生命科学学院知识库

在中国，半干旱黄土高原地区位于农牧交错带的中部，约占黄土高原总面积的 60%。这些地区降水量少，生态环境脆弱，农业生产能力低下已成为中国最大面积的贫困地区。目前，日益严峻的生态环境问题已经直接导致了半干旱黄土高原地区的粮食减产，使当地人民的生活更加困难。为了使卓有成效的粮食单产增产方法获得广泛推广，建立一种能够客观评价作物产量的方法是非常必要的。遥感技术具有客观、及时的特点，可以在短期内连续获取大范围的地面信息。作物生长动态模型能够详细描述作物生育期内光合、呼吸、蒸腾等重要生理生态过程。将卫星遥感数据和作物生长模型结合起来可以弥补它们各自存在的缺憾，在作物监测和产量预测方面发挥它们难以替代的作用。本论文在总结前人工作的基础上，主要做了以下三个方面的工作：1. 使用高分辨率的卫星影像（QuickBird）和产量效益模型相结合预测了榆中县中连川乡农作物的产量。并用 80 个遍布全乡的实收产量数据验证了模型的模拟结果。2. 通过使用数码相机提取植被盖度的方法来探索一种快速、可信度高和成本收益性好的植被盖度估算方法，以弥补高分辨率卫星影像成本太高的缺憾。在该方法中，首先将数码相机获取的影像从 RGB 空间转换到 HSI 色度空间，然后再使用面向对象的方法来进行绿色植被覆盖部分的提取。3. 首先使用植被指数转换模型来分别计算中巴地球资源卫星（CBERS）和QuickBird 影像的植被覆盖度数据；然后用第四章介绍的方法从数码照片中提取前面影像中同样样点的盖度值。比较各样点三种影像提取的盖度值的关系，以期建立以 QuickBird 影像为桥梁的数码影像盖度向 CBERS 盖度的尺度上推方法。在上面的研究中，我们得出了如下结论：
1. 产量效益模型中 NPP 的变异性主要受以下三个因素影响：入射的 PAR，fAPAR和 LUE。其中后两个因素因为卫星的高分辨率而得到改善。 fAPAR 和光谱植被指数（SVI）有很好的相关性，而 SVI 能够直接从 QuickBird 影像计算得到。LUE 的值也是基于从 QuickBird 影像提取的精准土地覆盖类型单个计算的。
2. 我们使用从农户调查的、遍布全乡的 80 个点的实收产量来验证模型的结果，预测值和实收值之间有很好的一致性，其决定系数为r2 = 0.86。结果证明将这种方法应用于监控集约式管理农田的作物生长模式是可行的。
3. 我们将提取的盖度数据与拍摄照片时同时取得的LAI值进行了比较。发现在每一个时期，不同种植密度下作物盖度和LAI值都存在较好的相关性（播种后 61天、80 天、90 天、100 天四个时相的决定系数依次为：r2 = 0.88，0.96，0.97和 0.92）。也就是说，这种方法提取的植被盖度数据是可信的。
4. 为了探究整个生长季作物盖度和籽粒产量的联系，我们分别计算了每一个小区不同时相的影像的盖度值，在小麦成熟时获得籽粒产量和地上生物量。不同密度下的作物盖度显示了相似的时间趋势。比较不同时相，作物盖度和籽粒产量或地上生物量相关关系的决定系数后发现，播后 80-90 天的作物盖度数据与它的最终产量和生物量关系最好。这个时候的盖度数据能够用来很好地估算作物的产量。
5. 植被盖度作为反映陆地表面植被数量的一个重要参数和生态系统变化的重要指标，已经受到了广泛的关注。目前，已有相当多的各种空间、时间分辨率的遥感数据用于植被盖度的估算。在实际应用中，往往需要将不同测量平台上的高空间分辨率数据与时间分辨率高的低空间分辨率数据整合，形成不同空间分辨率的影像数据层次体系，为地表现象多尺度描述或应用提供丰富的数据源。
6. 从数码相机影像获取的植被盖度数据通过 QuickBird 影像盖度，尺度上推到CBERS 影像。探索了一种基于高分辨率卫星影像的可行的尺度上推方法，同时也为 CBERS 影像的进一步应用提供了实验基础。

In the semi-arid Loess Plateau of China,numerous poor people and the minority live in this place,and they depend on the barren land for food. Moreover,fragmentation and degradation of the ecological environment have been accelerated due to the increasing population pressure and intensification of land use.Declining crop yield,as a part of the ongoing land degradation process,is considered a severe environmental problem.To address this,many research efforts have focused on increasing the crop yield. Prior to extend the techniques,an accurate estimation of crop yield is essential.Therefore,a method that could estimate crop yield over large geographic areas would be highly desirable.Remote sensing offers great potential for regional production monitoring and estimates.A production efficiency model(PEM)can be used to estimate crop growth based on the use of satellite remote sensing of canopy light absorption.By combing with remote sensing data and PEM are potentially useful for NPP monitoring or yield forecasting at regional scale.In this dissertation, we have carried out some original studies as follows:
1. To describe a method of integrating QuickBird imagery with a PEM to estimate crop yield in Zhonglianchuan,a hilly area on Loess Plateau.Based on the high spatial resolution QuickBird imagery,a site-specific land cover classification is used to study the variability of light-use efficiency (LUE).The fraction of absorbed photosynthetically active radiation (fAPAR) is related to spectral vegetation indices (SVIs),which are also derived from the QuickBird imagery.80 plots scattered throughout the village were used to validate the model outputs at the village level.
2. A method to accurately and conveniently estimate crop cover from digital camera images is presented.The digital camera images were converted from RGB to HSI color space.Hue segmentation technique enhanced the characteristics of plant tissues,and identified green tissues easily.Vegetation and non-vegetation binary pseudo-color images were produced and crop cover was calculated for each plot.Percent of pixels corresponding to the vegetation was then calculated and used as the percent coverage for each plot.
3. At first,China-Brazil Earth Resource Satellite (CBERS) and QuickBird images were used to derive the vegetation cover data at 62 plots by Vegetation Index Transform Model.The vegetation cover data at same plots were also derived from digital camera images by the method of Chapter 4 mentioned. When comparing the vegetation cover data derived from digital camera images with a CBERS 20 m pixel,information from QuickBird 2.4 m pixels provides a useful tool for scaling from point to area.
The conclusions are as follows:
1.Variations in NPP for the PEM model arise from changes in three factors:incident PAR,fAPAR, and LUE.The fAPAR is related to SVI,which is calculated from high spatial resolution QuickBird imagery.LUE varies among crop type and is calculated based on site-specific,reliable land cover classification, which is also derived from the QuickBird imagery.
2. Farmer-reported yields at 80 plots scattered throughout the village were used to validate the model outputs at the village level.The prediction approximated well with the harvested yield (r2 = 0.86).Our results demonstrate that the method used to monitor crop growth and yield in intensively managed agricultural lands is effective.3.Crop cover data and LAI value were compared at different stages,to validate the accuracy of the abovementioned method. There were high correlations between crop cover and LAI at four stages (61,80,90,and 100 DAS;r2 = 0.88,0.96,0.97,and 0.92,respectively).
4. To explore the association between crop cover and grain yields throughout the season,crop cover derived for each plot and imagery date were correlated with their respective grain yields. Crop cover at different densities showed similar temporal trends.By comparison the coefficients of determination(r2) of the relationships between aboveground biomass or crop yield and crop cover determined at different growth stages,the results clearly indicate that crop cover at 80–90 DAS was a good predictor of wheat yield.
5. The ground cover of vegetation is an important element of models that attempt to account for the exchanges of carbon, water, and energy at the land surface.The cover is also a sensitive indicator of land degradation and desertification in arid and semi-arid regions. Remote sensing data with multiple resolutions from different sensors have been extensively used to derive the vegetation cover.We have used high-resolution satellite (QuickBird) data in conjunction with the vegetation cover derived from digital camera to test the feasibility of CBERS image.6.As high resolution QuickBird data asconjunction.we developed a relationship between high spatial resolution Digital Camera imagery (3 mm) and CBERS based data (20 m) that will serve to provide vegetation cover for various modeling scales.