Auto-Calibration of Global Flood Forecasting Systems using Artificial Intelligence

洪水是最常见的自然灾害，也是全世界自然灾害死亡的主要原因。1980年至2009年间，洪水造成539，811人死亡（范围：510，941人至568，680人），361，974人受伤，2，821，895，005人受到洪水影响（资料来源：PLOS）。在全球范围内，洪水造成的经济损失从20世纪80年代的每年约70亿美元增加到2001- 2011年的每年240亿美元（经通货膨胀调整）。洪水预报是水文学中最具挑战性和最困难的问题之一。它也是水文学中最重要的问题之一，因为它在减少经济和人类损失方面做出了重要贡献。在世界许多地区，洪水预报是管理洪水的少数可行办法之一。近年来，由于气象和水文建模能力的整合、通过卫星观测改进数据收集以及分析和交流不确定性的知识和算法的进步，预测的可靠性有所提高。然而，在这方面，洪水预报技术的可扩展性仍然受到校准（或调整）所需时间和资源的限制水文模型中的参数，以便预测是准确的。该项目旨在通过利用人工智能和机器学习的最新进展以及沿着丰富的数据来解决这个可扩展性问题，以自动化校准过程。我们将使用概率规划方法这使得数据，专业的人类知识和机器学习能够协同工作，提供最佳，严格和可重复的自动校准。实现水文模型的自动机器学习校准将消除当前的瓶颈，实现更快，更准确的洪水预报。这将有助于对更广泛地区进行更多的洪水预报，从而减少经济损失，降低洪水给人民生活带来的风险和负面影响。

Floods are the most common natural disaster and the leading cause of natural disaster fatalities worldwide. There were 539,811 deaths (range: 510,941 to 568,680), 361,974 injuries and 2,821,895,005 people affected by floods between 1980 and 2009 (Source: PLOS). Globally, economic losses due to flooding increased from roughly US$7 billion per year in the 1980s to US$24 billion per year in 2001-11 (adjusted for inflation).Flood forecasting is one of the most challenging and difficult problems in hydrology. It is also one of the most important problems in hydrology due to its critical contribution in reducing economic and human damages. In many regions of the world, flood forecasting is one of the few feasible options to manage floods. Reliability of forecasts has increased in recent years due to the integration of meteorological and hydrological modelling capabilities, improvements in data collection through satellite observations, and advancements in knowledge and algorithms for analysis and communication of uncertainties. However, scalability of flood forecasting technologies is still limited by the time and resources it takes to calibrate (or tune) parameters in hydrological models such that the forecasts are accurate.This project aims to address this question of scalability by utilising the recent advancements in AI and Machine Learning along with the wealth of data now available to automate the process of calibration.We will use the probabilistic programming approach which enables data, expert human knowledge and machine learning to work together to provide optimal, rigorous and reproducible automatic calibrations.Achieving an automatic Machine Learning calibration of hydrological models would remove the current bottleneck, enabling faster and more accurate flood forecasts. This will permit more flood forecasts over wider areas, all of which leads to less economic damage and a reduction in the risk and negative impact that flooding can bring to peoples' lives.