Spatial variations of crop yields are commonly observed in typical rainfed systems worldwide. It is accepted that such variations are likely to be associated, among other factors, with water spatial variations due to lateral water flows occurring in fields with undulating topography. However, some of the main processes governing water spatial distribution such as lateral flow are not entirely considered by the most commonly adopted crop simulation models. This brings uncertainty to the process of yield simulation at field-scale, especially under water-limited conditions. Although it is expected that lateral water movement determines spatial variations of crop yields, it is still unclear what is the net contribution of lateral water inflows (LIF) to spatial variations of rainfed yields in fields of undulating topography. In this sense, by combining field experimentation, simulation models (HYDRUS-1D and AquaCrop), and the use of artificial neural networks, we assessed the occurrence and magnitude of LIF, and their impact on wheat yields in Cordoba, Spain, over a 30-year period. Seasonal precipitation varied over 30 years from 212.8 to 759.5 mm, and cumulative LIF ranged from 30 to 125 mm. The ratio of seasonal cumulative LIF divided by seasonal precipitation varied from 10.7% to 38.9% over the 30 years. The net contribution of LIF to spatial variations of rainfed potential yields showed to be relevant but highly irregular among years. Despite the inter-annual variability, typical of Mediterranean conditions, the occurrence of LIF caused simulated wheat yields to vary + 16% from up to downslope areas of the field. The net yield responses to LIF, in downslope areas were on average 383 kg grain yield (GY) ha⁻¹, and the LIF marginal water productivity reached 24.6 ( ± 13.2) kg GY ha⁻¹ mm⁻¹ in years of maximum responsiveness. Decision makers are encouraged to take water spatial variations into account when adjusting management to different potential yielding zones within the same field. However, this process is expected to benefit from further advances in in-season weather forecasting that should be coupled with a methodological approach such as the one presented here.

在全球典型的雨养系统中，普遍观察到作物产量的空间变化。人们普遍认为，除其他因素外，这种变化很可能与由于地形起伏的田地中发生的横向水流引起的水空间变化有关。然而，最常用的作物模拟模型并未完全考虑控制水空间分布的一些主要过程，例如侧向流。这给田间规模的产量模拟过程带来了不确定性，特别是在限水条件下。尽管预计横向水运动决定了作物产量的空间变化，但仍不清楚横向水流入（LIF）对起伏地形领域雨养产量空间变化的净贡献是什么。在这个意义上说，通过结合现场试验、模拟模型（HYDRUS-1D 和 AquaCrop）以及人工神经网络的使用，我们评估了 LIF 的发生和程度，以及它们对西班牙科尔多瓦小麦产量的 30 年影响。30 年间的季节性降水量从 212.8 到 759.5 毫米不等，累积 LIF 范围从 30 到 125 毫米。30 年来，季节性累积 LIF 除以季节性降水的比率从 10.7% 到 38.9% 不等。LIF 对雨养潜在产量空间变化的净贡献显示出相关但在年份之间高度不规则。尽管具有典型的地中海条件的年际变化，但 LIF 的发生导致模拟小麦产量从田地的上坡到下坡区域变化 + 16%。对 LIF 的净收益率响应，^-1，LIF 边际水生产力在最大响应年份达到 24.6 (± 13.2) kg GY ha ^-1 mm ^-1。鼓励决策者在对同一油田内不同的潜在产区进行管理调整时考虑水的空间变化。然而，预计这一过程将受益于季节性天气预报的进一步进展，该进展应与此处介绍的方法相结合。