Can weather and climate data from many sites forecast how drought hits soil, streams and vegetation?
Missing rain dries the soil first. If the shortfall lasts, it shows up later in rivers and in plants, and each of those stages is watched by a different monitoring system. Since summer 2026 I’ve been putting those systems on one scale to see how far ahead drought impacts can be forecast.
The Illinois Climate Network measures soil moisture and weather at 19 stations across the state, and some of its records go back to 1987. Each needle on the map is one station, reaching down into the soil it measures.
Each stage is recorded by a different system: soil sensors at the stations, USGS stream gauges and Landsat satellite images. I built the pipelines that bring all three onto a common scale: more than 760,000 observations.
Soil moisture changes slowly, so repeating the latest reading is a hard forecast to beat. I made the test harder: the model never sees the most recent 14 days of soil readings, and the baseline repeats that two-week-old reading.
On held-out 2025 data, the model cut 5 cm soil-moisture error by 41.8%. Streamflow forecasts beat the same kind of baseline in all 25 watersheds, with a median error reduction of 27.7%.
Share of day-to-day variation each source explained on its own:
“Old soil” is the two-week-old reading. From the August 2026 abstract.
At 50 cm, soil moisture averaged 0.30 below its seasonal normal, but models fed only weather predicted about normal. Adding the two-week-old 50 cm reading shrank that miss from 0.32 to 0.10.
In an early test of 14-day forecasts, error at the hardest of 12 stations was 2.2× the easiest. That is why I report results station by station and by depth, not as one average.
The framework needs inputs that exist well beyond Illinois: weather and soil stations, stream gauges and satellite images. Where stations are sparse, reanalysis data such as ERA5-Land fills in every ~9 km grid cell, so I tested it against the station records and helped validate an ERA5-Land extraction pipeline at NCSA.
Interviews with 12 Illinois farmers pointed to the same need: soil-moisture information for where they farm.
Drought monitors describe conditions now. A forecast that says how soil, streams and plants will respond in the coming weeks gives people time to plan.
What comes next“Tracking Drought Propagation Across Soil Moisture, Streamflow, and Vegetation in Illinois Using Machine Learning,” 41st Conference on Hydrology. First author, with Kyla Wolski and Dr. Allison Goodwell.
Mentor: Dr. Allison Goodwell, Illinois State Water Survey. Co-author: Kyla Wolski, Department of Civil and Environmental Engineering, University of Illinois Urbana-Champaign.