Performing Department
(N/A)
Non Technical Summary
Rangelands in western Great Plains are experiencing more frequent extreme drought, posing grand challenges to ecosystem services of this agriculturally important region. Monitoring and forecasting drought and predicting its impacts on forage production are critical for informing management decisions on sustaining ecosystem services via flexible livestock stocking under rapidly changing climates. Here, we will develop a high-resolution (30-m), near real-time monitoring and forecasting system of agroecosystem drought and forage production for the western Great Plains. Using soil moisture as a measure of agroecosystem drought and aboveground net primary productivity (ANPP) as the indicator of forage production, our work is unique and differs from previous approaches by (1) leveraging multiple streams of satellite and ground observation data, (2) employing different machine learning methods, and (3) co-producing tools with the ranching community to provide novel insights into the understanding of drought and its agroecosystem impacts. This monitoring and forecasting system will be built into the existing web tool, GrassCast, to provide weekly updated maps of monitoring and forecasting of agroecosystem drought and forage production. With our translational goal in providing a highly useful tool for monitoring and predicting agroecosystem drought and its impacts on forage production, this research meets the goal of Priority Area "Sustainable Agroecosystems: Health, Functions, Processes and Management Program" (A1451) to "improve ecosystem services in managed natural and agricultural systems".
Animal Health Component
0%
Research Effort Categories
Basic
50%
Applied
50%
Developmental
(N/A)
Goals / Objectives
This project aims to address drought and food security challenges faced by the western Great Plains rangeland agroecosystems by developing new insight and data products that overcome the limitations of past research. The overarching goal of our project is to develop, test and improve a monitoring and forecasting system of agroecosystem drought and forage production that is of high spatial resolution and accuracy, regularly updated, and easy to use.Specifically, we will use multi-stream satellite observations, ground data, and machine learning models to develop an empirical monitoring and forecasting framework for agroecosystem drought (indicated by soil moisture) and its impact on rangeland forage production (indicated by herbaceous aboveground net primary productivity, ANPP). Importantly, we will co-produce this framework by fostering key partnerships between researchers, extension specialists, and rangeland managers. This projectis supported by three specifical objectives: (1) New high-resolution (30-m) soil moisture products. These will be created by merging NASA's Soil Moisture Active-Passive mission (SMAP) data with Landsat-8 Operational Land Imager (OLI) imagery and by synthesizing multiple satellite soil moisture data products, each related to additional key environmental variables, to generate machine learning based forecasting models of agroecosystem drought. (2) Machine learning models that can predict herbaceous ANPP (forage production) based on climatic, soil, and vegetation variables; (3) A co-production approach by experts including scientists, ranchers, and other concerned stakeholders to collectively diagnose rangelands vulnerable to drought stress and mitigate drought impact.
Project Methods
This project includes three specific objectives, each with its particular methods.This first objective aims to generate new high-resolution soil moisture monitoring and forecasting products based on different satellite data, field data, and land surface modeling; while the forecasting framework will be based on machine learning models of soil moisture and associated climate, soil, and vegetation information.Here we will primarily employ soil moisture monitoring from the NASA SMAP mission to develop fine spatio-temporal resolution soil moisture monitoring maps for western Great Plains, which will also be complemented with existing synthesized data products.We will adopt two approaches to down-scale the original 9-km resolution SMAP L4_SM product to a 30-m resolution. The first approach is through fusing SMAP L4_SM with 30-m resolution Landsat-8 OLI data.In the second approach, we will use a method that integrates SMAP L3 data with the HydroBlocks land surface model to develop a high resolution (30-m resolution) integrated soil moisture product (SMAP-HB hereafter) for the western Great Plains.Furthermore, using this down-scaled 30-m resolution soil moisture products we would develop for historical periods, we will build a forecasting framework of soil moisture with machine learning methods, which will relate soil moisture to biogeophysical forcing variables.The second objective is to develop and test forage production monitoring and forecasting for western Great Plains rangeland using satellite data, machine learning, combined with weather and soil moisture forecasting.Here we will primarily use NDVI datasets generated from the Landsat-8 OLI (30-m resolution), and a Contiguous Solar-Induced chlorophyll Fluorescence (CSIF) data (0.05 o resolution, 2000-present) which was integrated from OCO-2 SIF and MODIS reflectance data to provide near real-time monitoring of forage production.In addition, we will use estimates of forage production measured in the field from ARS rangeland sites and those included in the LTER, LTAR, and NEON networks, to develop empirical relationships between NDVI or CSIF and forage production. Thiscan then be used to convert satellite monitored NDVI or CSIF to forage production.For the development of forecasting models, we will assess multiple climatic variables and soil and topographic information in this study.Satellite-derived production proxies and direct field measurements of forage production will also be used for developing and validating forecasting models.This third objective focuses on visualizing our products and results via public-accessible website and ensuring the easy-use of our products by the ranching community and other stakeholders.After training and validating, we will integrate our machine learning models into a modified version of GrassCast and compare predictions of NDVI and forage production with those made by the current implementation of GrassCast. We will compare these two versions of GrassCast in two ways. First, we will "backcast" and compare prediction accuracies of historical (1981- present) NDVI and forage production data using historical climate conditions. We will conduct this analysis at the site, grid and regional scales. Second, we will use the modified and original GrassCast versions in 2024 to nearcast forage production driven by near-term NOAA forecasts.We will adopt a co-production approach throughout the project lifespan. Here, stakeholders will not be only "informed" about the results and product outcome of the research. Rather, they will be actively participating from the very beginning of the project.We will further strengthen the researcher-rancher collaborative network through field visits to meet ranchers and interested stakeholders at the sites, discussing their needs and ensuring that the products are easily understood and meet their needs.