油菜冻害卫星遥感监测与评估方法研究

发布时间：2018-01-05 04:32

本文关键词：油菜冻害卫星遥感监测与评估方法研究　出处：《浙江大学》2017年博士论文　论文类型：学位论文

【摘要】：油菜在我国位居油料作物之首,我国油菜的播种面积和产量均居世界第二位,其中约90%为冬油菜。越冬期是冬油菜必然经历的物候阶段,因而具有发生冻害的风险。传统意义上的油菜冻害监测与评估都是由工作人员到田间实地调查得到冻害指数,不仅耗时、成本高,且代表性差。遥感是及时获取大范围地表信息的有效有段,在作物灾害监测与评估领域具有明显的优势。本研究选取传统的冬油菜主产区——安徽省作为研究区域,基于中等分辨率成像光谱仪MODIS/MERIS、国产环境减灾卫星HJ-CCD/IRS等多种遥感数据源,以明确记载有油菜冻害发生的典型年份作为研究案例,探索针对油菜冻害的多角度评估方法,研究时相包括2003-2004、2009-2010和2010-2011三个生长季。主要的研究内容以及取得的主要成果包括以下几个方面:(1)基于日最低气温指标的油菜冻害发生区域判断本文从油菜冻害的致灾因子——日最低气温出发,综合利用MODIS遥感数据、地面气象观测资料和基础地理数据,采用星地多源数据相结合的解决思路,依据各等级油菜冻害的致灾温度国家标准,对于可能会有冻害发生的区域进行判断。以2004年1月下旬及2月上旬发生的油菜冻害为例,论文分晴空与云覆盖两种情况讨论了对研究区域全覆盖的日最低气温分布数据的获取方法。对于晴空条件,本文通过建立日最低气温与晴空下LST、NDVI及儒略日(DOY)之间的多元线性回归方程来估算其空间分布,以2000-2014年间冬季三个月每日获取的数据为输入用于建模及验证。结果显示对于日最低气温的估算效果,基于前一天夜间所获取的Terra-LST建立的估算方程具有最佳的表现,其次为当日白天获取的Terra-LST、白天Aqua-LST和夜间Aqua-LST。对于阴天条件,本文采用旬低温(旬内平均日最低气温)背景网格推算+冻害年份低温距平调整+残差空间化方法逐步拟合云下的最低气温分布场,该方法首先基于经度、纬度、海拔、坡度和坡向等地理要素数据来推算安徽全省冬季各旬(12月上旬至次年3月上旬)常年低温水平的空间分布。在HANTS滤波的支持下基于8天合成的LST产品(MOD11A2)来模拟冻害发生年份指定旬与常年(2000-2014年)同时期低温程度的距平,而后利用安徽省78个县级气象站的气温观测数据将各低温日期的最低气温与灾害年份旬低温水平之间的残差进行空间化。本文以地理要素数据为输入分别考查了多元线性回归以及随机森林非线性回归方法对低温残差的估算效果,结果显示随机森林模型对残差的估算精度明显要优于多元线性回归方法,故采用该方法将各站点处的低温残差扩展到整个研究区域。以晴空下的最低气温估算结果为主体,利用云下的推算结果作为补充,从而得到每个低温日期上对研究区域全覆盖的1 km分辨率的最低气温分布。由于国家标准中尚未对低温的持续时间进行定义,本文采用冷积温指标,即低温时段内每日最低气温的累加值来判断可能会有油菜冻害发生的区域。结果显示冷积温能够捕捉到出现在皖北地区(主要是宿州市一带)以及江淮地区(主要为合肥及滁州地区)的低温分布状况,且通过与各气象站的观测值进行比较(相关系数r=0.810,P0.01,RMSE=8.6℃),冷积温的估算取得了较好的效果。(2)油菜的遥感识别与提取油菜从始花期到盛花期绿度出现下降,在植被指数时间序列剖面上呈现为"谷"的特征,区别于同时期生长的其他冬季作物(主要为冬小麦),基于这两个物候时期油菜所特有的绿度变化趋势可以实现其种植区域的提取。考虑到不同地区生长的油菜物候期并不同步,本文基于8天合成的MODIS-NDVI时间序列数据,在S-G滤波平滑的基础上,通过识别不同地区始花期和盛花期的开始时间从而分区域提取出安徽全省油菜的分布,并且依据统计年鉴所记录的市级油菜种植面积数据对提取结果进行了检验。当冻害较为严重时,受冻后的油菜在花期时植株趋于低矮、花朵稀疏,在30 m像元分辨率尺度上与冬小麦容易混淆,极易造成对实际种植面积的低估,此外农业灾害评估又具有现势性要求。结合合肥当地油菜生产的实际情况,针对该问题本文提出了基准种植区域+越冬作物掩膜调整方法,采用相邻正常生长季(未有冻害发生)的油菜分布作为基准,基于越冬作物从出苗到越冬前生长的特性建立掩膜来调整基准种植区域,将得到的结果用来表示冻害发生年份越冬时期的油菜分布,并且依据油菜播种面积统计数据对其进行检验。(3)油菜冻害灾情的遥感评估对于2004年发生的油菜冻害,在MODIS像元尺度上借助于时间序列剖面分析了 4种广泛采用的植被指数对于冻害的敏感性。结果显示,NDVI和GNDVI的表现要优于EVI和SAVI这两个植被指数,相比之下NDVI对于冻害影响的表现更为充分。MODIS数据具有较长的时间序列连续性,适合采用植被指数距平法评估冻害灾情。选取相邻几个正常生长季(2001-2006不含2004年)同时期的平均长势水平作为基准,采用NDVI年际标准差来表示长势的正常年际波动,将冻害影响后的长势水平相对于基准水平的百分比差异用来描述冻害的灾情程度。由于缺乏田间实测的灾情数据,本文选取全省47个油菜主产县区的平均冻害程度与相应的冷积温进行相关性分析,结果表明二者之间具有显著的相关性(r= 0.378,P0.01),从而一定程度上支持了灾情的评估结果。MODIS空间分辨率较低,适合开展大范围油菜冻害的筛查,但对于市(县)区域的油菜冻害遥感监测与评估,需要采用更高空间分辨率的数据进行研究。综合考虑现有遥感资料的空间和时间分辨率,对于2010年和2011年初发生的油菜冻害,本文尝试采用国产环境减灾卫星数据进行冻害评估。选取由HJ-CCD的4个工作波段构建的8个常用的植被指数作为候选的冻害评估指标。利用2009-2010生长季灾后相对于灾前时相各植被指数的变化量较正常生长季(2008-2009)同时期变化量的归一化差异直方图曲线来判断不同植被指数对于冻害的敏感性。结果表明在30m空间分辨率尺度上,NDVI和GNDVI较其他植被指数同样具有更高的敏感性,相比较而言GNDVI的敏感程度更高,因而适合作为冻害的评估指标。基于灾后时相相对于基准水平的GNDVI百分比差异来判断冻害发生年份的灾情等级,对于2011年1月份发生的油菜冻害,本文选取合肥市9个苗情监测点实测的田间冻害指数对结果进行了验证,分析表明遥感评估得到的冻害程度与田间实测冻害指数之间具有较高的相关度(r =-0.698,P0.05)。采用灰色关联分析方法和统计分析方法,本文考查了一些自然因素和作物自身条件对冻害灾情程度的影响,结果显示南坡向和西坡向种植的油菜灾情相对更为严重,几个因素对灾情影响程度的大小由高到低依次为:灾前长势、土壤湿度、最冷日期的地表温度和海拔高度。(4)新的冻害敏感指数的构建本文基于中等分辨率成像光谱仪MODIS和MERIS波段所构建的几个生理变量敏感植被指数,在1 km级分辨率尺度上考察了这些指数对于低温胁迫的响应模式及其冻害敏感性差异。所选取的指数包括MODIS-PRI(表征光合速率)、NDWI(表征冠层湿度),以及MERIS-MTCI(表征冠层叶绿素含量)、REP(红边位置)与LAI(叶面积指数)。结果显示PRI对于低温胁迫响应非常强烈,间接表明光合作用对于冻害敏感;油菜受冻之后NDWI呈现为异常上升的趋势,据此推测冻害导致了冠层失水;LAI对于冻害响应迟钝,其数值不仅降幅较小且低温过后迅速恢复;低温过程并未引起MERIS-MTCI与REP的降低或者削弱,推测冻害可能并未引起叶绿素水平的降低。低温显著地抑制了 MTCI数值水平的上升,而REP几乎未受到低温的影响,引入红边波段构建的红边NDVI对于低温胁迫的敏感程度要低于标准NDVI。基于上述发现,本文选取MODIS-L1B级数据中对植被光合作用以及冠层湿度较为敏感的7个波段,在不同冻害等级的随机分层抽样样本的支持下,通过最佳指数(theoptimal index factor,OIF)分析方法筛选出其第5、11和12波段组合构建并且提出了新的冻害敏感植被指数MFISI(MODIS Freezing Injury Sensitive Index)。该指数能够同时响应植被的光合作用与冠层湿度参数,具有明确的生理学意义;且由于两种生理指标在遥感水平上对于冻害均敏感,新指数能够明确地展现冻害的影响,其数值的空间分布图具有丰富的地物细节信息;新指数在冻害灾情的预判以及冻害发展趋势的预测方面,其表现要优于广泛应用的NDVI。
[Abstract]:In our country in the rapeseed oil crop in the first, China's rapeseed acreage and output ranked second in the world, of which about 90% of winter rapeseed. Winter is phenological stages of winter rape inevitable, thus has the risk of freezing injury. The traditional sense of the damage monitoring and assessment of rape are from the staff the field investigation by freezing index, not only time-consuming, high cost and poor representation. Remote sensing is a effective and timely access to a wide range of surface information, has obvious advantages in the field of monitoring and assessment of crop disasters. This study selected the traditional winter rapeseed in Anhui Province as the study area, the medium resolution imaging spectrometer based on MODIS/MERIS, HJ-CCD/IRS and other domestic environmental mitigation satellite remote sensing data source, with clear records of typical years rape frost damage occurs as a case study, to explore for rape and freeze Multi angle evaluation method of phase including 2003-20042009-2010 and 2010-2011 three in growing season. The main research contents and main results obtained include the following aspects: (1) to determine the area of injury occurred rape daily minimum temperature index from the rape frost disaster causing factors -- Based on the minimum temperature, the comprehensive utilization of MODIS remote sensing data and the ground meteorological observation data and the basic geographic data, using the solution with multi-source data satellite phase, according to the level of damage caused by the national standard of rape disaster for temperature, there may be frost damage area to damage judgment. Rape took place in late January 2004 and early February as an example, the clear sky and cloud cover two a discussion on the methods of obtaining the study area full coverage of the daily minimum temperature distribution data. For the clear condition, based on the minimum Warm and clear sky under LST, NDVI and Julian day (DOY) between the multiple linear regression equation to estimate the spatial distribution of the 2000-2014 during the winters of three months daily data as input for modeling and verification. The results show that for the estimation of the effect of daily minimum temperature, the equation for estimating the day before the establishment of Terra-LST night based on the best performance, followed by the date obtained during the day Terra-LST, day Aqua-LST and night Aqua-LST. for the cloudy conditions, at low temperature (within ten days average daily minimum temperature) + low temperature freezing background grid calculation year anomaly adjustment + residual space method step by step fitting under the cloud minimum temperature distribution field, firstly based on the longitude, latitude, altitude, slope and aspect of geographic data elements to calculate each year in Anhui province in winter (from early December to early March next year) the spatial distribution of perennial cold water flat. 8 days of synthesis of LST products based on HANTS filtering support (MOD11A2) to simulate the damage occurrence in late specified and perennial (2000-2014 years) during the same period of low temperature anomaly, and then use the 78 Anhui county meteorological station temperature observation data of the residual between the date of the minimum temperature and the low temperature disaster year ten low level of space. According to the geographic elements as input data are examined to estimate the effect of multiple linear regression and nonlinear regression method of random forest to low temperature residuals, results show that the estimation accuracy of the residual random forest model was obviously superior to the multiple linear regression method, so the use of the method of low temperature residual each site is extended to the entire study area. The minimum temperature estimates under the sky as the main body, the calculation results under the cloud as a supplement, in order to get on each date on low temperature The minimum temperature distribution of full coverage of the 1 km resolution. The national standard definition has not been on the low temperature duration, the cold accumulated temperature index, which is the sum of low temperature period daily minimum temperature value to determine the likely rape the frozen disaster area. The results show that cold accumulated temperature can capture in Northern Anhui (mainly Suzhou area) and Jianghuai region (mainly in Hefei and Chuzhou area) the temperature distribution, and through the observation and the weather station were compared (correlation coefficient r=0.810, P0.01, RMSE=8.6, c) estimation of cold temperature and achieved good results. (2) the remote sensing identification and extraction of rape rape from flowering to flowering green declined, showing "Valley" feature in the time series of vegetation index profile, different from the other and winter crop growth period (mainly winter wheat), based on the two The planting area extraction green trend peculiar to a period of rape. The phenological growth taking into account the different phenological period of rape is not synchronized, the 8 day composite MODIS-NDVI based on the time series data, based on S-G filtering, through different flowering and flowering area identification start time to the region extract the distribution of Anhui Province rape, rape and municipal according to statistical yearbook recorded data on the planting area extraction results were tested. When the damage is more serious, after freezing rape during flowering plants tended to be low, the flower is sparse, at 30 m resolution on the scale of winter wheat and easily confused, extremely easy to cause underestimation of the actual planting area, in addition to agricultural disaster assessment also has new requirements. Combined with the actual situation of Hefei local rape production, is proposed in this paper to solve this problem The planting area of winter crop base + mask adjustment method, the adjacent normal growth season (no frost damage) the distribution of rape as a benchmark, from emergence to winter crops before winter growth characteristics are set up to adjust the reference mask based on the planting area, the result will be used to describe the distribution of frost occurrence in winter rape period. And on the basis of inspection of the rape planting area statistics. (3) remote sensing assessment for freezing rape rape frost disaster occurred on 2004, in MODIS pixel scale by time sequence profile analysis of 4 kinds of widely used for freezing vegetation index sensitivity. The results showed that NDVI and GNDVI outperformed EVI and this two SAVI vegetation index, compared with NDVI for the freezing effect performance is more continuous full time series.MODIS data has a long, suitable for vegetation index Anomaly assessment method of freeze injury. Select several adjacent normal growth season (2001-2006 excluding 2004) the average growth level of the same period as the benchmark, the NDVI annual standard deviation of normal annual fluctuation growth, the freezing effect after growth levels relative to the benchmark percentage difference is used to describe the extent of the disaster damage due to disaster data. The lack of field test, the average degree of damage the province's 47 counties and rapeseed cold accumulated temperature corresponding correlation analysis, a significant correlation between the results showed that two (r= 0.378, P0.01), to a certain extent to support the disaster assessment results of.MODIS low spatial resolution, suitable for screening large range of rape frost, but for the city (county) rape damage remote sensing monitoring and evaluation of the region, need to adopt a higher spatial resolution data. Considering the existing remote sensing data in space and time resolution for freezing rape occurred in 2010 and early 2011, this paper attempts to evaluate the damage of domestic environmental mitigation satellite data. Select 8 commonly used vegetation index constructed by the 4 band HJ-CCD as the candidate of the freezing indexes. Using the 2009-2010 growing season changes compared to post disaster disaster when each vegetation index is less than the normal growing season (2008-2009) with changes in the amount of normalized difference histogram curve to determine the different vegetation index for damage sensitivity. The results show that the spatial resolution in the 30m scale, NDVI and GNDVI as compared with other vegetation index has higher sensitivity, contrast sensitivity GNDVI the more high, and therefore suitable as evaluation index. Based on the freezing disaster phase relative to the GNDVI benchmark to determine the percentage difference of hair damage Grade students in the disaster, frost damage occurred in January 2011 for rape, this paper selects Hefei city 9 less monitoring points measured field freezing index to verify the results, the correlation analysis showed that a high degree of damage assessment by remote sensing and field measured freezing index (R =-0.698, P0.05). By using grey relational analysis method and statistical analysis method, this paper investigated the influence of some natural factors and their crop conditions on freeze injury degree, results showed that the south slope and the west slope to rape planting to disaster is relatively more serious, several factors influence the size of the disaster from high to low: pre disaster growth, soil moisture, the most the date of the cold surface temperature and altitude. (4) several physiological variables this paper constructs new frost sensitive index which constructs the moderate resolution imaging spectroradiometer and MERIS band based on MODIS Sensitive vegetation index, at the 1 km level resolution scale to examine the index in response to low temperature stress and damage mode sensitivity differences. The selected index (including MODIS-PRI, NDWI characterization of photosynthetic rate (canopy humidity) characterization and MERIS-MTCI (chlorophyll content), canopy, REP (characterization) and LAI (red edge position) the leaf area index). The results showed that PRI in response to cold stress is very strong, suggesting that photosynthesis for frost sensitive; NDWI showed abnormal rapeseed cold after the rise, presumably freezing leads to canopy water loss; LAI damage response to slow, not only the numerical decline is small and low temperature after rapid recovery; low temperature process did not cause MERIS-MTCI with the decrease of REP that may damage or weaken, did not cause the lower chlorophyll levels. Low temperature rise of MTCI was inhibited by the numerical level, and almost no REP Affected by low temperature, the red edge band of red edge NDVI sensitivity to low temperature stress is lower than the standard NDVI. based on the above findings, this paper selects 7 more sensitive to vegetation photosynthesis and canopy humidity level data in the MODIS-L1B band, in the random stratified sample of different damage level under the support of the best index (theoptimal index factor, OIF) analysis method and screened at 5,11 and 12 band combination construction and puts forward the frost sensitive vegetation index MFISI (MODIS Freezing Injury Sensitive new Index). The index can also response of photosynthesis and canopy humidity parameters of vegetation, has definite physiological significance; and as a result of the two physiological indexes in remote sensing on the level of damage is sensitive, clearly show the damage can affect the new index, the spatial distribution map of its value has rich terrain details The performance of the new index is better than the widely used NDVI. in the prediction of the frost damage and the prediction of the development trend of the frost damage.

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：S565.4;S426;S127

【相似文献】