基于阴阳叶分解光能利用率模型和云覆盖信息的全天候水稻遥感估产

Timely and accurate prediction of rice and other crops yield information at regional scale is closely related to the people's livelihood, which has been attached great importance by all levels of government. The widely used empirical statistical models on rice yield prediction lack universality, and crop growth and remote sensing coupled models are too complicated. Although light use efficiency models have the characteristics of the mechanism and practical applicability, the application of the model has been limited because the commonly-used light use efficiency models highly lumped canopy different response characteristics of sunlit leaves and shade leaves to light, temperature, and cannot estimate rice yield with all-weather conditions due to ignoring the difference of light use efficiency with different sky conditions such as clear and cloudy days. In view of the above problems, based on the extraction of high-purity pixels of rice field, the project will to build an improved light use efficiency model with sunlit and shaded leaves separation for non-cloudy sky conditions, considering the differences of response characteristics of sunlit and shaded leaves to radiation and temperature, and to build an improved light use efficiency model based on per unit leaf area index for overcast sky conditions; and, at the same time, to establish the adaptation function of light use efficiency in different weather conditions by radiation intensity, making the light use efficiency model to be applied to the changeable weather conditions. And on the basis of comprehensive evaluation of cloud cover information extracted from MODIS Terra and Aqua data, the calculation method for net primary productivity is selected to calculate the net primary productivity in each pixel for whole growth period. And combined with the harvest index and other parameters, rice yield can be estimated. Thus, a rice yield estimation method with all-weather conditions is established by a daily step, and the use of the method is not be limited by weather conditions. The method not only is expected to further improve the yield estimation accuracy and stability, but also can achieve an integrated monitoring of quantitative rice growth status during full growing period and rice yield.

水稻遥感估产研究目前广泛使用的经验模型缺乏普适性、作物生长耦合模型过于复杂，虽然光能利用率（LUE）估产模型兼具了机理性与实用性的特点，但当前使用的LUE模型高度集总了冠层阳叶与阴叶对光、温的不同响应特征，并忽略了晴阴等不同天气状况下LUE的差别，难以全天候估产。针对以上问题，本研究在水稻高纯度像元提取的基础上，拟构建针对非阴天状态的考虑冠层阴叶与阳叶光、温响应差别的阴阳叶分解LUE模型和针对阴天状态的单位叶面积指数LUE模型，并建立不同天气状态下辐射量对LUE的调节函数，使LUE模型能够应用于多变的天气条件；再依据每天MODIS上午星和下午星云覆盖信息综合评判的天气状态选择LUE模型，计算生育期内每个像元净初级生产力，并结合收获指数等参数估算水稻单产，从而建立一种以"天"为步长且不受天气状况影响的全天候水稻估产方法，其有望进一步提高估产的精度和稳定性，且能够实现全生育定量长势的同步监测。

水稻遥感估产能够及时准确地获取大区域水稻产量信息，对国家的粮食生产、贸易及粮食安全预警具有重要意义。目前常用的水稻遥感估产模型中经验估产模型缺乏普适性、作物生长模型复杂难用，虽然光能利用率GPP/NPP估产模型兼具了机理性与实用性的特点，但是作为半经验模型的光能利用率模型，在反应生态系统过程和机理方面仍有不足。该模型在冠层尺度上未考虑冠层阴叶、阳叶之间的差别，在时间上也未考虑日内及日间的气象要素变化对光合作用的影响，因此基于传统光能利用率GPP/NPP估产模型往往会存在较大误差。为此本研究将光合机理模型与经验耦合构建了一个基于阴阳叶分解的GPP/NPP水稻估产新模型，并在小区尺度和区域尺度对该模型进行了验证。具体研究进展结果如下：.（1）基于Landsat 8数据，利用最大似然法对江苏区域进行监督分类，地级市水稻面积提取精度在85%左右。基于MODIS数据的地级市水稻面积整体精度比较低，但能够提取研究区面积比较大的田块，可用于确定水稻估产使用的抽样像元。.（2）采用MOD09A1数据，计算不同时期的水稻EVI，建立水稻估产的统计模型进行产量估算，各个地级市的水稻单产估算误差基本在10%以内。并利用水稻种植期内NDVI最大值数据实现了将地级市水稻单产空间化为像元单产。.（3）在浙江省德清县开展了针对新水稻估产模型验证的小区试验，该试验设置了两个主栽品种的S1（嘉58）和S2（秀优378）。研究发现模型中S2收获系数大于S1，利用新水稻估产模型对S1和S2产量估算误差分别为6.5%和3.0%。.（4）在小区试验验证的基础上，选择江苏省为研究区开展区域尺度估产，同时考虑收获系数随时间的线性变化特征，动态调节收获系数。分别对2004-2014年江苏省各地级市单产进行预测，江苏省地级市平均估产精度基本上优于96%，表明该模型能够较好的预测了区域水稻单产。. 本研究发表文章7篇，其中SCI论文5篇，获得授权发明专利3项。研究中所构建的水稻估产新模型提供了一种实用的水稻产量估算方法，该方法不仅模型结构件相对简单，而且具有较好的理论依据，为遥感数据支持下的广域尺度长时间序列水稻估产提供必要的技术支持，将进一步促进水稻遥感估产应用和发展。