A new computer modeling technique developed by scientists at the US National Center for Atmospheric Research (NCAR) offers the potential to generate summertime drought forecasts across the Western USA months-ahead, with the capability of differentiating between dry conditions at locations just a couple of miles apart.
The technique uses statistical methods and machine learning to analyze key drought indicators during the winter and spring and correlate them with the likelihood of dryness throughout the landscape the following summer. The scientists say this approach, if adapted for use by forecasters, could provide important information for such priorities as management of water resources, wildland fire and fuels, and agriculture.
Ronnie Abolafia-Rosenzweig, NCAR scientist and lead author of a new paper describing the technique, said, “This approach forecasts drought conditions before they have the largest impact. It gives managers an additional tool that they can use to prepare and guide the decisions they are making.”
Abolafia-Rosenzweig and his co-authors found that predictions issued one to three months in advance could correctly identify the occurrence of summer drought in about 81-94% of cases at a resolution of 4km (2.5 miles) across the western third of the USA. The predictions proved most accurate in regions of persistent drought, showing how upcoming dry conditions may vary from a cultivated field to a nearby mountainside or forested area. In regions where dry spells were punctuated by periods of heavy summer precipitation, however, the predictions proved less accurate.
To strengthen societal resilience, scientists are working to improve computer modeling techniques that produce months-ahead predictions of drought.
According to NCAR, current drought forecasts have a relatively coarse resolution of around 10km, which fails to capture the varying degrees of drying across different landscape features in the western USA.
NCAR scientists recently produced a new data set in collaboration with the US Geological Survey which helped open the way for more detailed drought forecasts. The data set is named CONUS404 because it contains simulations of hydrological and climate conditions at 4km resolution across the continental USA (CONUS) over the past 40+ years. Abolafia-Rosenzweig and his co-authors also drew on an equally high-resolution US Department of Agriculture data set, Parameter elevation Regression on Independent Slopes Model (PRISM) for meteorological observations.
These data sets enabled the scientists to identify complex relationships, at a 4km resolution, between climate and drought conditions in late autumn and winter and the extent of drying during the following summer. To identify these relationships, they used machine learning techniques that trained specialized statistical models.
The scientists focused on pre-summer climate variables such as temperature, precipitation and humidity, as well as distant ocean-atmosphere patterns such as the Pacific Multidecadal Oscillation that have far-reaching effects on climate. They found that commonly used drought measures, the Palmer Drought Severity Index and Soil Moisture Percentiles, have strong persistence from winter and spring into the summer, making pre-summer drought severity an especially important predictor of summer drought conditions.
Abolafia-Rosenzweig said the drought forecasting method can augment a fire prediction technique that he and his co-authors had developed last year. Combining the drought and fire models offers the potential for a very detailed look at fire hazard across the US West.
“The West is in a very unique period in terms of both drought and fire with records being broken that go back thousands of years,” he said. “The climate projections are showing drought conditions will continue to intensify in the future. Having tools that can better inform management is becoming increasingly important.”
