This study presents an investigation on the correlations between soiling losses and environmental parameters at 41 locations in the United States, with the aim of analyzing the possibility of predicting soiling losses at a site even when soiling data are not available. The results of this work, which considers the largest pool of soiling data points systematically investigated so far, confirm that a single‐variable regression based on particulate matter concentration returns the best correlations with soiling, with adjusted coefficients of determination up to 70%, corresponding to RMSE as low as 0.9%. Among the various particulate matter datasets investigated, a gridded Environment Protection Agency dataset is for the first time found to return correlations similar to those obtained by interpolating particulate matter monitoring station data. We discuss in detail the different interpolation techniques used to process the particulate matter concentrations because they can greatly impact the correlations. Specifically, the correlation coefficients between soiling and particulate matter range between 70% and less than 20%, depending on the interpolation methods and monitoring distance. Spatial interpolation methods based on inverse distance weighting are found to return better correlations than a nearest neighbor or a simple average approach, especially when large distances are considered. Similarly, the effects of different rain thresholds used to calculate the length of the dry periods are examined. An enhanced two‐variable regression is found to achieve higher‐quality correlations, with adjusted R² of 90% (RMSE = 0.55%), also suggesting that high and low soiling locations might be differentiated depending on fixed particulate matter or rainfall thresholds.

中文翻译：

---

使用环境参数预测光伏污染损失：一项更新

这项研究对美国41个地点的污染损失与环境参数之间的相关性进行了调查，目的是分析即使没有污染数据也可以预测某个地点的污染损失的可能性。这项工作的结果考虑了迄今为止系统研究的最大污染数据点，证实了基于颗粒物浓度的单变量回归与污染具有最佳相关性，调整后的测定系数高达70％，对应RMSE低至0.9％。在所研究的各种颗粒物数据集中，首次发现了网格化的环境保护局数据集，其返回的相关性类似于通过对颗粒物监测站数据进行插值获得的相关性。我们将详细讨论用于处理颗粒物浓度的不同插值技术，因为它们会极大地影响相关性。具体地，取决于插值方法和监测距离，污染与颗粒物之间的相关系数在70％至小于20％的范围内。发现基于逆距离权重的空间插值方法比最近邻或简单的平均方法能返回更好的相关性，尤其是在考虑较大距离的情况下。同样，检查了用于计算干旱期长度的不同降雨阈值的影响。调整后的增强型二变量回归可实现更高质量的相关性 取决于插值方法和监测距离，污物与颗粒物之间的相关系数范围在70％至小于20％之间。发现基于逆距离加权的空间插值方法比最近邻或简单的平均方法能返回更好的相关性，尤其是在考虑到较大距离的情况下。同样，检查了用于计算干旱期长度的不同降雨阈值的影响。调整后的增强型二变量回归可实现更高质量的相关性 取决于插值方法和监测距离，污物与颗粒物之间的相关系数范围在70％至小于20％之间。发现基于逆距离权重的空间插值方法比最近邻或简单的平均方法能返回更好的相关性，尤其是在考虑较大距离的情况下。同样，检查了用于计算干旱期长度的不同降雨阈值的影响。调整后的增强型二变量回归可实现更高质量的相关性 发现基于逆距离权重的空间插值方法比最近邻或简单的平均方法能返回更好的相关性，尤其是在考虑较大距离的情况下。同样，检查了用于计算干旱期长度的不同降雨阈值的影响。调整后的增强型二变量回归可实现更高质量的相关性 发现基于逆距离权重的空间插值方法比最近邻或简单的平均方法能返回更好的相关性，尤其是在考虑较大距离的情况下。同样，检查了用于计算干旱期长度的不同降雨阈值的影响。调整后的增强型二变量回归可实现更高质量的相关性R ²为90％（RMSE = 0.55％），也表明根据固定的颗粒物或降雨阈值，高低污染位置可能有所不同。