The lack of location-specific information for both model input and model constraints thus is the largest uncertainty in quantifying field-level carbon outcomes . For any technology used for carbon outcome quantification, there is a trade off between cost and accuracy . Although no clear criterion has been established so far to accept or reject a technology, for any quantification technology to be scalable, its per-acre operational cost must be meaningfully lower or significantly lower than the expected monetized carbon values from adopting climate-smart practices. In the current U.S. agriculture carbon market with a carbon price of roughly US $20/t CO2e, for example, this criterion, based on the DOE ARPA-E estimation , means costs should be significantly lower than $10/acre/year for soil carbon and $50/acre/year for N2O quantification for large-scale deployment, including installation, calibration, operation, and hardware lifetime and at the same time, the technology should be able to achieve less than 20% error at the field level . No single existing technology can meet both of these expectations. Instead, we propose that a more viable path for quantification of field-level carbon outcomes in agricultural soils is through an integration of sampling, sensing, and modeling, defined as the “System-of-Systems” solution. The “System-of-Systems” concept means that the complex problem of quantifying agroecosystem carbon outcomes cannot be solved by using a single sensor or a model alone, but only can be solved by effectively integrating various approaches . Such a “System-of-Systems” solution should simultaneously comprise the following features : scalable collection of ground truth data and cross-scale sensing of E, M, and C at the local field level; advanced modeling with necessary processes to support the quantification of carbon outcomes; systematic Model-Data Fusion , i.e. robust and efficient methods to integrate sensing data and models at each local farmland level; high computation efficiency and AI to scale to millions of individual fields with low cost; robust and multi-tier validation systems and infrastructures to test model/solution’s scalability, defined as the ability of a solution to perform robustly with accepted accuracy on all targeted fields.

Thus the “System-of-Systems” solution is a holistic framework including multiple sub-systems for sensing, monitoring, modeling, and model-data fusion,hydroponi bucket targeting to assure field-level accuracy, scalability, and cost-effectiveness. The “System-of-Systems” approach is so far the only pathway to implement the mass-balance approach to quantify SOC changes, which requires various localized observations and the integration of observations/data with models to accurately estimate each term in the massbalance equation and achieve the field-level accuracy. Compared with existing approaches , there are several advantages of using the mass-balance approach to quantify the change of SOC. First, all of the carbon budget terms are measurable, although some being costly, and can be used to verify model accuracy and provide a basis for confidence. Second, all the carbon budget terms can be measured and verified at relatively short time scales, i.e. from sub-hourly scale to annual time scale , which enables the quantification of annual change of SOC. In contrast, soil sampling is generally not able to detect annual changes, as the uncertainty of soil sampling is usually much larger than the annual change of SOC. Third, those carbon budget terms for calculating the carbon input to soil can be estimated using advanced remote sensing technologies , which offers an efficient and scalable way to achieve the field-level observational constraints in a large region due to the ubiquitous coverage of remote sensing technologies. Fourth, the carbon mass balance approach provides a holistic picture of the overall carbon budget of farmland soils, which enables a mechanistic understanding of differential impacts of management practices on SOC from field to field and from year to year, thereby could help farmers to improve their management practices along with the changing climate. Scalably sensing/estimating local information of E, M, and C at the field level is the first step of a “System-of-Systems” solution, which involves two seemingly different but inherently connected tasks: ground truth collection, and cross-scale sensing. Ground truth here is broadly defined as information that is collected on the ground to train, constrain or validate models. Agricultural ground truth is scarce and expensive to collect.

For example, collecting carbon flux data requires eddy-covariance flux towers, which are generally costly to set up and operate. However, we also have to face the reality that even with low-cost sensing technology or crowd sourcing efforts, one cannot collect ground truth for every field. Instead, we propose to develop “cross-scale sensing” approaches, especially those enabled by remote sensing, to scale-up “ground truth” collection to large scales. Cross-scale sensing can be demonstrated by the most recent development of deriving field-level photosynthesis information. Photosynthesis is the only term for land carbon input and also the largest carbon budget term . Ecosystem photosynthesis is the primary driver for crop litter and thus significantly contributes to the long-term change in SOC, as demonstrated in Section 2.3. Correctly quantifying photosynthesis at the field level puts significant constraint and reduces uncertainty on simulated crop carbon dynamics, crop litter and soil carbon dynamics . A recent breakthrough in the remote sensing of photosynthesis was made possible by full integration of leaf level chamber/sensor measurements, canopy-level hyperspectral sensing , and regional-scale mapping through satellite fusion data . The cross-scale sensing here is guided by the domain knowledge of plant physiology, radiative transfer modeling, and hyperspectral theories; ground truth data – in particular, leaf-level samples and eddy-covariance flux tower data – are extensively used in the model development stage, but once the translation from ground-truth data to satellite-scale signals can be robustly developed, satellite fusion data can expand the photosynthesis information for every single field every day since 2000 to present . Another advance in cross-scale sensing is the use of intermediate sensing to augment traditional ground truth collection, and enable the scaling from leaf-level or plot-level ground measurements to coarse satellite pixel size – a classic problem in the area of remote sensing. A typical example is the use of airborne hyperspectral imaging . Hyperspectral imaging can provide estimates of certain soil and plant traits with high accuracy , although its application for scalable mapping has been limited by its high cost.

A novel use of AHI is to treat AHI data as an intermediate bridge between ground truth collection and satellite scale-up. A general procedure is to first develop robust methods to translate AHI signals with targeted estimates based on data from intensive lab and field experiments; and then to use AHI as a strategic sampler to selectively “sample” over space and time, serving as a bridge from granular resolution of ground truth to large satellite pixels; and finally, to use satellite data overlaid with the AHI sampled area to translate satellite multi-spectral signals along with environmental variables into plant and soil trait estimation, thus deriving targeted E, M, C variables ubiquitously using satellite data. Though similar approaches have achieved success in mapping forests canopy bio-geochemistry , they have rarely been used in agroecosystems. Once advanced and automated pipelines are established to conduct AHI collection and data processing , AHI can be applied to estimate crop canopy nitrogen content, cover crop biomass, and crop residue fraction and tillage practices. Fig. 7 demonstrates how AHI is used to scale up the estimation of crop residue fraction and tillage intensity at the regional scale. Other sensing approaches, such as mobile vehicle sensing , IoT sensing network and robotics , could also achieve a similar function to augment ground truth collection and enable satellite scaling-up to regional scales. Table 1 provides a non-inclusive list of different critical E, M, C variables that currently have been estimated using cross-scale sensing technologies. Have sufficient and necessary processes represented. Coupled carbon-nutrient-water-energy cycling over farmland is the foundation for field-level carbon outcome quantification,stackable planters thus models should include a sufficient number of mechanistic pathways that clearly track the input, output and storage of water, carbon, nutrient and energy in crop lands under the interference of agricultural management. For the plant component, simulating the responses of crop carbon uptake and water use to different abiotic and biotic stresses is necessary as they largely determine the crop production and carbon input to the soil. From this perspective, proper representation of canopy energy balance, stomatal conductance, uptake and transport of water and nutrients from soil to canopy are needed to mechanistically simulate the crop carbon and nutrient uptake and crop water use . Many of the existing process-based models may lack critical processes or use over-simplified processes to model specific carbon outcomes. One obvious example, following our prior discussion on the importance of the holistic carbon budget of agroecosystems, is that most existing process-based models lack sufficient mechanisms that can model plant carbon processes as emergent phenomenon , resulting in significant errors when quantifying the downstream ΔSOC. For example, lack of explicit modeling of photosynthesis , plant stomatal responses to environmental stresses , and reproductive processes for yield can cause huge uncertainty of the modeled carbon input to the soil pools, contributing significant error to the simulated ΔSOC. For the below ground part, soil temperature, water, oxygen, and pH dynamics, bio-geochemical reactions related to carbon, nitrogen and phosphorus cycling, microbial activities and their regulation on SOM formation and stabilization as well as GHG emissions are core processes that need to be simulated. For example, recent studies identified two distinct pathways of SOM stabilization from litter decomposition, i.e. the DOM-microbial pathway in the early stage of decomposition, and the physical transfer pathway in the final stage of decomposition .

This work emphasized the importance of dissolved organic matter and microbial activities, and necromass in stabilizing SOM . Having those mechanisms and their interactions with related environmental drivers well represented in the soil carbon models is essential to accurately simulate the dynamics of SOC and its physical fractionations. Besides these biophysical and bio-geochemical processes, representing the farming management practices and their impacts on coupled carbon nutrient-water-energy cycling over farmland is critically needed to quantify the carbon outcomes. Neverthless, there should be a good balance between model complexity and practicality. Any model used for operational carbon outcomes quantification should have necessary complexity and processes, and new theoretical advances in science should be ultimately incorporated into existing models to improve representations of relevant processes. However, we also need to acknowledge that models with new mechanistic representations are not always better than simpler models in practice, especially when there is not enough data to constrain those new mechanistic representations. When evaluating the appropriate model structures for agricultural carbon outcomes, we should focus on two fundamental questions: Is a specific process indispensable for simulating the specific outcome and also achieving the desired accuracy? Is there sufficient data to parameterize that specific process at both field and regional scales? If the answer to either question is no, then including the new process may not necessarily benefit the quantification of carbon outcome for now. Maximum use of mechanistic process representation. To simulate biogeochemical and biogeophysical processes, many existing models use multiplication factors , law of the minimum , and empirically-derived response functions , all of which are ad hoc by nature. One consequence of these non-mechanistic modeling approaches is that different researchers applying the same method to a given process will obtain different mathematical representations, which then lead to a loose foundation to implement that particular process in these models . Moreover, non-mechanistic representation which lacks support from physical laws also limits the generality and scalability of the model simulations, especially when a model is used to extrapolate beyond the environmental and management conditions under which the model is previously developed or calibrated. For example, many models use the empirically-derived soil water stress functions to depict the down-regulation of crop carbon uptake and water use under water stress conditions, which causes inconsistencies and discrepancies in multi-model intercomparison simulations . A more mechanistic way to account for crop soil water stress would be to explicitly represent the plant-hydraulic-stomatal-photosynthetic coordination from soils to plant, and to atmosphere . Similarly, most models formulate soil carbon decomposition rate by assuming different controlling factors independently and multiplicatively scale the decomposition rate ; in reality, these factors are interacting and intertwined following specific mechanistic pathways to lead to decomposition rate, but very few existing models include such interactions and mechanistic pathways . Another example is how the impacts of different tillage practices are represented on soil physical and biogeochemical processes.