A holistic upscaling from the point source to the landscape scale requires incorporation of several interacting, complex components, adding substantial complexity above and beyond the agricultural system itself. Thus, a major consideration in environmental modeling is how to best capture essential interactions while maintaining models that are feasible to implement with available data and computational resources. Fig. 4 illustrates the various components linking point to landscape scales. A first element for the linkage from point to landscape is estimation of surface and subsurface fluxes and ecological transitions along the lateral scale. Coupling with landscape microclimate models provides the vertical inputs used by the agricultural systems models, as well as gradients along the landscape. Coupling with hydrological models provides water flow paths like surface runoff, vertical and lateral groundwater flow, and interactions between vadose and groundwater zones and with adjacent surface water bodies . Water quality models provides sediment and solute transport along the landscape controlled by water flows , and other effects like wind erosion. Integration and upscaling of landscapes into the watershed scale requires 3-dimensional coupling of the surface and subsurface water, energy and mass transfers. At this scale, the groundwater aquifer system typically transcends the boundaries of the watershed and necessitates analysis at the regional scale to evaluate not only the impacts of the cropping and animal production systems on water quantity and quality, but also feed backs from the hydrological system in the agricultural system . Further, mesoscale rainfall and evapotranspiration distribution models control the local surface and subsurface flow intensities,hydroponic pots pollution and abatement. At this scale, human effects through land-use changes as well as ecological dynamics and transitions on natural or protected lands are also an important and critical component to evaluate the overall sustainability of the agricultural system.
Current crop modeling upscaling approaches based on land use maps can be considered an efficient first-order approximation of the environmental linkages. For example, in the USA the US Geological Survey hierarchical system of Hydrologic Units Codes is commonly used as the reference spatial mapping system to link spatially-explicit hydrological and crop yield simulations. Srinivasan et al. applied the Soil and Water Assessment Tool model to 8-digit subbasin HUCs in the Upper Mississippi River Basin and compared yields of the main crops with observed county-level USDA National Agricultural Statistical Survey data obtained for 1991–2001. SWAT uses spatially distributed watershed inputs to simulate hydrology, sediment and contaminant transport and cycling in soils and streams, and crop/vegetative uptake, growth and yields. Because many counties in the NASS database have missing data it was necessary to aggregate the crop yield data and simulation results to 4-digit sub-region HUCs . In general SWAT predicted crop yields satisfactorily over the long-term average for most 4-digit HUC , although errors greater than 20% were found for 14% of the HUCs studied. Further information on crop management may improve SWAT’s perform conclude that these errors stem likely from those predicting AET and soil moisture storage at these large aggregated scale, and “one could extend the validity and confidence in the model prediction of AET and soil moisture using a well-compared model on crop yield” . Thus, next generation models should consider the lateral connections through the landscape and regional scales to evaluate the sustainability of the integrated system, including effects on water and soil resources quality and quantity and ecological value. Although model complexity has increased in recent years and is a natural outcome of the proposed next generation integrated modeling, there has been little work to rigorously characterize the threshold of relevance in integrated and complex models. Formally assessing the relevance of the model in the face of increasing complexity would be valuable because there is growing unease among developers and users of complex models about the cumulative effects of various sources of uncertainty on model outputs .
New approaches have been proposed recently to evaluate the uncertainty-complexity relevance modeling trilemma , or to identify which parts of a model are redundant in particular simulations . Innovative approaches to simplify model outcomes to make them relevant in decision-making will be central to the next generation modeling efforts. For example, the identification of processes that do not influence particular scenarios, and the use of meta models, could allow simplification without affecting results .This thesis is divided into two sections. In the first section, a technical primer is given to provide a starting point for readers interested in sensor, printing, and machine learning technologies. In the second section, these three technologies are combined to demonstrate my dissertation work in developing nitrogen sensor nodes for precision farming applications. Grain growers apply on the order of a hundred to a few hundred pounds of nitrogen per acre, depending on the crop and field conditions. At a cost of tens of cents to a dollar per pound, with prices rapidly increasing in recent months, it is the second highest cost for many crops, outdone only by seeds. Nitrate fertilizer is conventionally applied uniformly across a field despite studies that have shown existing nitrate concentration in the soil can vary significantly on the order of tens of meters. Precision agriculture practitioners aim to designate site-specific management zones to direct more efficient nitrogen application, but the tools they have to gather data are limited. Optical remote sensing can be used to estimate nitrogen in growing plant material, but to get measurements of nitrate in the soil, a soil sample must be collected and taken back to a laboratory, for analysis via chromatography or spectrographic methods. Such measurements are highly accurate, but they are also expensive, labor-intensive, and give data for only one point in time and space. Nitrate is highly mobile, so concentrations change over time. Models can be developed to estimate nitrate fluxes based on measurements at the beginning and end of a season, but these rely on many estimations and assumptions. Environmental quality monitoring and precision agriculture require nitrate sensors that are robust enough to survive field deployment and soil insertion, can be mass-produced, and involve few or no moving parts. Additionally, the data must be simple to read. Printed solid-state potentiometric ion-selective electrode sensors have the potential to meet these criteria. The use of printing methods for sensor fabrication offers several advantages such as low cost, high throughput, and ease of fabrication.
In order to realize the benefits of printing and enable large-scale sensor deployment, both electrodes must be printed.Previous works have shown printed nitrate ISEs for use in aqueous environments and agriculture. Dam et al.demonstrated potentiometric nitrate sensors having a screen-printed nitrate ISE paired with a commercial RE for agriculture applications. Similarly, inkjet-printed nitrate ISEs were reported by Jiang et al. using a commercial Ag/AgCl reference electrode during measurements. In this work, we demonstrate fully printed, potentiometric nitrate sensors and characterized their sensitivity, selectivity, and stability. We then integrated the sensors into a wireless sensor node and characterized its sensitivity to nitrate concentration and moisture levels in the soil. We then replaced the components of the nitrate sensor node with naturally degradable components and characterized the devices. We propose a model-driven paradigm of measuring these sensors using swarms of UAV drones whose flight paths are optimized using machine learning. Finally, we demonstrate the need for sensor arrays to account for the interference that different analytes could cause,grow pot and provide preliminary results for a nitrogen sensor array that measures nitrate, nitrite, and ammonium concentration in aqueous solutions.A sensor is a device that is able to detect and measure some physical quantity of interest and communicate that information to another device or person. A common example of a sensor that you might recognize is the liquid-in-glass thermometer, shown in Figure 1.1. In this type of thermometer, a thin glass tube is filled with a small volume of liquid mercury that collects in a bulb at the bottom. Then, as temperature changes, the mercury expands or contracts in response, causing the peak of the mercury to move up or down the long stem. The stem, as you may know, is calibrated and marked with numbers corresponding to the temperature in Fahrenheit, Celsius, or both. In this example, the sensor detects the change in temperature by the liquid expanding or contracting in the glass tube. The temperature is measured and communicated to a person by the numbered ticks on the thermometer stem. A sensor should not be mistaken for a detector, which is able to detect and communicate a physical quantity, but fails to measure it. Consider for example a smoke detector, such as the one shown in Figure 1.1B. A smoke detector is able to detect whether or not there is smoke, but it doesn’t measure how much smoke there is. To a smoke detector, there is no distinction between a blazing house fire and overcooked salmon: both cause it to brazenly communicate its detection of smoke.Contrary to the thermometer example above, most modern sensors are electronic devices, which will be the type of sensor that will be discussed in this dissertation.
Many electronic sensors work by having some material that is sensitive to the physical quantity that is being measured, causing a property of that material to change with respect to the physical quantity. Other electronic sensors take advantage of natural laws, such as conductive metal wires arranged in a loop to measure the strength of the magnetic field that it’s in. Later in Section 1.2, we will go over the various types of sensors and the transduction mechanisms that a sensor might use. However, regardless of the mechanism or the physical property that is being measured, all electronic sensors have a sensing element that converts the signal of the physical quantity to an electric signal. As a quick aside, the opposite of a sensor is an actuator, which converts an electric signal into a mechanical action. Some common examples of actuators are electrical motors, hydraulic pumps, or pneumatic valves. Sensors and actuators are both transducers, which is a device that converts energy from one form to another. The distinction here is the intended purpose of the device: sensors measure and detect, while actuators perform an action. A more quantitative way of thinking of this is to look at energy conversion efficiency. The efficiency of energy conversion for sensors is immaterial because their purpose is to detect and measure. For example, if one sensor is 10% efficient at energy conversion but is less accurate than a second sensor that is only 2% efficient, then the second sensor is still an objectively better sensor of the two sensors because it is better at detecting and measuring. Contrarily, efficiency is an important metric for actuators because their purpose is to perform an action. An electric motor with a 2W load is objectively better than a motor with a 5W load, assuming they perform the same task equally well. The organization and classification of sensors vary throughout the academic literature and commercial marketplace. This is because there really is no perfect form of organization, as there are many ‘one-off’ devices that sense for some unique purpose or by some unique method. Further, there are many ways to categorize sensors: sensor specifications, sensor materials, transduction mechanisms, the quantity being measured, the field of application, whether the sensor is active or passive, direct or complex, and many, many more. It is analogous to classifying humans: humans can be classified by their age, gender, race, nationality, preferred sports team, favorite color, or the size of their ears. Similarly, sensors can be classified in many such ways.Sensors can also be classified as passive or active types. The distinction is simple: active sensors provide their own energy source to operate, while passive sensors use naturally available energy. An interesting example of both an active and a passive sensor is a camera. In a brightly lit location, natural light will illuminate the photographed subjects and then reflect toward the camera lens, where the camera simply records the radiation provided . In a dark room, however, there won’t be enough ambient light for the camera to record the subjects adequately. Instead, the camera uses its own energy source – the ‘flash’ – to illuminate the subjects and record the radiation reflected off them .