基于多源卫星数据的东北主要农作物适宜播种期动态制图方法研究

本项目以黑龙江、辽宁、吉林和内蒙古为研究区，以玉米和大豆播种期遥感监测为目标。该区2019年玉米总产量占全国的44%，2018年大豆总产量占全国的32%，在全国的玉米和大豆生产中具有重要地位。但热量条件有限，播种迟会增加后期遭遇低温冷害的风险；因此，开展研究区玉米和大豆适宜播种期遥感监测对于玉米和大豆高产稳产、降低灾害风险具有重要应用价值。项目将优选基于被动微波的陆地表面温度（LST）反演方法；研究基于热红外和被动微波遥感数据反演的LST的数据融合方法，构建研究区全覆盖、高时空分辨率的LST数据集；利用 LST、植被指数、云量、土地覆盖、经度、纬度、海拔高度、坡度和坡向等因子作为自变量，构建基于深度学习的土壤日平均温度遥感估算模型；利用实测的播种期和土壤温度数据，确定玉米和大豆适宜播种的土壤温度指标；研制作物适宜播种期遥感制图原型系统，实现适宜播种期制图自动化，用于指导农业生产。

The study area of this project includes Heilongjiang, Liaoning, Jilin and Inner Mongolia province (autonomous region), in the northeast China. The research object is to develop framework to automatic mapping of suitable sowing dates（SSDs）for the corn and soybean using satellite data. Corn is one of the staple grain crops with the highest production in China. In 2019, the corn planting area in the study area account for 40% of the total corn planting area in China. In 2018, the soybean planting area in study area account for 45% of the soybean planting area in China. The corn and soybean production in the study area plays an important role in ensuring the national food security. However, the study area locates in high latitude zone and the frost-free periods are short. If sowing late, the risk of cold damage will increase, and will lead to the decrease of yield and quality. Therefore, sowing on time is the basis of increasing the rate of corn and soybean emergence, ensuring crop normal growth and development, promoting high and stable yield, and reducing disaster risk..Soil temperature is the key indicator of the suitable sowing forecast of corn and soybean. At present, the soil temperature used for forecasting the crop planting dates is mainly observed by meteorological stations, but the number of stations is limited and their distributions are uneven. In addition, the soil temperature varies greatly due to the topographic effect. Therefore, it is difficult to accurately forecast crop sowing dates spatially based on the soil temperature data from the small amount of meteorological stations to meet the need of precise planting and management..The land surface temperature (LST) derived from thermal infrared or passive microwave satellite data are the main independent variables in the construction of the soil temperature estimation models. Although the LST derived from thermal infrared sensors has high precision, it is susceptible to cloud and rain, which results in data loss. The LST inversed from passive microwave are all-weather, but the accuracy is lower than the LST from thermal infrared sensors. The project will select the optimum LST retrieval method based on passive microwave, and make full use of the thermal infrared LST and passive microwave LST to construct the full coverage daily LST with high spatial-temporal resolution (1km) in the study area..At present, statistical regression model is the mainly methods used to estimate soil temperature in different depths based on LST. It is easy but its accuracy is limited. As deep learning can simulate the complex nonlinear relationship, it provides the possibility to improve the accuracy of soil temperature estimation by remote sensing. Therefore, the different deep learning models such as Long Short Term Memory or Transformer will be used to construct the soil temperature models using LSTs, vegetation index, cloud cover, land cover, longitude, latitude, altitude, slope and aspect as independent variables. The results will be compared with those of baseline methods such as multivariate regression, Support Vector Machine, and Random Forest. The best deep learning model for soil temperature estimation will be selected..Based on the measured data of the sowing date and soil temperature from meteorological stations, the suitable indicators for corn and soybean sowing date will be determined. The SSDs of corn and soybean will be dynamically mapped..The project will develop a remote sensing mapping prototype system for crop SSD including remote sensing data preprocessing, passive microwave LST retrieving, data fusing with multiple satellite data, full cover LST data set establishing, soil temperature estimation, planting dates mapping. The SSDs of corn and soybean will be map automatically from satellite data to the results based on the prototype system.