Mapping and Monitoring Rice Agriculture with Multisensor Temporal Mixture Models

Rice feeds more humans than any other crop on Earth. Accurate prediction of the timing and volume of rice harvests therefore has considerable global importance for food security and economic stability, especially in the developing world. Optical and thermal satellite imagery can provide critical constraints on the spatial extent of rice planting and the timing of rice phenology. We present a novel approach to the mapping & monitoring of rice agriculture using Temporal Mixture Models (TMMs) derived from parallel spatiotemporal analyses of coincident optical and thermal Landsat image time series. Using the Sacramento Valley of California as a test area, we characterize regional rice phenology in terms of both fractional vegetation abundance (Fv) and brightness temperature (Tb). We compare satellite Tb retrievals to station data and find uncorrected Tb to compare with the upper bound of the envelope of air temperature observations to within 3°C on average. Results from parallel spatiotemporal analyses of coincident Fv and Tb image time series over the 2016 & 2017 growing seasons suggest that TMMs based on single year image time series can provide simple and accurate maps of crop timing, while TMMs based on dual year image time series can provide comparable maps of year-to-year crop conversion. Fv time series data show particular promise for estimating crop timing, while Tb appears particularly well suited for discriminating between rice and other crops. We also build a sample model midway through the 2018 growing season to illustrate a potential near-realtime monitoring application. Field validation confirms that the mid-2018 monitoring model provides an accurate upper bound estimate of the spatial extent and relative timing of the rice crop, even under conditions of relative data scarcity. The implications of these results could have potential utility for further analyses of precision agriculture, pest management, evapotranspiration (ET) and cropping practice verification.

稻米作为地球上滋养人类最多的农作物，其收获时间与数量的准确预测对于全球粮食安全与经济稳定具有重大意义，尤其在发展中国家。光学与热红外卫星影像能够为稻米种植的空间范围及物候期提供关键的限制条件。本研究提出一种基于并行时空分析的光学与热红外Landsat影像时间序列的时序混合模型（TMMs）来绘制与监测稻米农业的新方法。以加利福尼亚州萨克拉门托河谷为测试区域，我们从分数植被丰度（Fv）和亮度温度（Tb）两个维度对区域稻米物候期进行表征。我们将卫星Tb反演结果与地面数据进行比较，发现未经校正的Tb与空气温度观测值的上限之间平均误差在3°C以内。对2016年至2017年生长季节期间同步的Fv和Tb影像时间序列的并行时空分析结果表明，基于单年影像时间序列的TMMs能够提供简单而准确的作物生长时间地图，而基于双年影像时间序列的TMMs则能提供年度间作物转换的相似地图。Fv时间序列数据在估算作物生长时间方面展现出巨大潜力，而Tb则特别适用于区分稻米与其他作物。我们还在2018年生长季节中途建立了一个样本模型，以展示一种潜在的近实时监测应用。田间验证证实，2018年中期的监测模型即使在相对数据稀缺的条件下，也能提供稻米种植空间范围和相对生长时间的准确上限估计。这些结果对精准农业分析、病虫害管理、蒸散量（ET）以及作物种植实践验证具有潜在的应用价值。