苹果始花期是苹果生长发育的一个重要物候参数，也是与苹果产量密切相关的关键物候期。准确预测苹果始花期，对果园的疏花工作、霜冻灾害防御等精细化生产管理具有重要意义。由于现有的始花期预报方法多为基于气象站点数据预报，而地面气象站点分布稀疏，难以获取空间上连续的始花期物候信息。同时，苹果种植区多位于暖温带地区，且地形多为丘陵和山地，受复杂地形环境的影响，点尺度气象站点数据的空间代表性会降低全区域覆盖的始花期预报精度。针对上述问题，本研究以陕西省洛川县苹果优势种植区为研究区，提出一种基于高时空遥感地表温度数据预测苹果始花期的方法，以获取复杂气象和地形环境下空间连续的始花期物候信息。主要研究内容和结论如下：

（1）日尺度气温驱动的始花期预报模型构建研究。针对已有花期物候模型预报始花期过程中受逐小时气温数据时间分辨率限制的现状，本研究利用2019年10月1日-2020年4月30日的格网气象数据和花期物候序列模型，探索日冷/热积量与日最高气温Tamax、日最低气温Tamin、日平均气温Taavg之间的敏感性关系，并通过随机森林算法构建日尺度气温驱动的始花期预报模型。结果表明，利用Tamax+Taavg、Tamin+Taavg和Tamax+Tamin+Taavg 3组日气温特征因子构建的始花期预报模型在估算日冷热积量时结果较优，模型精度最高的为Tamax+Tamin+Taavg日气温特征因子组合。当利用该组合下始花期预报模型预报苹果始花期时，预报值与实测值间的决定系数（Coefficient of Determination，R²）为0.92，均方根误差（Root Mean Square Error，RMSE）为3.44天。上述结果表明利用本研究构建的苹果始花期预报模型可以有效将输入气温数据从逐小时尺度转换为日尺度，这在后续苹果始花期预报的工作中具有较好的应用价值和潜力。

（2）地表温度时空重建算法研究。为了避免受云、雨等天气的影响而导致原始地表温度数据时空不连续的问题，本研究提出了一种地表温度时空重建（Spatio-Temperal Reconstruction，STR）算法，通过计算有效地表温度像素对之间的空间距离因子和时间相似性因子构建加权函数，生成地表温度初步重建结果，然后利用时间序列谐波分析（Harmonic Analysis of Time Series，HANTS）算法得到研究区内时空无缝的地表温度数据集。结果表明，STR算法可以有效地填补不同缺失率下的地表温度，且重建后的地表温度与原始地表温度之间具有较高的空间连续性。通过对原始地表温度进行模拟缺失，利用STR算法重建后的白天时刻地表温度、夜间时刻地表温度和日平均地表温度与地表温度验证值之间的RMSE分别为1.21 ℃、1.10 ℃和1.07 ℃，上述结果表明该方法可以重建存在时空缺失的每日地表温度数据，并保证重建后数据的时空连续性。

（3）基于重建后地表温度的苹果始花期预报研究。以原始地表温度、格网气象数据、高程和经纬度作为变量，利用随机森林算法构建日气温估算模型，结合重建后地表温度数据集生成2019年10月1日-2020年4月30日研究区内包含每日最高气温、日最低气温和日平均气温的高时空连续气温数据集，并利用日尺度气温驱动的始花期预报模型进行苹果始花期预报。苹果始花期预测值与实测值之间的R²为0.72，RMSE为2.96天，结果表明该方法预报的始花期精度与传统基于真实逐小时气温数据驱动花期物候序列模型预报的精度相当。

本研究通过结合地表温度、气象数据与花期物候序列模型，针对传统基于气象站点数据预报苹果始花期方法中难以获得空间连续物候信息的问题，提出了一种遥感地温-气温协同下的苹果始花期预报方法，这可对未来遥感量化下果树物候期的预报研究提供参考价值。同时本研究提出的方法也为高效获得苹果始花期的空间分布提供了有效支撑，在果园精细化生产管理方面具有重要应用价值。

The first flowering date (FFD) of apple is an important parameter for apple growth and development, and is also a key phenology period closely related to apple yield. Accurate prediction of apple FFD is of great importance to the fine production management such as flowering thinning, frost disaster prevention in orchards. Most of the existing prediction methods are based on meteorological station data, it is difficult to obtain spatially continuous information on the FFD because of the sparse distribution of ground meteorological stations. At the same time, most of the apple growing areas are located in warm temperate regions, and the topography is mostly hilly and mountainous, so the spatial representativeness of point meteorological station data will reduce the accuracy of FFD prediction for the whole region. Aiming at the above problems, this paper takes the apple dominant growing areas in Luochuan, as the study area, and proposes a method to predict the apple first flowering date based on high spatial-temporal remote sensing land surface temperature data, it can obtain spatially continuous information on FFD in complex meteorological and topographic environments. The main research contents and conclusions of this thesis are as follows.

(1) Study on the construction of prediction model for the first flowering date driven by daily air temperature. In view of the current situation that existing flowering phenology models are limited by the temporal resolution of hourly air temperature data in predicting the FFD, based on the gridded meteorological data from October 1, 2019 to April 30, 2020 and the flowering phenology sequential model, the sensitivity relationships between daily chill and heat accumulation and daily maximum temperature (Tamax), daily minimum temperature (Tamin), and daily average temperature (Taavg) were explored firstly, and then constructed a prediction model for the first flowering date driven by daily air temperature using random forest algorithm. The results show that the model constructed using three groups of daily temperature characteristic factors (Tamax+Taavg, Tamin+Taavg and Tamax+Tamin+Taavg) have better results in estimating daily chill and heat accumulation, and the highest model accuracy was obtained for the combination of Tamax+Tamin+Taavg daily temperature characteristic factors. When Tamax+Tamin+Taavg is used to predict the apple FFD, the R² and RMSE between the predicted and measured values are 0.92 and 3.44 days, respectively. The model constructed in this study can effectively convert the input air temperature data from hourly scale to daily scale, which has good application value and potential in the subsequent work of apple FFD prediction.

(2) Study on the spatio-temporal reconstruction algorithm of land surface temperature. In order to avoid the problem of spatio-temporal discontinuity of land surface temperature data due to missing original data caused by clouds and rain, a Spatio-Temperal Reconstruction (STR) algorithm is proposed in this study. Firstly, a preliminary land surface temperature reconstruction result was generated by calculating the spatial distance factor and temporal similarity factor between effective land surface temperature pixel pairs, and then a spatio-temporal seamless land surface temperature dataset was obtained in the study area based on Harmonic Analysis of Time Series (HANTS). The results show that the STR algorithm can effectively fill in the land surface temperature at different missing rates, and the reconstructed land surface temperature have high spatial continuity with the original land surface temperature. By simulating the missing original land surface temperature, the RMSE between the reconstructed daytime, nighttime and daily average land surface temperature and the original land surface temperature using the STR algorithm are 1.21 ℃, 1.10 ℃ and 1.07 ℃, respectively. The above results show that the method can reconstruct daily land surface temperature data with spatial and temporal missing rates, and at the same time can ensure the spatial and temporal continuity of the reconstructed data.

(3) Study on apple first flowering date prediction based on reconstructed land surface temperature. Using the original land surface temperature, gridded meteorological data, elevation, latitude and longitude as variables, a daily air temperature estimation model was constructed using the random forest algorithm. Next, the reconstructed land surface temperature dataset was combined to generate a high spatial and temporal continuous daily air temperature dataset containing Tamax, Tamin and Taavg in the study area from October 1, 2019 to April 30, 2020. Then the prediction model for the first flowering date was used to predicting the apple FFD based on the daily air temperature dataset. The results show that the R² between the predicted and measured values of apple FFD is 0.72 and the RMSE is 2.96 days, which indicates that the accuracy of the FFD prediction by this method is comparable to the result by using the traditional sequential model based on real hourly air temperature data driven.

By combining the land surface temperature, meteorological data and the phenology sequential model, in response to the problem that it is difficult to obtain spatially continuous phenological information in the traditional method of predicting the apple first flowering date based on meteorological station data, this study proposes a method to predict regional scale apple first flowering date prediction based on collaboration of remote sensing land surface temperature and meteorological data, which can provide a reference value for future study of fruit tree phenological period prediction under remote sensing quantification. And the method proposed in this study also provides an effective support to obtain the spatial distribution of the apple FFD, which has important application value for orchard refinement production management.