The role of tree traits in explaining throughfall variability in US forests

Abstract Throughfall, precipitation that passes through the plant canopy or drips from canopy surfaces, represents the dominant input of water to most forests and is extremely spatially and temporally variable. Many plant traits can influence throughfall, but because measuring throughfall and a wide variety of plant traits is labor intensive, most previous studies have been limited to a small number of nearby sites and a small subset of plant traits. As a result, it remains unclear which tree traits are most important in influencing throughfall across a wide variety of forests. Here, this gap is addressed by combining three large datasets: (1) standardized multiyear throughfall data from across the United States; (2) high‐resolution remotely sensed tree data, including individual crown delineations and tree species identity; and (3) plant traits data, resulting in colocated throughfall and trait data for 8385 rainy days from 53 collectors at 16 sites below 24 different tree species. This was used to train two machine learning models to predict throughfall ratio (throughfall/precipitation). The RF‐1 model was driven solely by precipitation amount, the primary determinant of throughfall amount, while RF‐20 retained precipitation amount but also included season and 17 tree crown, leaf, and life history traits. RF‐20 had the best performance, predicting ~1‐m 2 resolution daily throughfall ratio to ±22.5% of observed (20% accuracy improvement relative to RF‐1), especially for precipitation amounts ≤17 mm, and exceeded the performance of RF‐1 at 74% of the throughfall collectors. Precipitation amount, season, and five crown metrics including crown area, height, and height:width were, in order, the most important variables for model performance. Most of the remaining traits were moderately important. These findings can help optimize future studies when deciding which traits to monitor and which parameters might result in the biggest improvement of hydrological models that partition precipitation. Moreover, since several traits altered the amount of precipitation that throughfall accounted for by 5–10 percentage points, the findings can inform tree species selection to minimize the risk of unintended consequences for planted forests, which comprise almost 300 million ha globally with 3 million ha planted annually.