Developing a new methodology for assessing future food production based on machine learning, remote sensing and crop models

该博士学位结合了遥感，作物模型和机器学习，以改进对未来粮食生产的预测。从历史上看，研究只使用其中一种方法，最近在利兹和其他地方的进展集中在组合，如作物模型和遥感;或机器学习和作物模型。遥感农田数据通常用于绘制作物种植模式和区分作物类型。这些地图反过来又被acdemics和工业界用于一系列用途，从商品和供应链监测到预测气候变化的未来影响。全球农田地图包括（例如Monfreda等人，2008年）。汇集这些数据集所涉及的资源意味着这些数据集不会经常更新。同时，基础方法不断得到改进（Orynbaikyzy等人，2019年，还有很多事情要做。例如，区分一年生作物和多年生作物是联合利华确定的一个关键问题。2014年，Chelseor，2009年），首先模拟天气和气候对产量的影响。然而，只有在作物种植面积也已知或估计的情况下，才能对作物总产量进行任何形式的评估。了解这些未来的种植面积并不简单，因为它们将在一定程度上取决于气候变化导致的作物适宜性变化。作物适宜性模型可用于了解未来气候对特定作物的适宜程度（例如Rippke等人，2016）。然而，这些模型通常不与基于过程的作物模型一起使用。相反，对未来作物产量的评估要么假设作物面积等于一些历史数据集的面积，要么使用对未来土地使用的现有评估（例如LUH 2 [1]），其通常不考虑作物适宜性。2017年）现在是一个无处不在的技术和方法在许多领域，包括大数据，健康，环境，机器人等有一个巨大的需求，学术界和工业界的博士与机器学习的专业知识。ML已经在利兹作为潜在的下一代作物模型进行了测试[2]。在这个学生项目中，您将利用所有主管的专业知识和资源，以便：评估和使用遥感数据来识别作物的具体位置和生产强度。进行范围研究，以确定最有前途的作物，区域，数据集和模型，用于预测未来的作物位置和强度。使用和开发ML方法来改善农田数据集，通过利用该领域的持续进展（例如Orynbaikyzy等人，2019).探索作物产量模型、适宜性模型、遥感数据和机器学习的组合使用，以预测未来现实的农田地图[1] https：//luh.umd.edu/data.shtml [2] https://environment.leeds.ac.uk/see/pgr/2584/joe-gallear

This PhD combines remote sensing, crop models and machine learning in order to produce improved projections of future food production. Historically, studies have used only one of these methods, with recent progress at Leeds and elsewhere focusing on combinations, such as crop models and remote sensing; or machine learning and crop models.Remotely-sensed croplands data are routinely used to map crop cultivation patterns and distinguish crop types from each other. These maps are in turn used by acdemics and by industry for a range of uses, from commodity and supply chain monitoring to projecting the future impacts of climate change. Global croplands maps include (e.g. Monfreda et al., 2008). The resources involved in putting together such datasets mean that they are not frequently updated. Whilst, the underpinning methodologies are constantly being improved (Orynbaikyzy et al., 2019), much remains to be done. Distinguishing annual crops from perennials, for example, is one key issue identified by Unilever.Crop models are regularly used to develop options to adapt to climate change (e.g. Webber et al., 2014, Challinor, 2009), by simulating first the impacts of weather and climate on yields. However, it is only when the areas under crop cultivation are also known or estimated that any kind of assessment of total crop production can be made. Knowing these future cropped areas is not trivial, since they will be determined to some extent by changes in crop suitability driven by climate change. Models of crop suitability can be used to see how suitable future climates are for particular crops (e.g. Rippke et al., 2016). However, these models are not generally used alongside process-based crop models. Rather, assessments of future crop production either assume that cropped areas equal that of some historical dataset, or else they use an existing assessment of future land use (e.g. LUH2[1]), which generally do not account for crop suitability.Machine learning (ML; see Witten et al., 2017) is now a ubiquitous technique and methodology in many fields, including big data, health, environment, robotics etc. There is a huge demand by both academia and industry for PhDs with machine learning expertise. ML is already being tested at Leeds as potential next-generation crop models[2].In this studentship, you will draw on the expertise and resources of all supervisors in order to:Evaluate and use remote sensing data for identification of crop specific locations and production intensities.Conduct a scoping study to determine the most promising crops, regions, datasets and models to use in the PhD for predicting future crop locations and intensities.Use and develop ML methods to improve croplands datasets, by drawing on ongoing progress in this area (e.g. Orynbaikyzy et al., 2019).Explore the combined use of crop yield models, suitablitity models, remotely-sensed data and ML for projecting realistic future croplands maps[1] https://luh.umd.edu/data.shtml[2] https://environment.leeds.ac.uk/see/pgr/2584/joe-gallear