Abstract The growth and yield of crops within a farm largely vary among fields. Farms are increasing in size by acquiring smaller land parcels from different farmers who have different management strategies. As a result, between-field variability increases and understanding such variability is a necessity for precision farming. New data analysis techniques are needed in this context, especially given the trend that more farms are collecting more data. Therefore, this study has the objective to provide a data analysis methodology to analyze within-year variability and identify year-independent factors that influence growth. As a second objective, we applied this novel methodology to a case study, where we analyzed potato growth data of four successive years of a farm in the south of the Netherlands. The methodology consists of three main steps: (1) describing growth using mixed models, (2) clustering and explaining growth by correlating the clusters to (a) yield, (b) other plant characteristics and (c) to defining, limiting and reducing variables, and (3) predicting growth by automatically selecting a regression model. By applying our method on the potato growth data, we obtained the following results. The main results of the work are: (1) the estimated growth curves of the stems, haulm and tubers explain the between-field variability in growth well very well ( R 2 of 0.85 , 0.74 and 0.89 , respectively), (2) clusters with a stem length between 110 and 130 cm have the highest average yield, (3) deeper groundwater level and sugar beet or grass as previously cultivated crop positively influence growth, and (4) N and K fertilization must be adjusted for optimal growth. Concluding, this study responds to the quest for new data-based methods for sustainable intensification, and is the first to explicitly analyze and explain differences in crop growth between fields in practice. In addition, clear management advice could be provided, showing the scientific and practical potential of our methodology.

中文翻译：

---

农场层面的数据科学：解释和预测马铃薯生长和产量的农场内变异

摘要 农场内作物的生长和产量因田地而异。通过从具有不同管理策略的不同农民那里获得较小的地块，农场的规模不断扩大。因此，田间变异性增加，了解这种变异性是精准农业的必要条件。在这种情况下需要新的数据分析技术，特别是考虑到越来越多的农场正在收集更多数据的趋势。因此，本研究的目的是提供一种数据分析方法来分析年内变异并确定影响增长的与年份无关的因素。作为第二个目标，我们将这种新颖的方法应用于案例研究，我们分析了荷兰南部一个农场连续四年的马铃薯生长数据。该方法包括三个主要步骤：(1) 使用混合模型描述生长，(2) 通过将聚类与 (a) 产量，(b) 其他植物特征和 (c) 定义、限制和减少变量相关联来聚类和解释生长，以及 (3) 预测生长通过自动选择回归模型。通过将我们的方法应用于马铃薯生长数据，我们获得了以下结果。工作的主要结果是：（1）茎、茎和块茎的估计生长曲线很好地解释了生长的田间变异性（R 2 分别为 0.85、0.74 和 0.89），（2）簇茎长在 110 到 130 厘米之间的平均产量最高，(3) 地下水位更深，以前种植的甜菜或草对生长有积极影响，(4) 必须调整施氮肥和钾肥以获得最佳生长。总结，这项研究响应了对可持续集约化新数据方法的探索，并且是第一个明确分析和解释实践中田间作物生长差异的研究。此外，可以提供明确的管理建议，展示我们方法的科学和实践潜力。