To study the effects of heat stress in dairy cattle, animal performance records are merged with weather data from nearby meteorological stations. However, the stations location and distribution may render them less reliable. Alternatively, weather information from several meteorological stations can be interpolated using empirical algorithms to produce gridded estimates of weather parameters at desired spatio-temporal resolutions. The objective of this study was to compare and apply observational weather station data (WSD) and gridded numerical weather prediction (NWP) model data with high spatio-temporal resolution to identify thermal stress thresholds for milk production traits. Weather data were from the German meteorological service and reanalysis was based on the COnsortium for Small-Scale MOdelling (COSMO)-REA6 model. Milk performance data were from over 16 million monthly test-day records for the period 2010 to 2019 in southern Bavaria, Germany. Individual cow records were transformed to herd averages resulting to 797,455 herd test-days from 9,726 herds. These were merged with temperature and temperature humidity index (THI) data from 53 weather stations and corresponding gridded data. There was good agreement between WSD and NWP model data with correlation coefficients of 0.97 for both daily average temperature and THI and 0.84 for relative humidity. However, positive and negative biases were observed in the pre-Alpine regions. The average herd reaction norms were in good agreement and followed similar trends when estimated from WSD and NWP. The heat stress threshold at which milk yield and protein yield started to decline was 16.0 °C for temperature and 60 units for THI. The responses of fat yield, protein and fat contents were generally linear decline with no definite thresholds. Milk urea increased in a non-linear accelerating pattern and there was no undesired effects on SCS. The thresholds obtained in this study may be applied to implement necessary management strategies to mitigate losses.

为了研究热应激对奶牛的影响，动物性能记录与附近气象站的天气数据相结合。然而，站点的位置和分布可能会降低它们的可靠性。或者，可以使用经验算法对来自多个气象站的天气信息进行插值，以在所需的时空分辨率下生成天气参数的网格估计。本研究的目的是比较和应用具有高时空分辨率的观测气象站数据 (WSD) 和网格数值天气预报 (NWP) 模型数据，以确定产奶性状的热应力阈值。天气数据来自德国气象局，再分析基于小尺度建模联盟 (COSMO)-REA6 模型。牛奶性能数据来自德国南部巴伐利亚州 2010 年至 2019 年期间超过 1600 万个月度测试日记录。将个体奶牛记录转换为牧群平均值，从 9,726 个牛群中得出 797,455 个牛群测试日。这些数据与来自 53 个气象站的温度和温度湿度指数 (THI) 数据以及相应的网格数据合并。WSD 和 NWP 模型数据之间的一致性很好，日平均温度和 THI 的相关系数为 0.97，相对湿度的相关系数为 0.84。然而，在前高山地区观察到正偏差和负偏差。从 WSD 和 NWP 估计时，平均群体反应规范非常一致，并遵循类似的趋势。牛奶产量和蛋白质产量开始下降的热应激阈值为 16。温度为 0 °C，THI 为 60 个单位。脂肪产量、蛋白质和脂肪含量的反应一般呈线性下降，没有明确的阈值。牛奶尿素以非线性加速模式增加，并且对 SCS 没有不良影响。本研究中获得的阈值可用于实施必要的管理策略以减轻损失。