【作者】 高建峰；

【导师】 潘剑君；

【作者基本信息】 南京农业大学 ， 土壤学， 2007， 博士

【摘要】 近二十年来,我国经济发达地区土地利用发生了前所未有的快速变化。城市迅速扩展、农田大量流失,使得人地矛盾更加尖锐,伴随而来的环境恶化、土地质量下降等问题非常突出,并且制约了经济的快速协调发展。因此开展不同区域土地利用变化的动态监测研究是迫在眉睫的任务。但传统的统计方法不仅费时费力,且时效性、准确度都不高,遥感作为一门技术手段具有广域性、周期性、连续性和经济性的特点,可以快速有效地获取研究区域在某时间段内土地利用的变化信息。而新一代中等分辨率光谱仪（MODIS）具有高光谱和高时相分辨率的优势,且其空间分辨率也由NOAAAVHRR的1 km提高到250 m,非常有利于捕捉连续的土地利用变化信息,而且其对更高空间分辨率、低时相分辨率比较适合中小尺度土地利用变化研究的TM/ETM+影象具有明显的补益作用。因此,本研究以我国经济发达地区江苏省吴江市为例,使用1984、1991、1998和2005年TM/ETM+影象研究了水稻田近20年来的土地利用变化情况,着重探讨了MODIS数据与TM/ETM+相结合的方法来提取土地利用变化信息的方法,并进一步应用遥感解译的土地利用变化结果结合第二次全国土壤普查数据、2003年全国土壤定位监测数据及吴江市土壤年度定位监测资料研究了水稻表层土壤有机碳库的变化情况,主要研究结果如下:●根据遥感解译,吴江市1984-2005年土地利用分1984-1991年,1991-1998年和1998-2005年三个时段监测结果:1)吴江市1984年和1991年水稻种植面积分别为714.62 km²和629 km²,7年间下降了85.62 km²,其中建设用地、林地、水体和旱地分别占用稻田35.67km²、13.79 km²、19.51 km²和16.65 km²;2)1998年稻田面积为542.95km²,1992-1998年间水稻种植面积下降了86.05 km²,建设用地、林地、水体和旱地分别占用稻田面积为:47.31 km²、15.8 km²、13.01 km²和9.93km²;3)1998-2005年间土地利用变化有所加快,其中稻田转化为建设用地和林地的面积分别为:253.28km²和62.63km²,原1984年稻田在2005年土地利用分布为:稻田254.79 km²、建设用地336.26 km²、林地92.22 km²、旱地10.25 km²和水体21.10km²。●基于MODIS数据的多种分类特征选择和良好的分类器能提高区域尺度土地利用分类的精度。在不同分类器比较试验中,以应用MODIS植被指数、水分指数时序系列进行掩膜的分类方法分类效果优于其它分类方法,总体分类精度达到66.60%,决策树分类法也表现出良好效果,在分类树中引入近年发展起来的一种机器学习领域內的增强技术-boosting技术,能提高类别的分类准确率。而神经网络分类精度并不比传统分类方法好,单独使用MODIS影象在中小尺度难以取得较好的分类效果。本研究进一步探索使用2000-2005年MODIS影象与高空间分辨率遥感信息源2005年ETM+相结合的方法来提取土地利用变化信息,结果表明,辅以2005年ETM+的2000-2005年MODIS影象分类方法能够取得较好的分类效果,2000年MODIS影象相对于2000年TM影象的分类精度分别为:水体91.67%、稻田85.40%、建设用地92.09%、旱地70.80%和林地71.93%,该分类方法的总体分类精度达到了85.25%。根据辅以2005年ETM+的2000-2005年MODIS影象分类结果,吴江市2000-2005年的水稻种植面积分别为:368.36km²、315.21km²、297.50 km²、237.77 km²、261.64 km²和254.79 km²,其结果与吴江市农业统计资料比较,两者具有显著的相关性（R²=0.9873）,进一步验证了辅以ETM+影象的MODIS影象土地利用动态监测方法具有较高的利用潜力。●根据辅以2005年ETM+的2000-2005年MODIS影象分类结果,吴江市水稻种植面积在2003年以前下降幅度比较大,但是2003-2005年间又有小幅回升,而旱地面积在2003年前快速上升,而2003年后又明显下降,原因是2003前的建设用地包含了许多未利用土地,2003后政府采取了强有力的耕地保护措施,提高了土地利用效率,部分未利用耕地又得到重新恢复。建设用地占用稻田的面积从2000年258.98 km²增加到2005年336.26 km²,6年间增加了29.8%。水域面积2000-2005年基本保持稳定。林地占用稻田的面积也呈逐年增加趋势,从2000年的33.59 km²增加到2005年的92.22 km²。总体来看,吴江市的土地利用变化受人为因素影响很大。●根据2003年全国耕地质量调查与第二次土壤普查结果比较,吴江市水稻土有机碳含量普遍有上升的趋势,含量在15.20g.kg^-1的样本比例有了较大的提高,平均有机碳含量从1982年的16.63 g.kg^-1增加到2003年的16.95 g.kg^-1,年增长率约0.01g.kg^-1。稻田转化为林地和旱地后土壤有机碳含量則不同程度的下降,年下降率分别为:0.06g.kg^-1a^-1和0.03g.kg^-1a^-1。●结合遥感监测土地利用变化结果与土壤有机碳测定结果估算水稻土表层土壤有机碳库变化结果如下:2005年稻田表层土壤（0-20cm）有机碳储量为2686047 t,相对于1984年的有机碳储量2672884t,22年来共增加了0.49%,其中1984-1991年增加了9454.9t,1991-1998年增加了5834.1t,1998-2005年增加了2125.9t,增长率呈现不断下降态势,原因是稻田转化为建设用地和水域后不再具有固碳效用,而且水稻田转化为林地和旱地后对土壤有机碳还具有释放效应,所以当前的土地利用变化对增加土壤碳储量非常不利。总之,本研究中所采用的使用基于TM/ETM+影象与MODIS影象进行土地利用变化动态监测方法取得了较理想的结果,特别是辅以ETM+的MODIS影象土地利用变化监测方法表现出较好的利用潜力,该方法可以为我国其它经济发达地区特别是在更大空间尺度进行土地利用变化研究提供参考。根据基于遥感的土壤有机碳计算结果,吴江市土地利用变化对于增加土壤有机碳库非常不利,而基于第二次土壤普查的资料对发达地区土壤有机碳估算值与实测数据有较大的出入。因此在估算我国经济发达地区水稻土碳储量变化时,有必要借助遥感等技术手段获取适时的土地利用变化信息。更多还原

【Abstract】 The land-use in developed area of China has changed dramatically in recent 20 years,of which,agricultural land area has decreased substantially due to rapid industrialization and urbanization.Changes in land-use may have significant effects on enviroment,soil quality and may limit the increasing of economic development,so monitoring the changes of land-use has been the case.But the traditional statistical method for acquiring the land-use data is time-consuming and hard sledding;besides, calculation mistakes easily occur and could result in errors.Satellite images for land and crop classification are in common use and remote sensing has become a powerful tool in the real-time identification of rand-use classes.A new generation of advanced optical sensor-the Moderate Resolution Imaging Spectral-radiometer（MODIS） provide dally global imagery at spatial resolutions of 250 m（red and NIR1） and 500 m（blue,green,NIR2,SWIRl,SWIR2） and hay obvious advantages for identifying paddy fields pixels compared to NOAA AVHRR 1 km spatial resolutions.The Landsat TM and ETM++ imagery offer much better quality data sets but at infrequent intervals to capture the seasonal dynamics of crops.Therefore,the combined TM /ETM+ data and MODIS data may be more suitable for timely monitoring the land-use changes of paddy fields.In this study,we test the performance of combining MODIS and TM/ETM+ analysis to extract yearly land-use informations of Wujiang County during 2000-2005.Land-use classmaps derived from the TM/ETM+ and MODIS data were further used to estimate the topsoil SOC pool and the changes of sequestration potential of paddy soils in Wujiang County（one of the earliest open districts in theYangtze River Delta of China）.The soil organic carbon（SOC） data were obtained from the 2nd State Soil Survey（2SSS）,the nationwide arable soil monitoring system（NASMC） and the county level soil reconnaissance data.The major results are as follows:●According to the TM/ETM+ images analysis:1) The rice-plant area of 1984 paddy fields area has decreased 85.62 km² during 1984-1991.The construction land,woodland（including orchard land,mulberry field,nursery land）,upland and water body occupied 35.67 km²,13.79 km²,19.51 km² and 16.65 km², respectively.2) The rice-plant area of 1984 paddy fields was 542.95 km² in 1998, having reduced 86.05 km² during 1991-1998.The construction land,woodland, upland and water body occupied 47.31 km²,15.80 km²,13.01 km² and 9.93 km², respectively.The loss of the rice area during 1991-1998 is near to 1984-1991.But areas of the new-built of construction land and woodland during 1991-1998 were greater than 1984-1991,account for 54.98%and 18.36%of the total loss of paddy area,respectively.The areas of the new-built upland and water body during 1991-1998 were less than 1984-1991,account for 15.12%and 11.54%of the total loss of rice area,respectively;3) The land-use changes have speeded up during 1998-2005,the construction land and woodland occupied 253.28 km² and 62.63 km² of rice-plant area,respectively.The areas distribution of different land-use types of 1984 paddy fields in 2005 were as follows:rice 254.79km²,construction land 336.26 km²,woodland 92.22 km²,upland 10.25km² and water-logged land 21.10km²,respectively.●Classification feature derived from MODIS such as the vegetation indices,water indices could improve the accuracy in land-use classification.Among various classifiers,the new developed classifier by using the time series of vegetation indices and land surface water indices has the greatest potiential in monitoring the changes of land-use,and its accuracy achieved 66.60%.The decision tree classification,an algorithm based on human-computer interaction is applied to remote sensing image classification with better flexibility and robust compared with classic automatic classification technology,Classification accuracy based on neural network is not improved compared to Maximum Likelihood Classifier （MLC）.In all,using a good classifier and choosing suitable features are important in MODIS land-use classification.●Our newly-built land use classification method by combining MODIS and ETM+ data is favorable and feasible for land-use mapping.The total accuracy of 2000 MODIS land-use map is 85.25%when compare to the results of 2000 TM land-use classification.A significant linear relationship between 2000-2005 MODIS derived paddy fields areas and statistical data（R² is 0.9873） demonstrated the potential of the newly-built MODIS classification method. According to the results of 2000-2005 MODIS images land-use classification,the total paddy fields area has decreased by 30%（113.57 km2） during 2000-2005. The area of upland was estimated to be nearly doubled during 2000-2003 but dropped dramatically during 2004-2005.This is probably because the fallow land was forbidden by the government and the fallow land was returned to paddy fields again.The waterlogged area was relatively stable in recently six years.●The SOC contents for the plow layers in the paddy fields in 2003 follow a normal distribution.About 40%,30.6%and 12.5%of the soil samples of paddy soils had SOC contents ranging from 17.5 to 20 g·kg^-1,from 15 to 17.5 g·kg^-1 and 12.5 to 15 g·kg^-1 in 2003.In 1984,in contrast,24.06%,39.89%and 18.97%of the soil samples had SOC contents ranging from 17.5 to 20 g·kg^-1,from 15 to 17.5 g·kg^-1 and from 12.5 to 15 g·kg^-1,respectively.The percentage of samples with SOC contents greater than 20 g·kg^-1 or less than 12.5 g.kg^-1 in 2003 was relatively similar to that in 1984.However,the percentage of samples with SOC contents of 17.5-20 g·kg^-1 in 2003 was greater than that in 1984.During 1980 to 2003,the SOC contents of paddy soils have been increased by 0.32 g·kg^-1.This suggests that irrigation-based rice cultivation in Wujiang County has significantly increased SOC content compared to the data obtained by the 2^nd SSS.●The average SOC content of paddy soils exhibited an increase（ASOC） of 0.01g·kg^-1 per year during the past 20 years.On the contrary,obvious decreases in the SOC content were found in uplands and woodlands.The annual decrease rate（△SOC） in uplands and woodlands were 0.03 g·kg^-1 and 0.06 g·kg^-1, respectively.The changes of SOC content under different land-uses conversions reflects that land-use changes may play an important role in C sequestration.●The C stock in 2005 has increased by 0.49%during 22 years relative to the total of 2672884t C stock in topsoil of paddy fields in 1984.Of which,the C stocks have increased by 9454.9t during 1984.-1991,5834.1t during 1991-1998,2125.9t during 1998-2005.The increase rate of SOC stocks has reduced as a consequence of decreasing cultivation area of paddy rice since 1984.The total SOC loss has exceeded the total SOC sequestration since 2001 in 1984 paddy fields.Given that the SOC content of buried soils in construction land and newly converted fish-breeding water body keeps stable,the SOC was observed to decrease after paddy fields had been converted to plant nursery land during 1984 to 2005. Moreover,conversion of rice paddy to upland had been also shown decline in the SOC.In all,the method using MODIS and TM/ETM+ images for monitoring the changes of land-use is favorable in the study area during 1984 to 2005.Especially,the yearly land-use changes monitoring by using MODIS in addition to ETM+ images has obtained a reasonable results.The method used in this research could be regarded as a reference for land-use monitoring in other region or more large spatial scale in other developed area in China.According to the SOC stocks calculating results,the currently land-use changes in Wujiang County are not favorable for the increasing of the soil C stocks.Further more,the calculation of carbon pool based on data from 1980s soil survey could result in errors because dramatic change of land-use in China during the recent 20 years may has a great influence upon soil C stocks.To accurately estimate soil carbon pool in these areas,the timely informations about land-use changes based on the remote sensing are very useful.更多还原