Unearthing predictability: CISL-driven research examines soil moisture forecasts
by Shira Feldman
"If you predict precipitation two weeks in advance, those predictions are not very skillful. But you can predict the soil moisture several weeks in advance." —Sanjiv Kumar
A rural agricultural landscape shows a vibrant crop against the rich, dark soil of a field. Photo credit: Drazen Nesic.
New research is transforming how we predict soil moisture, particularly for forecasts looking two weeks to three months ahead. This period, known as subseasonal-to-seasonal (S2S), bridges the gap between short-term weather forecasts and longer seasonal outlooks.
Traditionally, predicting precipitation accurately several weeks in advance has been very difficult. However, exciting new findings show that even when rain forecasts aren't precise, soil moisture predictions can be remarkably accurate.
Study co-author Sanjiv Kumar. Photo Credit: Cole Sikes.
"This new science will help farmers and resource managers."
The paper provides strong evidence that starting with a precise description of the land's current state, known as "land initialization," greatly improves the accuracy of subseasonal soil moisture forecasts. Land initialization means measuring the current condition of the land to give the predictive model a realistic starting point, leading to more accurate predictions. This new research suggests that the initial land condition, especially soil moisture, is crucial for skillful predictions. In forecasting, "skill" means how much better a prediction is compared to a basic forecast, so a more skillful forecast is better at capturing observed conditions.
Sanjiv Kumar, a co-author of the study, explains the significance: "If you initialize the land correctly, then both the soil moisture as well as other water-related quantities, e.g., streamflow, can be more predictable,” said Kumar. “Land initialization has been paid less attention compared to atmosphere or ocean initialization.”
The research provides evidence that land initialization alone accounts for a remarkable 91% (±3%) of the total subseasonal forecast skill for root zone soil moisture in both summer and winter conditions. “It was important to put a quantitative estimate to initialization’s impact," said Kumar.
This research offers broad practical benefits: farmers can use soil moisture forecasts for irrigation and drought planning, and the findings are also relevant for wildfire prediction due to dry land conditions.
The study used the Community Earth System Model version 2 Sub to Seasonal climate forecast experiments (CESM2-S2S) and included eight different experiments to determine the individual contributions of land, atmosphere, and ocean initializations (Richter et al., 2024).
A framework underscoring the critical role of land surface initialization in developing skillful soil moisture forecasts. Photo Credit: Thomas M. Kavoo.
These experiments were conducted on NSF NCAR’s Cheyenne system as part of the CESM Community allocation. The results clearly showed that land consistently provides the most significant contribution to soil moisture forecast skill.
The data analysis conducted by Kumar’s team required powerful systems to handle the large amount of data."All these datasets were generated at NSF NCAR,” Kumar said. “It's an extensive set of sensitivity experiments that was done using the CESM climate model. All these datasets are at campaign storage or Casper, through which we directly accessed and analyzed the data. I could not have done this research without NSF NCAR resources because these data are very big."
Kumar also stressed how essential the advanced computational resources provided by NSF NCAR and CISL were:
"I have not found any customer service that is more helpful than CISL. Anytime you send them an email, someone always gives you a helpful reply."
"It has been a rewarding experience for me, working for almost 15 years now with NSF NCAR and CISL. I don't know what my career would have been without them. I would rate CISL and its computational service at the top of everything I have done."
Kumar served on the steering committee for NSF NCAR’s July 2025 S2S Predictability Workshop, which focused on the role of land-atmosphere interactions: "I'm very happy to see people paying attention to this issue," he said. Through this workshop, a team is gathering community input to further refine understanding of why predictions differ across models and how to improve them. “Modeling centers like NASA, DOE, and NOAA will all contribute to the dataset," Kumar stated.
Kumar also finds joy in working with students: "I'm a faculty member, so I always work with students. It’s exciting to bring their energy and their thought processes into the research, and to see how they develop in their careers."
He offered insightful advice for the next generation: “Get connected to the right group of people. If you want to do new or computationally intensive things, getting access to resources is very important, because that's a bottleneck.”
Duan, Y., Kumar, S., Maruf, M. et al. Enhancing sub-seasonal soil moisture forecasts through land initialization. npj Clim Atmos Sci 8, 100 (2025). https://doi.org/10.1038/s41612-025-00987-0
This study was supported by USDA-NIFA Awards 2020-67021-32476 and 2023-70003-41354.
