VPD into saturation and actual vapor pressure component, the absence of decadal VPD variability is due to misrepresentation of actual vapor pressure.

While CMIP6 models can reasonably reproduce the observed spectrum of saturation vapor pressure, all of them appear to misrepresent decadal components of actual vapor pressure anomalies in the southwest US.

Also valid in JRA55!

Area-averaged VPD = saturation VP minus actual VP

PDO

ENSO

VPD

9 of 15

Conclusion remarks

Increasing VPD trends in the western US are mainly caused by global warming.

The leading VPD mode displays a center-of-action in the southwest US with a pronounced decadal component.

Actual vapor pressure is the main driver of decadal VPD component.

In its dry phase, this VPD mode is characterized by reduced soil moisture, high-pressure systems, and La Niña-like conditions, and connected to decadal SST variability in the Pacific and Atlantic basins.

While saturation vapor pressure can be reasonably captured by SPEAR-LE and CMIP6 models, they all struggle to reproduce the decadal component of the actual vapor pressure, leading to the absence of decadal VPD component.

Implication: VPD variability in the southwest US might be more predictable.

Lou, J., Joh, Y., & Delworth, T. L., The role of long-term trends and internal variability in altering fire weather conditions in the western United States. (under internal review).

10 of 15

Supplementary materials

11 of 15

Background climate change

Schematic adapted from Moritz et al. (2012)

Ignitions: Lightning

Human behavior

Vegetation: Fuel aridity

Biodiversity

Growth rate

Mortality rate

Fire weather conditions:

Wind; Temperature

Humidity; Precipitation

Soil moisture

…

Background climate change alters fire activity

Oceanic

Controls on fire activity

12 of 15

Fire weather indices on the rise: which ones to use?

HEAT

HUMIDITY

WIND

ERA5-based fire danger indices (Vitolo et al. 2020; Scientific Data)

… …

From Quilcaille et al. (2023; Earth Syst. Sci. Data)

13 of 15

Fig. 6. Sensitivity of the signal-to-noise ratio (SNR) to the model-analog ensemble size. SNR of the area-averaged VPD in the southwest US (24-40^oN, 90-120^oW) for (a) DJF, (b) MAM, (c) JJA, and (d) SON, respectively. On the vertical axis displays the model-analog ensemble sizes drawn from each month ranging from 1 to 100. On the horizontal axis displays the time lags from the initial month (lag=0) to -12 months prior. The target state at lag 0 is pre-specified as the positive phase of the leading VPD mode. The ensemble means of the selected model-analogs represent the desired signal, and the standard deviations of them denote the ensemble spread (i.e., noise). SNR is then the fraction between signal and noise.

14 of 15