|
|
Development of a Geospatial Soil-Crop Inference Engine for Smallholder Farmers (India)
Research Team
Principal Investigator:
Sabine Grunwald, Department of Soil and Water Science, University of Florida
Co-Principal Investigators:
Scot E. Smith, Geomatics/School of Forest Resources and Conservation
Suhas Wani, International Crops Research Institute for the Semi-Arid Tropics, Patancheru, India
K. Ramesh Reddy, Department of Soil and Water Science, University of Florida
Graduate students:
Christopher Clingensmith, Department of Soil and Water Science, University of Florida
Collaborators:
Yaduraju,
International Crops Research Institute for the Semi-Arid Tropics, Patancheru, India
V. Balaji, Commonwealth of Learning, Vancouver, Canada
Walter Bowen, IFAS International Office, University of Florida
Time: 1/2012 to 1/2014
Funding Source: National Science Foundation (NSF) - EAGER
Overview:
In this project we develop an inference engine to transform smallholder farm production systems for resource optimization by utilizing spectral technology and geospatial modeling. The intellectual merit focuses on fusing of innovative sensor technologies into a holistic engine aiming to enhance conservation of soil quality and optimize crop yield to support smallholder farmers. The overarching goal is to build a Geospatial Soil-Crop Inference Engine (GeoSCIE) that fuses ground-based and remotely-sensed spectral data with geo-referenced observations to derive critical metrics for crop and soil management. Several hypotheses are tested in two smallholder farm settings in Karnataka and Andhra Pradesh, India, to provide a proof-of-concept for a novel multi-spectral, spatially-explicit engine that links research models to pixels of smallholder farms.
The objectives of this project are to:
(1) Develop and validate quantitative models that relate analytically-derived measures of soil indicators including soil texture, soil organic matter, macronutrients (nitrogen and phosphorus), micronutrients (sulfur, boron, and zinc) and soil spectral data derived from diffuse reflectance spectroscopy (DRS) (visible/near-infrared and mid-infrared spectral ranges);
(2) Design and implement a three-tiered multi-spectral system using ground-based spectral reflectance measurements, aerial sensors, and satellite multispectral scanners for a sequence of cropping seasons that aims to identify the optimal footprint or instantaneous field of view (IFOV) required for accurate assessment of crop-specific properties and stressors in smallholder farm settings;
(3a) Fuse soil DRS and remote sensing data into a GeoSCIE and calibrate and validate the engine by estimating a suite of critical soil and crop/vegetation-specific properties;
(3b) Use a genetic algorithm to derive indices from fused spectral data to infer on soil quality, fertility, water deficiency, and crop stress; and
(4) Translate results from GeoSCIE into Reusable Learning Objects (RLOs) to provide training/learning material to smallholder farmers and streamline GeoSCIE-based management recommendations into the agricultural-oriented social network application and information system AGROPEDIA used by smallholder farmers in India.
The methods entail:
(1) Inferential modeling to derive multiple soil properties from DRS spectra using ensemble tree regression methods, such as boosted regression trees and Random Forest (calibration and validation);
(2) Use of the Shuffled Complex Evolution – Universal Algorithm (SCE-UA) to derive soil indices from DRS spectra and soil observations;
(3) Derive crop-specific properties from various remote (IRS satellites and ADS40 aerial photographs) and a ground-based system and identify the optimal IFOV for smallholder settings;
(4) Fuse spectral and field/lab observations combining geostatistical methods (e.g. regression kriging) and SCE-UA; and
(5) Convert research findings into a format easily understood by smallholder farmers using audio and video recordings and develop a text-based tool to disseminate GeoSCIE findings to smallholder farmers via AGROPEDIA.
The research is expected to have broad impact by integrating research with education and knowledge sharing activities. To foster advancing discovery and understanding while promoting teaching, training and learning we will utilize the EcoLearnIT RLO System. RLOs are globally accessible and will not only reach smallholder farmers in the selected study areas in India, but can be also used for training by farmers elsewhere. Undergraduate students will be explicitly engaged in the educational component of this project. The aim is to create broad impact by engagement and networking activities which include integration of research findings and RLOs into short courses taught by PI and Co-PIs at the UF-ICRISAT Education Center in India, which targets underrepresented groups in South and South-East Asia, and other developing countries in Africa. Results from GeoSCIE will be streamlined into AGROPEDIA which is ideally suited to broaden participation of smallholder farmers, since it has shown success to link farmers and researchers.
Results [in progress]
Peer-reviewed Publications:
Oral and poster presentations:
|
|
