SBIR Phase I: Creating high-quality, lower-cost soil maps using machine learning algorithms

SBIR第一阶段项目的更广泛影响/商业潜力将是为农学家和农民制作高质量、准确、高分辨率的土壤地图。准确的土壤信息是更好、更有效的作物/土壤管理的根本驱动力。这项新技术将显著提高农场的盈利能力，降低食品成本，并改善环境保护和可持续性。使更高质量的土壤肥力制图易于获得和使用是本项目的目标。这项技术预计将提高作物产量，同时降低投入成本，从而在一个长期遭受低利润率困扰的行业带来更高的利润。预期的好处包括对环境更负责任的农场管理和更好的粪便管理规划、养分管理规划、精准农业、土地利用规划、种植决策、对植物的压力源评估、田间调节、作物轮作和产量预测/解释。这将增加农场的盈利能力，更有效地使用氮肥，并提高土壤的健康和植物的肥力。该项目推进了一项创新技术，该技术包括三个关键组成部分，用于绘制训练场内外的基本土壤养分地图--这些地图目前需要进行广泛的采样，但数据不够充分。首先是数字山坡位置，以选择最佳采样位置来表示整个景观中的土壤变异性，从而消除了采集不必要的土壤样本的需要。第二个元素利用先进的机器学习算法，对样本量不敏感。第三个要素是其选择适当的遥感信息(地形变化和卫星图像)的能力。该技术将选择适当的地形导数分析尺度，以捕捉所有潜在的土壤变异性。然后，它将根据空间、时间和光谱分辨率选择和使用适当的卫星图像波段，以减少过度拟合和计算时间的风险。与目前可用的方法不同，这项技术可以利用从训练场获得的土壤信息预测训练场内外的土壤养分--也就是说，这项技术具有预测邻近田地土壤性质的潜力--而不需要额外的样本。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/ commercial potential of this SBIR Phase I project will be to produce high-quality, accurate, high-resolution soil maps for agronomists and farmers. Accurate soil information is a fundamental driver of better, more efficient crop/soil management. The new technology would significantly increase farm profitability, lower food costs, and improve environmental protection and sustainability. Making higher quality soil fertility mapping readily available and usable is the goal of this project. This technology is expected to result in increased crop yield while allowing for decreased input costs, leading to higher profits in an industry that chronically suffers from low profit margins. The expected benefits include more environmentally responsible farm management and better manure-management planning, nutrient-management planning, precision farming, land use planning, planting decisions, evaluating stressors on plants, field conditioning, crop rotation, and prediction/interpretation of yields. These will result in increased farm profitability, more efficient application of nitrogen fertilizers, and increased soil health and fertility for plants. This project advances an innovative technology has three key components to produce maps of essential soil nutrients in training fields and beyond — maps that currently require extensive sampling while producing inadequate data. The first is a digital hill-slope position to select optimal sampling locations to represent the soil variability across the landscape, eliminating the need to take unnecessary soil samples. The second element leverages advanced machine-learning algorithms insensitive to the quantity of sample size. The third element is its ability to select suitable remotely sensed information (terrain derivatives and satellite imagery). The technology will select appropriate analysis scales of terrain derivatives to capture all potential soil variability. It will then select and use proper bands of satellite imagery, based on spatial, temporal, and spectral resolution, to decrease the risks of overfitting and computation time. Unlike currently available methods, this technology can predict the soil nutrients inside the training fields and beyond — i.e., this technology has the potential to predict soil properties in neighboring fields — using the soil information obtained from training fields—without the need for additional samples.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.