基于2012-2019年自动站雷暴大风观测实况和对应雷达回波,利用传统机器学习方法(决策树)和深度学习方法(CNN、YOLO)等三种机器学习方法分别建立雷暴大风自动识别模型.根据广东雷暴大风回波特征,选取50 dBZ高度、反射率因子强度梯度等5个回波参量作为决策树的特征因子;将1～9 km高度的雷达回波分为11层,作为YOLOv3的输入层,使其由原3个特征层扩展到11层,训练优化后的YOLOv3可更合理刻画雷暴大风的空间结构特征.经批量测试和业务试运行试验,检验结果表明:三种模型中基于决策树的模型虚警最高,基于CNN的模型漏报最多,基于YOLO的模型识别效果最好,其POD和CSI均最高.通过对广东2020年汛期5次系统性和5次局地性雷暴大风过程进行分类型自动识别效果评估,并选取任意天气下长达30天连续时段进行不间断识别检验,结果表明该算法对于不同类型的雷暴大风均有较好的识别能力,具备业务化应用前景.

利用龙门县内南昆山、气象观测基地和沈村3 个不同的大气负氧离子监测站数据,对龙门县负氧离子时空特征进行分析.结果表明:龙门县负氧离子资源丰富(年均浓度为2 989 cm-3),其空间分布与植被覆盖度对应呈现为西高东低趋势.高植被覆盖度区域南昆山站的负氧离子浓度日变化不明显,而低植被覆盖度区域气象观测基地和沈村站存在夜间、早晨高,下午、傍晚低的特征.季节变化来看,龙门县负氧离子浓度呈现春夏高、秋冬低的分布趋势.

利用揭西县8个大气负氧离子自动监测站数据,对揭西县大气负氧离子浓度空间分布与时间变化特征进行分析.结果 表明:揭西县大气负氧离子资源十分丰富,负氧离子浓度呈现从西往东、从北往南逐渐增加的趋势,东部生态区负氧离子浓度普遍较高,年平均浓度基本超过1 000 cm-3,全县年平均为1 609 cm-3,揭西县坪上镇石内村最高,高达2 609.1 cm-3.在居住区,一天之中负氧离子浓度在清晨前后最高,并有夜间高、白天低的变化规律;在生态区,负氧离子浓度有白天高、夜间低的变化规律,且最高浓度出现在上午.揭西县京溪园镇粗坑村夏季正午负氧离子浓度最高,有明显的年变化波动,1月负氧离子浓度最高,达2 728 cm-3.大北山森林公园冬季午后负氧离子最盛,年变化波动较小,12月负氧离子浓度最高,8、9月负氧离子浓度处于较低水平.

广东省植被类型丰富、气候条件复杂,开展气候因子对植被的影响研究,对于全省保护生态环境、应对气候变化具有重要意义.利用MODIS NDVI数据和地面气象观测数据,基于变化趋势分析、空间自相关分析、相关性分析等方法,研究广东省NDVI时空分布特征及其对气温、降水、日照时数等气候因子的响应.结果表明,2000—2018年广东省NDVI平均值为0.62,总体呈上升趋势,平均年增长值为0.0053,正增长区面积占比达94.50％.全省73.72％的区域植被呈现显著空间集聚性,其中高值集聚区面积占比48.19％,低值集聚区面积占比25.53％.不同类型植被的NDVI均呈增加趋势,其中常绿阔叶林、常绿针叶林、混交林的NDVI平均值最大,而农田、矮树灌木、混交林平均年增长值最大.NDVI具有明显的季节变化规律,夏季NDVI值最大、年增长值最小,冬季NDVI值最小、年增长值最大.年内NDVI最低值出现在2月,最高值出现在9月.月平均NDVI与气温、降水、日照时数相关性显著,其最大相关系数分别为0.8637、0.6394、0.6057.NDVI对日照时数的响应不存在滞后,对气温存在1个月的滞后,对降水存在1—2个月的滞后.日照时数对NDVI的影响仅持续1个月左右,温度、降水对NDVI的影响可持续4—5个月.

利用MODIS地表温度数据,计算城市热岛强度指数,分析近15年广州市城市热岛的时空分布特征及演变规律,并结合气象观测数据、社会统计数据定性分析其主要影响因素.结果表明:广州市城市热岛的空间分布受地形地貌影响明显,负热岛区主要分布于森林密集的北部山区,无热岛区主要分布于中部低山丘陵区域,热岛区主要分布于高度城市化的中南部平原区.关于城市热岛的日变化规律,白天热岛区、负热岛区面积均小于夜间,但白天热岛区强度、负热岛区强度大于夜间.关于城市热岛的季节变化规律,冬季热岛区面积最大,热岛强度最小,夏季热岛区面积最小,热岛强度最大;冬季负热岛区面积最小,负热岛强度最小,夏季负热岛区面积最大,负热岛强度最大.对于城市热岛的年际变化规律,近15年来广州市的热岛区、负热岛区占全市总面积的百分比呈上升趋势,无热岛区所占百分比呈下降趋势,人为热排放在城市中心区域的持续增长,加上区内建筑物密度大、植被覆盖度低,导致了热岛区的增加,而北部山区至中部丘陵山区的植被的持续好转,加上地理特征限制了该区域的城市化发展,导致了负热岛区的增加.

利用遥感方法可以在区域尺度反演地表植被的光合生理状况和生产力变化,但亚热带常绿林冠层结构季节变化较小,传统的光谱植被指数对植被光合作用难以准确捕捉。利用2014—2015年中国科学院广东省鼎湖山森林生态试验站多角度自动光谱观测系统的光谱反射数据,分别反演传统冠层结构型植被指数（NDVI）、光合生理生化型植被指数（CCI）和叶绿素荧光型植被指数(NDFI 685 和NDFI 760 ),并利用不同类型植被指数的组合,构建多元线性回归模型。结果表明:亚热带常绿针阔混交林三种类型植被指数均与GPP的动态变化有显著的相关性,其中,NDVI是表征GPP较优的植被指数（R~2=0.60,P＜0.01）,其次为CCI（R~2=0.55,P＜0.01）,而NDFI能够作为辅助指数,有效提高NDVI（R~2=0.68,P＜0.001）和CCI（R~2=0.67,P＜0.001）表征GPP的程度。多个植被指数参与构建的多元回归模型能够有效提高亚热带地区常绿林GPP季节动态变化的拟合精度,提升遥感精确评估亚热带森林生产力的能力。

Light use efficiency (LUE) is a critical parameter for remote sensing estimation of ecosystem productivity.Therefore,accurate estimates of spatio-temporal variation of LUE is essentially required.Photochemical reflectance index (PRI) has been widely accepted to hold the potential to track the LUE variations.Here we combined spectral and eddy covariance observations in a subtropical mixed forest to investigate the ability of PRI to estimate LUE at a half-hourly time-scale in different seasons.The forest is located at Dinghu mountain forest ecological experiment station in Zhaoqing of Guangdong province and the observations ranged from April 2014 to March 2015.The results showed that:(1) LUE was significantly correlated with PRI but the correlation strength was different in different seasons;The correlation was the strongest in winter (R2=0.40,P＜0.01) and the weakest in summer (R2=0.04,P＜0.01).(2) The standardized PRI (△PRI),which removes noises from canopy and soil background,did not well capture the dynamics of LUE.(3) Photosynthetic active radiation (PAR) had the largest influence on the relationship between LUE and PRI among various climatic variables.Our analysis featured the temporal limitation of PRI in estimating LUE in subtropical forests and similar efforts are needed to quantify PRI ability to estimate LUE in other ecosystem types.

冬种辣椒是广东冬季农业生产的支柱产业.但冬种辣椒生产常常受寒潮、连阴雨、干旱等灾害性天气的威胁.调查广东省冬季气象灾害的发生规律,结合冬种辣椒的生长和栽培特点,探讨广东气象条件与冬种辣椒生产的关系,分析冬种辣椒面临的主要气象灾害的发生规律、特征及其对辣椒生产的影响.辣椒属喜光喜温作物,同时根系较浅,对水分要求严格.寒害是冬季辣椒生产面临的主要气象灾害;2、3月份的低温阴雨天气主要发生在粤北和珠三角地区,容易给辣椒造成渍害和湿害;冬季干旱发生频率也较大,珠江口北部以及雷州半岛地区是重点防范区域;而在全球变化的背景下,冬季暴雨对辣椒生产的影响亦不容忽视.因此,在农业生产中需要采取有效的防灾措施,提高广东省冬种辣椒生产的综合抗灾能力,并针对不同的天气状况提出相应的应对措施.

Canopy is a major structural layer for vegetation to carry out ecological activities. The differences of light radiative transfer processes in canopies caused by forest canopy structure directly influence remote sensing inversion of forest canopy biochemical composition. Thus an analysis of spectral characteristics between different canopy structures contributes to improve the accuracy of remote sensing inversion of forest canopy biochemical components. Based on a Hyperion hyperspectral image in the north Slope of Changbai Mountain Nature Reserve, through FLAASH (the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes) atmospheric correction, different canopy reflectance spectra were extracted, and spectral transforms were carried out using continuum removal method and first derivative method for quantitative analysis of the spectral characteristics. A set of spectral indices were calculated, including NIR (near infrared reflectance), NDVI (normalized difference vegetation index), EVI (Enhanced Vegetation Index), NDNI (normalized difference nitrogen index), SPRI (normalized photochemical reflectance index) * NDVI and SPRI * EVI (vegetation productivity index). Combined with the broad foliar dominance index (BFDI), the relationships between the spectral indices and canopy structure composition were investigated. The characteristics of canopy structure composition impacting its spectral curve and indices were clarified in the temperate forest. The results showed that: (1) there existed significantly different spectral characteristics between different canopy structures: comparing to the spectrum of broad-leaved forest canopies, the red edge moved to the left and their slope decreased, blue edge and yellow edge features were also weakened, near-infrared reflectance decreased, normalized reflectance in visible region risen for the spectrum of conifer forest canopies; (2) the spectrum variation were controlled by BFDL The correlations between BFDI and the spectral indices were significant (P < 0.01). It was suggested the ratio of broad-leaved and conifer in canopy played an important role in variation of spectral indices. The coefficients of determination (R2) of BFDI and NDVI, EVI, SPRI * EVI, SPRI * NDVI and NDNI were 0.90, 0.83, 0.83, 0.81, 0.68 and 0.59 respectively. It was revealed that BFDI could control the variation of the canopy structure, greenness, leaf nitrogen concentration, leaf area index and productivity in temperate coniferous and broad-leaved mixed forests. Our findings were very significant foundation for accurate determination of forest type, quantitative extraction of canopy biochemical components, estimation of regional forest ecosystem productivity and other related researches.

The photochemical reflectance index (PRI) calculated from spectral reflectance has universally become a proxy for the light-use efficiency (LUE), which significantly improves the LUE-based estimation of ecosystem gross primary productivity on a large scale through upscaling. In this study, we observed the vegetation spectral reflectance of a planted subtropical coniferous forest from the top of a flux tower at Qianyanzhou Station, one of the ChinaFLUX sites, in September and December 2013, and simultaneously measured CO2 flux and meteorological variables for correlation and regression analysis. Results showed that PRI had a better correlation with LUE (R2 = 0.20, P< 0.001) than that of normalized difference vegetation index (NDVI), i.e., PRI was preferred in LUE retrieval. During the whole observation period, PRI and soil water content (SWC)-based bivariate regression model correlated well with LUE (R2 = 0.29, P < 0.001 and R2 = 0.30, P < 0.01 for daytime and midday observation, respectively), but in autumn the bivariate regression model of PRI and vapor pressure deficit (VPD) had a higher correlation with LUE (R2 = 0.448, P < 0.001) for midday observation, which showed that environmental factors, i.e., SWC and VPD, had a potential in improving the LUE retrieval from PRI, but the choice of appropriate environmental factors depended on season.

Eddy covariance (EC) technique is the most direct way to measure the exchanges of carbon dioxide (CO2), water vapor,and energy flux between terrestrial ecosystems and atmosphere, which can be used to explore CO2 exchanges between terrestrial ecosystems and atmosphere and its controlling mechanism. In this paper, we used the multivariate geographic clustering approach to generate flux-ecoregions with different clustering number (25,50,75,85,100,150,200 clusters) in China based on 11 variables affecting carbon flux,including meteorological factor,soil factor,abiotic factor of topography, actual vegetation (Leaf area index (LAI) and Enhanced vegetation index (EVI)) and vegetation productivity variables (Gross primary productivity, GPP). Based on the spatial distribution pattern of the existing flux observation stations in China and the comparative analysis between newly generated flux ecoregions and the existing geographical regionalization, the results showed that the existing 85 eddy covariance flux observation stations in China cannot reflect the spatial and temporal characteristics of carbon flux of all ecosystems because of the countrys complex topography and the diverse ecosystem types. It is also recommended that the number of the flux-ecoregions be 100-150. Considering the building and operating costs of the flux towers, the number of eddy flux tower stations can be added to 150 sites. Thus,the optimized flux network is supposed to represent major ecosystems and facilitate the integration of flux and remote sensing data,consequently, improve the accuracy of upscaling CO2 and water vapor flux observations from tower to regional scales to better exam the simulation result of the process based ecosystem model.