Accurate and cost-effective quantification of the carbon cycle for agroecosystems at decision-relevant scales is critical to mitigating climate change and ensuring sustainable food production. However, conventional process-based or data-driven modeling approaches alone have large prediction uncertainties due to the complex biogeochemical processes to model and the lack of observations to constrain many key state and flux variables. Here we propose a Knowledge-Guided Machine Learning (KGML) framework that addresses the above challenges by integrating knowledge embedded in a process-based model, high-resolution remote sensing observations, and machine learning (ML) techniques. Using the U.S. Corn Belt as a testbed, we demonstrate that KGML can outperform conventional process-based and black-box ML models in quantifying carbon cycle dynamics. Our high-resolution approach quantitatively reveals 86% more spatial detail of soil organic carbon changes than conventional coarse-resolution approaches. Moreover, we outline a protocol for improving KGML via various paths, which can be generalized to develop hybrid models to better predict complex earth system dynamics.

The hydrologic sciences and water resources management have long depended on a combination of in situ measurements and remotely sensed data for research and regulatory purposes. In situ measurements are constrained in their spatial distribution by both logistics and costs. Satellite‐ and manned aircraft‐based remote sensing provide spatially broad data, but often of a coarse resolution. Furthermore, all these data collection methods are often limited in their ability to respond to specific short‐term events when the data and imagery could inform real‐time assessment and decision‐making, especially during and immediately after natural disasters. Researchers and regulators are increasingly using unmanned aerial vehicles (UAVs) to fill these gaps in both spatial and temporal data resolution. UAVs are often more cost‐effective than traditional remote sensing or in situ data collection methodologies and the availability of new advancements in UAV design, power supply, payload capacity, and sensors has been driving rapid innovations in their use in the hydrological sciences. This in‐depth review explores these UAV technologies and how they are being employed for on‐demand and cost‐effective characterization, monitoring, assessment, and modeling of soil and water resources. Our discussion addresses both key opportunities and challenges in surface water and groundwater studies, land‐atmosphere interaction studies, water management in agriculture and forestry, ecosystem monitoring, and hydrological modeling.

Cover crops are gaining traction in many agricultural regions, partly driven by increased public subsidies and by private markets for ecosystem services. These payments are motivated by environmental benefits, including improved soil health, reduced erosion, and increased soil organic carbon. However, previous work based on experimental plots or crop modeling indicates cover crops may reduce crop yields. It remains unclear, though, how recent cover crop adoption has affected productivity in commercial agricultural systems. Here we perform the first large‐scale, field‐level analysis of observed yield impacts from cover cropping as implemented across the US Corn Belt. We use validated satellite data products at sub‐field scales to analyze maize and soybean yield outcomes for over 90,000 fields in 2019–2020. Because we lack data on cover crop species or timing, we seek to quantify the yield impacts of cover cropping as currently practiced in aggregate. Using causal forests analysis, we estimate an average maize yield loss of 5.5% on fields where cover crops were used for 3 or more years, compared with fields that did not adopt cover cropping. Maize yield losses were larger on fields with better soil ratings, cooler mid‐season temperatures, and lower spring rainfall. For soybeans, average yield losses were 3.5%, with larger impacts on fields with warmer June temperatures, lower spring and late‐season rainfall, and, to a lesser extent, better soils. Estimated impacts are consistent with multiple mechanisms indicated by experimental and simulation‐based studies, including the effects of cover crops on nitrogen dynamics, water consumption, and soil oxygen depletion. Our results suggest a need to improve cover crop management to reduce yield penalties, and a potential need to target subsidies based on likely yield impacts. Ultimately, avoiding substantial yield penalties is important for realizing widespread adoption and associated benefits for water quality, erosion, soil carbon, and greenhouse gas emissions.

The photosynthetic capacity or CO₂-saturated photosynthetic rate (V_max), chlorophyll, and nitrogen are closely linked leaf traits that determine C₄ crop photosynthesis and yield. Accurate, timely, rapid, and nondestructive approaches to predict leaf photosynthetic traits from hyperspectral reflectance are urgently needed for high-throughput crop monitoring. Therefore, this study thoroughly evaluated the state-of-the-art physically-based radiative transfer models (RTMs), data-driven partial-least-squares regression (PLSR), and generalized PLSR (gPLSR) models to estimate leaf traits from leaf-clip hyperspectral reflectance, which was collected from maize (Zea mays L.) plots with diverse genotypes, growth stages, treatments of nitrogen fertilizers and ozone stresses in three growing seasons. Results show that leaf RTMs considering bidirectional effects can give accurate estimates of chlorophyll content (Pearson correlation r = 0.95), while gPLSR enabled retrieval of leaf nitrogen concentration (r = 0.85). Using PLSR with field measurements for training, the cross-validation indicates that V_max can be well predicted from spectra (r = 0.81). The integration of chlorophyll content (strongly related to visible spectra) and nitrogen concentration (linked to shortwave infrared signals) can provide better predictions of V_max (r = 0.71) than only using either chlorophyll or nitrogen individually. This study highlights leaf chlorophyll content and nitrogen concentration have key and unique contributions to V_max prediction.

Substantial uncertainty exists in daily and sub-daily gross primary production (GPP) estimation, which dampens accurate monitoring of the global carbon cycle. Here we find that near-infrared radiance of vegetation (NIR_v,Rad), defined as the product of observed NIR radiance and normalized difference vegetation index, can accurately estimate corn and soybean GPP at daily and half-hourly time scales, benchmarked with multi-year tower-based GPP at three sites with different environmental and irrigation conditions. Overall, NIR_v,Rad explains 84% and 78% variations of half-hourly GPP for corn and soybean, respectively, outperforming NIR reflectance of vegetation (NIR_v,Ref), enhanced vegetation index (EVI), and far-red solar-induced fluorescence (SIF₇₆₀). The strong linear relationship between NIR_v,Rad and absorbed photosynthetically active radiation by green leaves (APAR_green), and that between APAR_green and GPP, explain the good NIR_v,Rad-GPP relationship. The NIR_v,Rad-GPP relationship is robust and consistent across sites. The scalability and simplicity of NIR_v,Rad indicate a great potential to estimate daily or sub-daily GPP from high-resolution and/or long-term satellite remote sensing data.

High-quality bathymetric maps of inland water bodies are a common requirement for hydraulic engineering and hydrological science applications. Remote sensing methods, such as space-borne and airborne multispectral imaging or lidar, have been developed to estimate water depth, but are ineffective for most inland water bodies, because of the attenuation of electromagnetic radiation in water, especially under turbid conditions. Surveys conducted with boats equipped with sonars can retrieve accurate water depths, but are expensive, time-consuming, and unsuitable for unnavigable water bodies. We develop and assess a novel approach to retrieve accurate and high-resolution bathymetry maps. We measured accurate water depths using a tethered floating sonar controlled by an unmanned aerial vehicle (UAV) in a lake and in two different rivers located in Denmark. The developed technique combines the advantages of remote sensing with the potential of bathymetric sonars. UAV surveys can be conducted also in unnavigable, inaccessible, or remote water bodies. The tethered sonar can measure bathymetry with an accuracy of ∼2.1 % of the actual depth for observations up to 35 m, without being significantly affected by water turbidity, bed form, or bed material.