两种机器学习方法在重庆夏季旱涝预测中的应用

利用1961—2010年重庆34个气象观测站夏季降水资料及国家气候中心130项环流指数,采用机器学习的决策树和随机森林方法建立重庆夏季旱涝预测模型,通过2011—2018年预测效果检验发现,夏季同期环流指数决策树模型和前冬海温指数决策树模型预测的8 a降水异常趋势均正确,比考虑单一指数的PC评分分别提高37.5%和12.5%。此外,用随机森林模型预测重庆2014—2018年的夏季降水,5 a平均PS、CC和PC评分分别是84.6、0.27和67.1,相比于业务发布预报质量均有明显提高,且随机森林的预测质量较为稳定。     

Monthly prediction of air temperature in Australia and New Zealand with machine learning algorithms

Long-term air temperature prediction is of major importance in a large number of applications, including climate-related studies, energy, agricultural, or medical. This paper examines the performance of two Machine Learning algorithms (Support Vector Regression (SVR) and Multi-layer Perceptron (MLP)) in a problem of monthly mean air temperature prediction, from the previous measured values in observational stations of Australia and New Zealand, and climate indices of importance in the region. The performance of the two considered algorithms is discussed in the paper and compared to alternative approaches. The results indicate that the SVR algorithm is able to obtain the best prediction performance among all the algorithms compared in the paper. Moreover, the results obtained have shown that the mean absolute error made by the two algorithms considered is significantly larger for the last 20 years than in the previous decades, in what can be interpreted as a change in the relationship among the prediction variables involved in the training of the algorithms.     

Spatial and temporal analysis of urban flood risk assessment

广东暴雨强度大、范围广、季节长,造成的灾害重、影响大。为合理、定量地评估广东暴雨洪涝过程强度及其损失,基于1994-2018年广东致灾暴雨过程和相应灾情资料,构建了广东暴雨过程综合强度评估模型和灾情指数模型,并采用百分位数法进行暴雨强度和灾情等级划分,以第60、第80、第90和第95百分位数为临界阈值,分别将致灾暴雨过程强度和灾情划分为弱（1级）、较弱（2级）、中等（3级）、较强（4级）、强（5级）和微灾、小灾、中灾、大灾、巨灾5个等级,进而分析了不同强度等级暴雨过程可能造成的人口、农作物、房屋和经济等承灾体损失。结果表明：（1）1994-2018年间,广东各等级致灾暴雨过程主要出现在4-9月的汛期,5-7月尤其多,要特别注意防御;（2）致灾暴雨过程强度等级与各类承灾体灾情指数存在显著正相关关系：随暴雨强度的增强,倒塌房屋数呈指数增长,受灾人口、死亡人数、农作物受灾面积和直接经济损失呈线性增长;（3）平均而言,当暴雨强度达到强（5级）等级时,受灾人口、死亡人数、农作物受灾面积、倒塌房屋数和直接经济损失标准分别约为187.19万人、22人、10.52×10⁴ hm²、1.12万间和13.07亿元。

     

基于遥感与GIS的浙江省洪涝灾害综合风险评估研究

暴雨洪涝是最为频发、多发的自然灾害之一,已成为我国实现可持续发展的严重障碍。开展暴雨洪涝灾害风险评估技术与应用研究,是当今防灾减灾中的一项迫切要求。本文基于应用场景时间顺序从灾前预评估、灾中跟踪评估、灾后评估三个方面系统回顾了前人的研究历史和当前研究所取得的成果,并在对研究现状进一步认识的基础上,提出了当今暴雨洪涝灾害风险评估研究中存在的主要问题,指出了有待进一步研究和发展的新方向。

     

1981-2020年贵阳市洪涝灾害多尺度时序特征分析

全球城市化进程正在不断快速推进,城市洪涝灾害灾害频繁发生,成为世界各国、特别是中国等发展中国家普遍面临的城市问题,分析洪涝现象的时间演变规律和发展趋势,对于采取科学的政策措施有效防治洪涝灾害具有重要的参考价值。基于贵阳市1961-2020年的逐日气象数据,利用SPEI指数、Mann-Kendall检验方法和小波变换,分析了近60年多尺度时序下的洪涝变化规律。结果表明: ①贵阳市月度和季节降水强度大,暴雨发生频次高,月度降水强度集中在大雨及以上,季节性降水强度集中在暴雨及以上,易发生极端降水事件,造成城市洪涝灾害;②多时间尺度下,SPEI值对于降水、气温等气象因素的响应敏感程度不一,日尺度敏感程度最强,其他尺度趋于平缓,同时,大尺度对应的洪涝事件占比逐渐减少。③贵阳市发生洪涝的年份占总年份的21.67%,SPEI-12呈上升趋势, SPEI-3突变发生在19世纪60年代,突变后呈上升趋势且秋冬两季上升趋势显著,说明季节性洪涝有增强趋势; ④SPEI-1的周期变化特征为：在33个月主周期条件下的周期是23个月;在14个月主周期条件下的周期是9个月。     

Estimation of aerosol acidity at a suburban site of Nanjing using machine learning method

Journal of Atmospheric Chemistry - Aerosol acidity is found to exert negative effects on ecosystem diversity and architectural appearance. Current analytical technology is unable to measure in-situ...     

Modeling soil respiration and variations in source components using a multi-factor global climate change experiment

Soil respiration is an important component of the global carbon cycle and is highly responsive to changes in soil temperature and moisture. Accurate prediction of soil respiration and its changes under future climatic conditions requires a clear understanding of the processes involved. Most current empirical soil respiration models incorporate just few of the underlying mechanisms that may influence its response. In this study, a new partially process-based component model that separately treated several source components of soil respiration was tested with data from a climate change experiment that manipulated atmospheric [CO₂], air temperature and soil moisture. Results from this model were compared to results from other widely used models with the parameters fitted using experimental data. Using the component model, we were able to estimate the relative proportions of heterotrophic and autotrophic respiration in total soil respiration for each of the different treatments. The value of the Q ₁₀ parameters for temperature response component of all of the models showed sensitivity to soil moisture. Estimated Q ₁₀ parameters were higher for wet treatments and lower for dry treatments compared to the values estimated using either the data from all treatments or from only the control treatments. Our results suggest that process-based models provide a better understanding of soil respiration dynamics under changing environmental conditions, but the extent and contribution of different source components need to be included in mechanistic and process-based soil respiration models at corresponding scales.     

基于机器学习方法的辽宁省初霜冻日期预测模型研究

基于前期ERA5逐月再分析数据,应用3种机器学习算法(Lasso回归、随机森林和神经网络)对辽宁省初霜冻日期进行预测评估。Lasso回归算法提取对初霜冻日期预测有重要指示意义的气象要素特征集,通过交叉验证和超参数调优建立初霜冻日期预测模型,利用均方根误差(RMSE)和距平同号率方法定量定性地评估模型的预测效果。结果表明：特征选择后的气象要素特征集建模提升了模型的泛化能力、可解释性和稳定性;Lasso回归模型在4月起报的预测效果最好(RMSE为6—8 d),神经网络模型在5月起报性能最好(RMSE为6—9 d),随机森林模型在3月起报性能最好(RMSE为8—9 d);辽宁全省大部分站点距平同号率为50%—70%,其中Lasso回归和神经网络模型为5月起报最高(约为68%),随机森林算法为3月起报最高(约为62%)。特征选择和敏感性实验结果发现,低植被覆盖比例是初霜冻日期预测关键预测因子,植被覆盖率越高越有利于地表含水量保持,降温容易产生霜冻,初霜冻日期也就越易提前,去掉低植被覆盖比例因子后模型预测效果显著下降,也表明该因子是模型建模的前期关键因子。     

基于多种机器学习模型的西北地区蒸散发模拟与趋势分析

基于机器学习方法和多源数据构建高精度蒸散发（Evapotranspiration,ET）产品对研究气候变化背景下干旱、半干旱地区陆地水循环变化具有重要意义。本文利用西北地区12个草地通量站点与卫星遥感产品,基于随机森林、极端梯度提升、支持向量回归和人工神经网络4种机器学习方法构建ET估算模型,制作5 km分辨率ET产品,并分析ET的长期变化趋势。交叉验证结果表明,4种模型的均方根误差都低于0.57 mm·d^-1,R²高达0.73～0.88。SHAP （SHapley Additive exPlanation）可解释性分析表明,4种模型均将净辐射、植被和土壤湿度作为ET估算的重要因子,也能刻画出土壤偏干时土壤水分对ET的限制作用,有较好的物理解释性。多模型集合的ET结果相比单一机器学习模型以及现有遥感产品误差分别降低7％～20％和45％～70％。趋势分析结果显示,西北地区非裸地下垫面在2001—2018年间整体呈现ET增加趋势,平均速率为19 mm/（10 a）。在河套平原和内蒙古中部和东北部地区,ET的增长速率超过降水,这可能会进一步加剧这些地区的干旱化。     

Urban vulnerability and adaptation to heatwaves: a case study of Graz (Austria)

The increasing frequency of heatwaves, particularly in urban contexts, is one of the perceptible consequences of climate change. A city’s vulnerability to these heatwaves must be determined to develop proper adaptation measures. This article addresses the vulnerability of a medium-sized city in Central Europe, Graz, to heatwaves. Based on secondary data and primary data gathered from expert interviews, we identified certain determinants of vulnerability for the city: temperature, proportion of open and green spaces to developed areas, construction period of buildings, distribution of age and poverty risk, adaptation strategies used, and risk perception levels assessed for decision makers in the city administration. Certain city districts can be classified as particularly vulnerable. A high level of risk perception was detected among all decision makers and some adaptation measures have already been enacted. In particular, inter-organizational collaboration in adaptation networks works effectively. A deficit in efficient communication between researchers, policy makers, and members of the public was perceived to be the main barrier. This case study exemplifies the assessment of a city’s vulnerability to heatwaves on the basis of particular determinants and can be applied to many other cities.

Policy relevance

The method applied revealed potential improvements and opportunities on the policy level. Strong networks for climate change adaptation are most effective if regular meetings take place, allowing trust and friendship to grow between decision makers. More target-group-oriented information is needed. Emergency organizations, in particular, need more information, because the perception of heatwave risks has only been based thus far on personal experiences. By establishing a central authority, more information could be provided on heatwaves in cities. The need to raise the perception of members of the population and motivate them to take personal responsibility during disasters was emphasized by interviewed decision makers. This can be supported by providing advice during heatwaves through newspapers, TV, and radio. People in risk groups and their relatives could be trained in workshops. City areas that are at high risk should be marked on maps to make relevant information more tangible for decision makers.     

改进的AHP在县域尺度暴雨洪涝风险评价的应用

以淮河流域为例,选取降水、土地利用、经济、人口等指标作为淮河流域暴雨洪涝灾害风险指标,利用信息熵改进的层次分析法确定淮河流域暴雨洪涝的风险评估指标权重,并应用于县域尺度淮河流域暴雨洪涝灾害风险评价。结果表明:(1)淮河流域暴雨洪涝灾害风险空间分布整体呈现南部高、北部低,东西高、中部次之的形态。(2)改进的层次分析法得到的高风险区比传统方法的面积减少,市县个数下降,而次高风险区、中风险区、次低以及低风险区面积比之传统方法均有增加。同时风险平均值升高,导致受灾程度可能加大。(3)改进方法得到的岳西县风险等级由高风险区降为次高风险区,低于金寨县风险等级。宿州市风险等级由次高风险区降为中风险区,较灵璧、泗县风险低,与实际情况更为相符,提高了淮河流域暴雨洪涝灾害风险评价精度。     

利用最大降水量拟合福州城市内涝积水深度的误差分析

利用2015—2016年福州城区内涝和地面气象观测资料,分析内涝成因、最大降雨量与积水深度的关系,建立城市内涝积水深度拟合方程,并对拟合效果进行了检验。结果表明:1)导致福州城市内涝的突发性原因占26%,是城市管网等数据更新困难的主因,非突发性原因占74%;2)在3次不同天气系统的降雨量拟合结果中,雨季降水过程拟合平均绝对误差最小,午后雷雨天气降雨过程拟合平均绝对误差最大;3)对内涝积水深度负拟合值进行归零处理和阈值限定后,其结果均与实况相符;4)基于最新数据资料所得拟合值的误差更小。     

基于机器学习的复杂地形下短期数值天气预报误差分析与订正

初步研发了一套基于机器学习方法XGBoost且考虑地形特征影响的数值预报多模式集成技术,并与传统的等权重平均和线性回归方法的集成效果进行了对比分析。利用北京地区快速更新循环数值预报系统每天8次循环预报给出的近地面2m温度、2 m相对湿度、10 m风速、10 m风向数据产品,分别基于机器学习方法XGBoost、等权重平均方法、线性回归方法构建了3种体现地形因子影响的多模式预报时间滞后集成模型。试验对比分析了暖季、冷季每日不同时刻的模式预报集成订正效果。结果表明:分季节试验中,基于XGBoost模型对2m温度、10m风速的集成预报结果相对原始最优预报结果误差明显优于其他两种传统方法。XGBoost对2 m温度集成的误差可降低11.02%—18.09%,10 m风速集成误差可降低31.23%33.22%,10 m风向集成误差可降低4.1%—8.23%。2 m相对湿度的集成预报误差与传统方法接近。基于XGBoost的多模式集成预报模型可以充分"挖掘"不同模式或不同时刻快速更新循环预报优点,有效降低模式的系统性误差,提供准确性更高的多模式集成确定性预报产品。     

金城江暴雨洪涝统计特征分析及防御措施

统计分析金城江暴雨洪涝分布特征,提出防御措施。     

Social learning towards a more adaptive paradigm? Reducing flood risk in Kristianstad municipality,Sweden

Social learning is often treated as an intervention, a designed process facilitated or even initiated by a third party. We investigated how a social learning process emerged spontaneously from inside Kristianstad, one of the most flood-prone municipalities in Sweden. Twenty key persons were interviewed over 8 years, many of them several times, to assess the process. A small action oriented group of technical professionals perceived the flood risk and were key drivers providing strategic innovative capacity. We identified the process attributes that fostered the learning, the knowledge generated and other learning outcomes adapting a model by Schusler et al. (2003). Despite some elements of double loop learning, this process was not able to change the prevailing stationary principle/paradigm, feeling safe behind the embankments and continuing building on low lying land. We argue that building resilience and adaptive capacity would require a mind shift to a paradigm of flood proofing/living with floods and preparing for the unexpected, acknowledging that water cannot be controlled at a certain level. We conclude that knowledge development is inhibited by the Swedish decentralisation approach and we call for a multilevel learning strategy including learning from international experience and emphasising more active coordination at the national level.     

Temperature projection in a tropical city using remote sensing and dynamic modeling

Recent temperature projections for urban areas have only been able to reflect the expected change due to greenhouse-induced warming, with little attempt to predict urbanisation effects. This research examines temperature changes due to both global warming and urbanisation independently and applies them differentially to urban and rural areas over a sub-tropical city, Hong Kong. The effect of global warming on temperature is estimated by regressing IPCC data from eight Global Climate Models against the background temperature recorded at a rural climate station. Results suggest a mean background temperature increase of 0.67 °C by 2039. To model temperature changes for different degrees of urbanization, long-term temperature records along with a measureable urbanisation parameter, plot ratio surrounding different automatic weather stations (AWS) were used. Models representing daytime and nighttime respectively were developed, and a logarithmic relationship between the rate of temperature change and plot ratio (degree of urbanisation) is observed. Baseline air temperature patterns over Hong Kong for 2009 were derived from two ASTER thermal satellite images, for summer daytime and nighttime respectively. Dynamic raster modeling was employed to project temperatures to 2039 in 10-year intervals on a per-pixel basis according to the degree of urbanization predicted. Daytime and nighttime temperatures in the highly urbanized areas are expected to rise by ca. 2 °C by 2039. Validation by projecting observed temperature trends at AWS, gave low average RMS errors of 0.19 °C for daytime and 0.14 °C for nighttime, and suggests the reliability of the method.     

宿迁地区水稻暴雨洪涝灾害保险气象理赔指数开发

利用宿迁地区104个乡镇2003—2013年水稻产量资料和2008—2013年保险理赔资料,计算水稻遭受暴雨洪涝灾害的灾害损失率;在自定义暴雨洪涝灾害指数的基础上,分析水稻灾损率与暴雨洪涝灾害指数的相关关系,建立各水稻种植区的气象理赔指数面板随机效应模型。结果得出:宿迁地区水稻遭受暴雨洪涝灾害影响的平均灾损率为-1.83%,达到理赔标准的乡镇比率为37.5%;受暴雨洪涝灾害影响最重的是III区,其次是I区,VII区和II区位居第三,其他几个区相对较轻;暴雨洪涝灾害主要影响水稻的孕穗扬花期和灌浆成熟期,对孕穗扬花期以前的生育期影响较小;暴雨洪涝灾害指数对水稻遭受暴雨洪涝灾害影响有较好的指示意义,利用这一指数制作水稻在灾损率预报准确率较高,达到90%以上;利用气象理赔指数,能较好的预报和模拟宿迁地区水稻遭受暴雨洪涝灾害的灾损率分布,对降低理赔风险,减少理赔费用有较强的指导意义。     

Assessing changing vulnerability to dengue in northeastern Brazil using a water-associated disease index approach

Water management practices and access to safe water supplies have major implications for human health. While a range of assessments has been developed to assess water vulnerability, limited work has extended these concepts to health and wellbeing. Water-associated disease cycles are characterized by complex linkages between social and ecological determinants, thus conceptualizing vulnerability in the context of health offers a useful framework for analysis. This paper applies a water associated disease index (WADI) as a tool to deepen understanding of changing vulnerability to dengue, comparing conditions in 2000 and 2010 in Pernambuco, Northeast Brazil. Multi-dimensional data were integrated into indicators of exposure and susceptibility using the WADI approach, including water access, land cover, climate, and solid waste collection, and outputs were validated and visualized in map form. The findings illustrate heterogeneous patterns of vulnerability to dengue in the region, and highlight trends of seasonal and long-term changes. Highest vulnerability was observed in densely populated Recife and the surrounding coastal region in both time periods, with climate conditions creating seasonal trends in exposure to dengue. While more remote areas in the semi-arid Sertão showed low vulnerability overall, increases were observed in some areas between 2000 and 2010 due to land use intensification and growing population densities. These findings suggest that interventions should consider the dynamic nature of social and ecological factors that contribute to health outcomes and address current as well as future populations vulnerable to dengue transmission. This vulnerability mapping approach can be applied to other water-associated diseases impacted by global environmental change to highlight priority areas for further investigation and contribute towards improving interventions.