High-throughput phenotyping (HTP) has the potential to revolutionize plant breeding by providing scientists with exponentially more data than was available through traditional observations. Even though data collection is rapidly increasing, the optimum use of this data and implementation in the breeding program has not been thoroughly explored. In an effort to apply HTP to the earliest stages of a plant breeding program, we extended field-based HTP pipelines to evaluate and extract data from spaced single plants. Using a panel of 340 winter wheat lines planted in full plots and grid-spaced single plants for two growing seasons, we evaluated relationships between single plants and full plot yields. Normalized difference vegetation index (NDVI) was collected multiple times through the growing season using an unmanned aerial vehicle. NDVI measurements during grain filling stage from both single plants and full plots were typically positively associated with their respective grain yield with correlation ranging from -0.22 to 0.74. The relationship between single plant NDVI and full plot yield, however, was variable between seasons ranging from -0.40 to 0.06. A genome wide association analysis (GWAS) identified the same significant markers for traits measured in both full plots and single plots, but also displayed variability between growing seasons. Strong genotype by environment interactions could impede selection on quantitative traits, yet these methods could provide an effective tool for plant breeding programs to quickly screen early-generation germplasm especially for qualitative traits. Effective use of early-generation, affordable HTP data could improve overall genetic gain in plant breeding.
Developing crops with better root system architecture and nitrogen use efficiency can lead to resilient crops capable of sustaining productivity in both optimum and stress environments. The increase in soybean global demand and its use in biodiesel and new soy-based products calls for new soybean cultivars that have higher yields, better nutritional values or desirable traits for specific use. The soybean germplasm collection at the USDA is a valuable resource in discovering novel allelic variations. This collection has not been screened extensively for the root nodule and root system architecture traits. We have already established root phenotyping platforms in our laboratory and we are proposing to screen a diverse pool of the USDA soybean collection for nodules and root system architecture traits. Availability of SNP data for this collection will let us run genome-wide analysis and identify QTLs responsible for different root traits. We will then use hairy root transformation to knockout/down some of the candidate genes in the loci identified in earlier reports and new QTLs identified in this project. This project will help us in deciphering the phenotypic variation in soybean root traits. Ultimately, better understanding of the regulatory mechanism controlling these traits can help us in developing resilient crops and a sustainable cropping system.
Crop yield is sensitive to climate change and has been projected to be negatively affected by future climate. To reduce yield loss and ensure food security in the context of climate change, it is critical to understand how climate variables interact with crop growth in agroecosystems. One important and widely used tool to study yield responses to climate is process-based modeling. However, using process-based models to simulate the climate impacts on crops is becoming challengeable as the future climate is characterized by more and more frequent extreme events, such as heatwaves, unpredictable rainfall, and droughts. Most existing crop models may not be capable of characterizing the impacts of such extreme events on crops simply because they usually do not simulate some critical processes that climate variables directly affect crop growth such as photosynthesis. Instead, they use a simplified approach–radiation-use efficiency (RUE) which is a coefficient to describe empirical relationships between intercepted radiation and biomass. The usage of RUE has simplified computation but also limited our understanding of interactions between climate variables (e.g., temperature, CO2, rainfall) and crop growth. Thus, we developed a module combining processes of radiative transfer and photosynthesis (RP) within the canopy to account for the impacts of climate variables on crop growth dynamically. Then, we integrated the RP module into a popular agricultural system model—the Environmental Policy Integrated Climate (EPIC) to assess its performance. The results show that its capabilities of predicting crop yield are comparable to the traditional RUE method. The correlation between observed and simulated biomass is 0.77 for the RUE method, while 0.76 for the RP method. But the RP method could show responses of biomass accumulation to changes in climate factors, which is almost overwhelming for RUE. For instance, the RP module could simulate how extremely high temperatures (which usually last several hours during a day) affect crop growth and also allow the EPIC to distinguish elevated CO2 impacts on C3 and C4 crops, while the default RUE method could not. Therefore, the RP module is promising to improve capabilities and extend functionalities of current process-based models, which is not only beneficial to the community of crop modeling but also enhances our ability to evaluate the impacts of climate change on the agroecosystem.
Genomic tools are increasingly being deployed to unlock factors affecting genetic gain. Here, we report the utility of a mid-density marker panel for genetic studies and other applications in cowpea breeding. The 2,602-marker panel was used to genotype 376 cowpea materials pooled from four different genetic backgrounds. The panel was informative with over 78% SNPs exceeding minor allele frequency of 0.20. The panel correctly deciphered co-ancestry among lines, identifying 0.04 % of all pairwise relationships as identical lines, 0.01% as parent-offspring, 0.12% as half-sibs, 39.19% as unrelated, and 60.64% with distant relationships. STRUCTURE, principal component analysis (PCA), and discriminant analysis of principal components (DAPC) inferred two major groups, with all the bi-parental RILs placed in a single gene pool. Excluding bi-parental RILs exposed sub-structure within the remaining diverse lines. Variance within populations was higher (16.64%) than that between populations (8.38%). Linkage disequilibrium (LD) decay was correctly depicted as being slower in bi-parental RILs than in other populations. Overall, LD dissipated to r2 = 0.1 at 1.25Mb. In addition, we mapped a region on chromosome VU07 known to be associated with both seed and flower colors in cowpea. This region harbors several genes including Vigun07g110700, a basic helix-loop-helix (bHLH) DNA-binding superfamily protein that regulates plant pigmentation. The panel revealed unexpected heterozygosity within some lines and authenticated the hybridity of F1 progenies. This study demonstrated the panel’s effectiveness for molecular applications in cowpea, and that the accessions that were used are genetically diverse and suitable for trait discovery and breeding.
Abiotic efflux of CO2 from soil is often attributed to dissolution of carbonates, and therefore not expected to occur in soils with a low pH. However, another abiotic source of CO2, less constrained by pH, may arise from reactions that oxidize natural soil organic matter and reduce metal oxides. Studies of redox reactions between phenolic compounds and Fe and Mn oxides in soil have been focused mainly on the environmental fate of both oxidants and reductants and formation of organic matter. We measured CO2 formed during 3-hour, room temperature (22±2 oC), incubations of samples of archived soils and from an ongoing crop diversity study. Subsamples (8 g. ODE) of each soil, were treated (5 ml) with water, or solutions of glucose (0.029 M), or gallic acid (0.025 M). For each soil, subsamples amended with H2O or with the glucose solution produced little CO2 and were nearly identical to each other, while CO2 quickly formed after treatment with gallic acid regardless of pH. The net increase in CO2 due to gallic acid, observed from the 18 archived soils, ranged from less than 0.5 to more than 80 mg CO2-C kg-1 soil. Significant treatment effects were observed in samples from the crop diversity study with more (Tukey’s P≤0.05) net CO2 from a small grain-fallow treatment compared a 5-year rotation treatment, 19.04 and 15.77 mg CO2-C kg-1 soil, respectively. This study suggests abiotic reactions capable of rapidly producing a burst of CO2 can occur in a wide range of soils following inputs of simple phenolic compounds and be impacted by management regimes. We suggest these are redox reactions in soil linked to Mn or Fe metal oxides and when considered together with fluctuations of carbon inputs to soil and redox cycling, might be a larger contributor to C emissions than previously accounted for.
Soil carbon is intimately related to the living part of the organic matter, as represented by the soil microbial biomass, which mediates the decomposition, mineralization, and immobilization of organic carbon available in soils under different land-use systems. Forest-to-agriculture conversion and land-use change often lead to a loss in microbial biomass carbon (MBC) and shifts in microbial activity, directly influencing the soil carbon dynamics. The main aim of this study was to evaluate the effects of land-use change and geographical distribution on the microbial and environmental patterns related to soil C-dynamics. We evaluated MBC and microbial respiration in soils under five different land-use systems and two contrasting seasons, at a regional scale in Santa Catarina State, Southern Brazil. At the west mesoregion, changes in the MBC were correlated to sampling season in forest and grassland systems. Yet at the plateau mesoregion, we observed a land-use effect, as MBC decreased in no-till and crop-livestock integration systems. At the two mesoregions, forest and grassland had presented the highest values of MBC and microbial activity, as represented by microbial respiration. The grassland sites have presented lower values of the metabolic quotient (qCO2) and higher values of the microbial quotient (qMic). The qCO2 was lower in winter for all land-use systems. The forest sites have shown the highest total and particulate organic carbon values. The chemical-physical characteristics have shown correlations with microbiological variables related to the soil microbial C-dynamics. The land-use intensity, season, and geographic location were the main drivers of changes in microbial C-dynamics.
The poor across the world is very vulnerable to floods and drought disasters and have a detrimental effect on the lives and livelihoods of the poor. Weather based index insurance is one of the ways of dealing with these disasters. Protecting against floods and providing risk cover against losses due to floods has been a major area of concern for any government. Risk transfer through insurance is an important component in managing agricultural risks from extreme flood events. The study developed the first of its kind of designing and implementation of an index-based flood insurance (IBFI) product with the advanced use of satellite data and flood models to estimate crop losses due to floods. IBFI insurance product uses two different data elements, the first one is based on the flood model using HEC-HMS and HEC-RAS that uses inputs from NASA GPM bias-corrected satellite rainfall estimates, observed water level and discharge data, river characteristics, and digital elevation model to generate flood depth and flood duration to develop pre-determined thresholds based on the historical flood events between 1991 to 2015 and the second IBFI product uses only satellite data from NASA MODIS Terra and Aqua satellite data and the Copernicus Sentinel-1 SAR data to generate flood depth and flood duration to develop pre-determined thresholds based on the historical flood events and economic losses. More than 7,000 farming households in Bihar (India) and northern Bangladesh have signed up for a pilot IBFI scheme, which went live in 2017. The participating farmers have received insurance compensation for crop losses of over USD 160,000. In addition to the insurance product implementation, the research evaluated the farming willingness to pay, developing business models for scaling; social equity, and economic benefits of derisk disasters. IBFI initiative promotes a closer linkage between risk transfer and risk reduction that could make this more sustainable and robust financial instruments for flood-affected communities and reducing the burden of post-disaster relief funds for the government. In summary, the index insurance using open-access satellite imagery is a win-win opportunity as it brings down the data development cost, lower insurance premium, quick settlement, and greater transparency among various users.
Increasing agricultural demand for freshwater in the face of a changing climate requires improved irrigation management to maximize resource efficiency. Soil water deficits can significantly reduce plant growth and development, directly impacting crop quantity and quality. Dendrometers are a plant-based tool that have shown potential to improve irrigation management in high-value woody perennial crops (e.g. trees and vines). A dendrometer continuously measures small fluctuations in stem diameter; this has been directly correlated to water stress. While plant-based measures of water deficits are the best indication of water stress, current dendrometers are imprecise due to mechanical hysteresis and thermal expansion. The high-precision dendrometer created at the OPEnS Lab alleviates these key failure points using zero-thermal expansion carbon fiber, zero friction via a spring tensioning approach, and a linear magnetic encoder. The device achieves 0.5-micron resolution, and thermal fluctuations are less than 1 micron over diurnal swings of 25°C. The cost of the device varies with build quantity; parts are $200 - $450 each and assembly requires 6 to 12 hours per system. Dendrometers are currently being deployed with telemetry based on LoRa, which is under evaluation. Without solar charging and telemetry, the battery is sufficient for over two years of operation. Mass deployment of these automated dendrometers has the potential to provide a continuous record of water stress driven changes in stems, providing valuable decision support for irrigation management.
Farmland mulching can maintain soil temperature and moisture, increase crop yields, which plays a positive role in agricultural planting in the arid area. However, the composition of the soil microbial community and the carbon cycle involved is not well understood under the ground cover scenario. Based on the Changwu Agro-Ecological Experiment Station, this study set up a long-term positioning Experiment to explore the impact of long-term land cover pattern on soil bacterial community structure and carbon metabolism capacity. In this study, we include five treatments: control uncovered (CK), plastic film mulching (F), high amount of straw mulching is carried out in July, August and September every year (St90), low amount of straw mulch throughout the growth period (S45), high amount of straw cover throughout the growth period (S90). We combined Illumina MiSeq sequencing and Biolog-ECO technology to analyze the functional characterization of soil bacterial community composition in terms of carbon source metabolism. The results indicated that the soil bacteria showed high degree of epistatic clustering after mulching treatments, and the operational taxonomic units (OTU) numbers of soil bacteria under different treatments decrease as St90, S45, F, CK, S90. Redundancy analysis (RDA) showed that AP, NH4+-N, MC were the main environmental factors affecting bacterial community composition. Proteobacteria, Acidobacteria and Actinobacteria are the main dominant flora in this area. Long-term surface cover resulted in significant differences in soil bacterial carbon metabolism, which was highest in the F treatment. Structural equation modeling (SEM) shows that surface cover has direct and indirect effects on bacterial carbon metabolism capacity and diversity through soil physicochemical properties. Reasonable mulching patterns can greatly change the community structure and carbon metabolism of soil bacteria, and play a positive role in the sustainable development of agro-ecosystems in arid area.
Modeling socio-ecological interactions are one of the essential requirements for water resource management in water-stressed areas. Mismanagement of water resource combined with extensive withdrawal by farmers in Indus Basin is putting pressure on freshwater resources. In some areas sever depletion of groundwater is evident. Waterlogging remains a bigger problems in the areas with higher surface water endowments, causing salinization; a greatest threat to long-term groundwater sustainability. Physical water management solution wouldn’t be a successful approach as it ignores water users’ behaviors; their interaction with each other and the feedback effects they receive from the system. We have developed an ABM model simulating the system by varying different agro-climatic parameters for water withdrawal behaviours of framers to substantiate a groundwater development framework in conjunction with the management of surface water. Overtime spatially distributed farmers’ caricaturized scenarios were built to include groundwater depth fluctuations for better management of water resources. Self-governing Rules (SGR) and Institutional Management Perspective(IMP) bring equity in water availability and prevent agriculture from worsening water quality parameters. However, consistency in the benefits may break down in extreme cases of climate change and spatio-physical conditions. Our water management perspectives provide improved outcomes of water withdrawal. SGR perspective managed to increase groundwater abstraction price 3 times more than the existing rates for the farmers located near water source. For the farmers located at tails IMP appears to manage resource better than other scenarios. Better and sustainable water withdrawal management requires to have area-wise policies and institutional support for promotion of norms.
Climate change is both a physical and social phenomenon in which individual understandings are contextualized within broader considerations: individuals are not ‘blank slates’ receiving information about climate change, but that information is always and inevitably filtered through values and worldviews. Personal experience, local knowledge, and social-learning influence climate risk perception and vary substantially among countries and regions. Likewise, they differently affect individuals and social groups at the regional and local scale, among whom exposures, attitudes, and capacities to manage risks vary greatly. A climate storyline approach is hence well-suited to study human observations, compound climate risks, and inform and conceptualize human–water systems interactions. Narrative storylines are used as input drivers to climate models, to represent different development pathways, which are usually characterized and applied at national and sub-national scales. Storylines aim to provide new social scenarios that address local human cognition uncertainties and improve human behavior modelling and robustness when addressing decision-making processes. Climate risks and hazards understanding can be communicated by presenting the experiences or a sequence of events, facts, and observations that are plausible and potentially critical for the system under study. Methods guiding storytelling are usually focused on conducting interviews with stakeholders, carrying out collective workshops, developing appropriate focal questions, and iterating between model results and key stakeholders. Therefore, can other data collection tools be used to reduce uncertainty in physical aspects of climate change from individuals’ local experience and perception? This contribution presents a triple-loop survey to detail the core elements of farmers’ perception and behavior when addressing climate change risk. We collect first-hand observations from northern Italian farmers about how climate change affects their activity and how extreme events are conditioning their adaptation capacity. Emphasis is placed on understanding the driving factors (risk awareness, perceived impacts, and adaptation measures and barriers) involved in the physically self-consistent past events and the plausibility of those factors. Moreover, we want to test if these factors can provide relevant implications for appropriately modelling storylines in decision-making processes. Tentative results can be useful to discuss the methodological framework of storylines building and narratives modelling, and at which point surveys can be an alternative and complementary way of dealing with deep uncertainty within climate risk management and social scenarios modelling.
Climate change is arguably the most severe and complex challenge facing today’s society, a cross-cutting issue affecting many sectors and connected to other global challenges, such as ensuring sustainable water management and food security. Agricultural systems are adversely influenced by climate change through increased water stress, change in run-off patterns, seasonality fluctuation, and temperature variations. Farmers are, hence, a valuable source of first-hand observations of climate change as they may provide a deeper understanding of their manifestation, relevance, and effects. Social and behavioural sciences have investigated the influence of farmers’ experiences in increasing climate change adaptation capability and improving decision-making processes at the system level. The conclusion is that local perceptions provide sufficient baseline information for understanding individual and collective exposure to climate risks, an essential element for effective policy formulation and implementation. Traditional management approaches based on simple, linear growth optimization strategies, overseen by command-and-control policies, have proven inadequate for effective adaptation to climate change. Conversely, accurate bottom-up approaches focused on social learning can complement the system transformation by building collaborative problem solving among individuals, stakeholders, and decision-makers. In this context, deepening social perception becomes fundamental for two main reasons: i) it is a key component of the socio-political context, and ii) it is an essential step for behaviour transformation and attitude change. In this line, associative processing methods, such as interviews and surveys, have been discussed for their ability to monitor the nature, extent, significance, and influence of personal experience on climate change adaptation. Also, modelling techniques have been recognized in social sciences as effective mechanisms to simulate the social influence in decision-making processes. System dynamics (e.g., causal loop diagrams, CLD) and Agent-Based Models (ABM) can include feedback between social and physical environments, define individuals’ and stakeholders’ narratives, and map the social network with agents’ interactions. This proposal aims at testing how qualitative data can enable policy-makers and managers to understand and re-think water management and climate change policies at the local level, which is essential to address agricultural risks. From a system dynamics approach, we examine how ABMs can most effectively integrate behavioural data collected from semi-structured interviews and surveys to increase robustness in decision-making processes while attending to farmers’ behaviour on climate change adaptation. We surveyed 460 farmers and semi-structured interviews with 13 irrigation consortiums from northern Italy to deepen a triple loop analysis on climate change awareness, perceived impacts, and adaptive capacity.
The Surface Biology and Geology (SBG) mission is one of the core missions of NASA’s Earth System Observatory (ESO). SBG will acquire high resolution solar-reflected spectroscopy and thermal infrared observations at a data rate of ~10 TB/day and generate products at ~75 TB/day. As the per-day volume is greater than NASA’s total extant airborne hyperspectral data collection, collecting, processing/re- processing, disseminating, and exploiting the SBG data presents new challenges. To address these challenges, we are developing a prototype science pipeline and a full-volume global hyperspectral synthetic data set to help prepare for SBG’s flight. Our science pipeline is based on the science processing operations technology developed for the Kepler and TESS planet-hunting missions. The pipeline infrastructure, Ziggy, provides a scalable architecture for robust, repeatable, and replicable science and application products that can be run on a range of systems from a laptop to the cloud or an on-site supercomputer. Our effort began by ingesting data and applying workflows from the EO- 1/Hyperion 17-year mission archive that provides globally sampled visible through shortwave infrared spectra that are representative of SBG data types and volumes. We have fully implemented the first stage of processing, from the raw data (Level 0) to top-of-the-atmosphere radiance (Level 1R). We plan to begin reprocessing the entire 55 TB Hyperion data set by the end of 2021. Work to implement an atmospheric correction module to convert the L1R data to surface reflectance (Level 2) is also underway. Additionally, an effort to develop a hybrid High Performance Computing (HPC)/cloud processing framework has been started to help optimize the cost, processing throughput and overall system resiliency for SBG’s science data system (SDS). Separately, we have developed a method for generating full-volume synthetic data sets for SBG based on MODIS data and have made the first version of this data set available to the community on the data portal of NASA’s Advanced Supercomputing Division at NASA Ames Research Center. The synthetic data will make it possible to test parts of the pipeline infrastructure and other software to be applied for product generation.
Soil moisture plays an essential role in the complex eco-hydrologic processes, such as infiltration, rainfall-evapotranspiration-runoff circulation, photosynthesis, and groundwater recharge. However, the accurate estimation of soil moisture (SM) at regional or larger scale is difficult because SM varies highly over space and time due to heterogeneous land cover and soil properties, and ground measurements are often time-consuming and expensive. Currently, Bangladesh Meteorological Department (BMD) measures SM only at twelve stations which is quite inadequate for assessing large-scale spatial and temporal variation of SM. Thus, satellite-derived soil moisture data products or Global Land Data Assimilation System simulated (GLDAS-2.2) soil moisture dataset with the Gravity Recovery and Climate Experiment Data Assimilation (GRACE-DA) can be promising alternatives to the in-situ measurement for this data-scarce region. In this study, the spatial and temporal variations of SM from GLDAS and Soil Moisture Active Passive (SMAP) satellite were compared against the in-situ measurements from seven agrometeorological stations of Bangladesh. The GLDAS and SMAP products overpredicted the in-situ SM for most of the stations and could capture the temporal dynamics of observed SM with correlation coefficient (R) of 0.36 and 0.17, respectively. Later an Artificial Neural Network model was developed based on soil moisture from both sources (SMAP and GLDAS) and terrestrial water storage from GLDAS to obtain more accurate estimation of SM for this data-scarce region. The ANN model shows an improvement in estimation and predicted SM with R = 0.63 (considering all stations). The results were more promising when separate model is developed for each study site. Incorporating additional climate data (such as precipitation with different lag times) as input improved the accuracy marginally. This study suggests that the release of daily GRACE gravity field solutions in near-real-time may provide a reasonable and continuous estimate of soil moisture in this data-scarce region.
Soil and Water Assessment Tool (SWAT) is one of the widely used hydrological models, especially it has been successfully applied for the assessment of the impact of land use land cover and best management practices scenarios. But it is less applicable for type of research that requires integration or optimization with other models since it cannot update the land-use and best management practice information efficiently and it should be run separately when results for the multiple watersheds are needed. These days, the attention on the water security has been growing and interdisciplinary works desires integration of the hydrological model with other models are being highlighted. Thus, there are needs of development of the surrogate model which is computationally efficient and applicable to multiple watersheds. In this research, we propose the surrogate model of the SWAT with novel machine learning techniques such as random forest model. As a first step, the models are trained with the SWAT data from one watershed, which is a Maumee River Watersheds. Models for flow, mineral phosphorus, total nitrogen, total phosphorus, and sediment transport are built separately, and the model performance was above satisfactory level based on R-squared value, Nash-Sutcliffe efficiency, and percent bias. In addition, the surrogate models were tested for the different best management practices adoption scenarios and were trained additional data to make the model valid for the wide range of the best management practices adoption ratios. Finally, the surrogate models were expanded to multiple watersheds, by training SWAT results from Huron River Watersheds and River Raisin and they evaluated with the R-squared value. High R-squared values indicated that the surrogate model could be used in place of SWAT.
Nitrous oxide (N2O) is one of the important greenhouse gases (GHGs), with its global warming potential 265 times greater than that of carbon dioxide (CO2). About 60% of the anthropogenic N2O emission is from agriculture production. To date, estimating N2O emissions from cropland remains a challenging task because the related microbial origin processes (e.g. incomplete nitrification and denitrification) are controlled by a diverse factors of climate, soil, plant and human activities. In this study, we developed a ML model with physical/biogeochemical domain knowledge, namely knowledge guided machine learning (KGML), for simulating daily N2O fluxes from the agriculture ecosystem. The Gated Recurrent Unit (GRU) was used as the basis to build the model structure. A range of ideas have been implemented to optimize the model performance, including 1) hierarchical structure based on variable causal relations, 2) intermediate variable (IMV) prediction and transfer, 3) inputting IMV initials for constraints, 4) model pretrain/retrain, and 5) multitask learning. The developed KGML was pre-trained by millions of synthetic data generated by an advanced PB model, ecosys, and then re-trained by observations from six mesocosm chambers during three growing seasons. Six other pure ML models were developed using the same data from mesocosm chambers to serve as the benchmark for the KGML model. The results show that KGML can always outperform the PB model in efficiency and ML models in prediction accuracy of capturing N2O flux magnitude and dynamics. Besides, the reasonable predictions of IMVs increase the interpretability of KGML. We believe the footprint of KGML development in this study will stimulate a new body of research on interpretable machine learning for biogeochemistry and other related geoscience processes.
Improving the estimation of CO2 exchange between the atmosphere and terrestrial ecosystems is critical to reducing the large uncertainty in the global carbon budget. Large amounts of the atmospheric CO2 assimilated by plants return to the atmosphere by ecosystem respiration (Reco), including plant autotrophic respiration (Ra) and soil microbial heterotrophic respiration (Rh). However, Ra and Rh are challenging to be estimated at large regional scales because of the limited understanding of the complex interactions among physical, chemical, and biological processes and the resulting high spatio-temporal dynamics. Traditional approaches for estimating Reco including process-based (PB) models are limited by human knowledge resulting in limited accuracy and efficiency. Accumulation of the in situ observation of net ecosystem exchange (NEE), weather, and soil, and satellite data of GPP, LAI and soil moisture make it possible for applying data driven machine learning (ML) approaches. But the ML model approach has disadvantages of omission of domain knowledge and lack of interpretability. Here we propose a novel knowledge guided machine learning (KGML) method for predicting daily Ra and Rh in the US crop fields. With Gated Recurrent Unit (GRU) as the basis, we develop the KGML models constructing the hierarchical structure of ML with a mass balance constraint. The KGML models were pre-trained using synthetic data generated by an advanced agroecosystem model, ecosys, and re-trained with real-world FLUXNET observation data. We extrapolate the best KGML model to crop fields over the US with the help of satellite data, reanalysis climate forcings, and soil database to reveal the spatio-temporal variations and key controlling factors. We believe this study advances the interpretable machine learning concept for carbon cycle estimation and will shed light on many other process-based biogeochemistry research.