Recipient Organization
UNIVERSITY OF COLORADO
(N/A)
BOULDER,CO 80309
Performing Department
School of Arts and Sciences
Non Technical Summary
Nearly one third of soils globally are degraded and an even larger fraction is at risk of degradation. The decline in soil health and function threatens food security, ecological function and places large portions of the global south dependent on agriculture at risk of economic and ecological decline. Our work will provide a completely new approach to the monitoring of degradation and restoration while training a new cohort of engineering and environmental science graduate students in the US and a cross-disciplinary group of postdoctoral fellows in the UK bringing together soil and data science. At both the US and UK sites, we will engage in new forms of science communication including a live link to the data (and interpretation of changes) at the sites under active management.In this project focus on detecting soil degradation and restoration through a novel multi-functional soil sensing platform that combines conventional and newly created sensors and a machine learning framework. Our proposed work directly addresses the Signals in the Soil call to 'advance our understanding of dynamic soil processes that operate at different temporal/spatial scales.' Through the creation of an innovative new approach to capturing and analyzing high frequency data from in-situ sensors, this project will predict the rate and direction of soil system functions for sites undergoing degradation or restoration. To do this, we will build and train a new mechanistically-informed machine learning system to turn high frequency data on multiple soil functions, such as water infiltration, CO2 production, and surface soil movement, into predictions of longer term changes in soil health including the status of microbial processes, soil organic matter (SOM) content, and other properties and processes. Such an approach could be transformative: a system that will allow short-term sensor data to be used to evaluate longer term soil transformations in key ecosystem functions. We will start our work with a suite of off-the-shelf sensors observing multiple soil functions that can be installed quickly. These data will allow us to rapidly initiate development and training of a novel mechanistically informed machine learning framework. In parallel we will develop two new soil health sensors focused on in-situ real time measurement of decomposition rates and transformation of soil color that reflects the accumulation or loss of SOM. We will then link these new sensors with a suite of conventional sensors in a novel data collection and networking system coupled to the Swarm satellite network to create a low cost sensor array that can be deployed in remote areas and used to support studies of soil degradation or progress toward restoration worldwide.This proposal addresses one of the most pressing issues in the global environmental community by creating a novel multi-functional soil sensing platform that can be used for the early detection of soil degradation and restoration. The work proposed here will generate new insights into the behavior of soils that are undergoing degradation and restoration while also providing a completely new approach to site monitoring, evaluation and management.
Animal Health Component
50%
Research Effort Categories
Basic
20%
Applied
50%
Developmental
30%
Goals / Objectives
Our research program is organized around the deep and complementary integration of research teams and study sites close to the University of Colorado, Boulder, US, and in the Northwest of England, UK. Our group includes a multidisciplinary team of soil scientists, ecologists, engineers, and statisticians that will develop and deploy a wholly new approach to the use of in-situ soil data streams and analytics to evaluate the rate and direction of soil system change. Our proposed work is organized into four objectives (work packages) with specific engineering or data system deliverables and associated hypothesesObjective 1 - Develop and deploy high frequency in-situ soil sensing platform: Deploy a new in-situ high frequency measurement package to sites in the US and UK to capture high frequency physical, chemical, and biological soil properties. This package will include a mix of conventional (moisture, temperature, CO2, hydrologic) sensors as well an unconventional capability developed by this group to use digital cameras to monitor surface soil particle movement.Objective 2 - MIMLA: Develop and deploy a new Mechanistically-Informed Machine Learning Approach (MIMLA) capable of assimilating multiple high temporal frequency data streams and combining these with mechanistic understanding to generate predictive analytics on the rate and direction of soil system change. MIMLA will be trained initially with the data from prior work and the high frequency data stream developed in Objective 1. The model will be tested against (through the experimental period) conventional point data (e.g. not from sensors) focused on a suite of soil health indicators.Objective 3 - Novel in-situ sensors and Internet of Things (IoT) connectivity. We will develop two new sensors focused on capturing key attributes of soil health. These include a new in-situ color detection sensor and in-situ decomposition sensor. These sensors, plus a select set of conventional sensors, will be linked together in a custom designed sensor array and data collection system developed at CU Boulder and networked to the SWARM satellite network to enable cloud-based data collection and processing.Objective 4 - Testing and scaling: Evaluate the success of the system developed in Objectives 1-3 in US and UK sites with contrasting soil and environmental conditions, and management characteristics. This test of the ability of our system to capture soil change in new (e.g. untrained) sites is a key step toward our longer term goal of developing a low cost, networked system that can be deployed in remote sites around the globe
Project Methods
MIMLA: The MIMLA will be developed from multiple soil multi-functionality building blocks. A building block is a mathematical representation of a well understood process in the soil. In an ecosystem model, the building blocks would be subroutines or equations embedded within the overall model structure. In the approach proposed here, a building block is a mathematical conceptualization of a well understood aspect of soil science that is run prior to the introduction of this information to the MIMLA. A building block also serves to provide the analytical linkages between high and low frequency soil response variables. As information from the sensor network enters the MIMLA, it moves through the building blocks prior to introduction to the machine learning portion of the MIMLA and finally through statistical 'changepoint detection' algorithms to evaluate the rate of change in processed variables. Through this process, the MIMLA combines these building block models to create a larger picture of soil health. At each time point, the machine learning framework selects an appropriate combination of the building blocks, taking account of the correlation between the blocks, to predict that status and direction of change in soil health. This approach balances the conceptual power of mechanistic insight with the predictive power of machine learning.Degradation sensor: The proposed approach is to use composite conductors formed from blends of degradable polymeric binders, such as poly(lactic acid), poly(caprolactone), poly(hydroxyalkanoates), poly(vinyl alcohol), poly(ethylene oxide), starches, celluloses, etc. and conductive, water-soluble, oxide-forming metal particles such as tungsten, zinc, magnesium or iron. These conductive composites will be printed onto substrates and encapsulated within films of materials (such as those described above) that degrade due to microbial activity. The sensor is buried in the soil at an appropriate depth, and when the substrate or encapsulating film is breached the conductive trace will fail, providing a simple binary signal that can be read either passively either in a chipless fashion as a change in the properties of a reflected RF signal from an inductively coupled reader coil (Huang et al. 2016), using a passive RFID approach (Mei et al. 2017), or combined with a power source and an appropriate integrated circuit for active wireless readout.Soil color sensor: A small, low-cost, complementary metal-oxide semiconductor (CMOS) camera, similar to those found in smartphones will be used alongside an LED light source to capture images of soil. Depending on requirements, these cameras and light sources can be combined with a power source and RF communication electronics and mounted onto a fixed post to take images of the soil surface or placed in a housing and buried in the soil at an appropriate depth (likely 6-12 inches). Soil images will be captured at various illumination levels in order to account for different reflectance properties and ambient lighting conditions.IOT System: The basic structure of the system iwill network accessory sensors in a hub and spoke approach with our custom engineered hardware system. The "hub" is the base unit that is powered with a solar panel, rechargeable battery, has a microcontroller, a satellite or cellular radio, and a low bandwidth accessory radio. This base unit is mounted outdoors, with a clear view of the sky without any plumbing, electrical or data hookups required. We then apply an accessory sensor approach, which are either self-powered, low powered, or solar powered, and use low power wireless transmission to the base unit. In this work, we will adapt off-the-shelf soil moisture, temperature as well as our newly developed decomposition sensors as IoT enabled accessory sensors, enabling remote monitoring and analytics of site-specific data.Field Testing: Our project will be deployed at two sites: one in the US and one in the UK. The US site is a heavily degraded semi-arid grassland outside Boulder, CO, whereas the UK focal site is long-term grassland restoration site in Yorkshire Dales, northern England, designed to identify optimal approaches for ecosystem service restoration on intensively managed, degraded grassland. The sites were selected for their proximity to the two lead institutions and because both sites are undergoing periods of rapid soil change as described below. These rapid changes will provide an ideal test case for the core hypothesis in this proposal. Instrumentation will be deployed at both sites in the following design. In summer of 2020, two paired instrumentation arrays will be deployed into a previously selected location at both sites. The two array design allows for sensor redundancy and ensure continuous measurements during the critical early stages of this proposal. The first arrays will use conventional soil hydrologic, thermal, chemical, and CO2 sensors and will generate data to train the MIMLA. The newly designed sensors and IoT Swarm satellite networked stations will be deployed as they become available.