数据与计算方式
这一页说明本站用了哪些数据、降雪量与雪质分数怎么算出来的，以及这些数字到哪里为止还可信。每一个数字都是机械产生的，用同一套方法套用在所有地方。
測量的范圍
全球 3,069 座滑雪场的冬季降雪量。其中 397 座在日本，而日本的每一座，我们不只量下多少，也量下的是什么——下雪时的气温、下雪的频率、改下雨的频率，以及风。
日本环境省指定的 79 处国民保养温泉地，也用完全相同的方法測量。温泉地的清单到处都有；用和滑雪场同一把尺量出来的清单，没有別的地方有。
重点不是哪一个地方被仔细研究过，而是每一个地方都用同样的方式量过。 在这里，能不能互相比较，比单一数值的精确度更要紧。
什么算是「一座滑雪场」，各地不同。 白马山谷（Hakuba Valley）的十座分开计算，因为即使有共通券，每一座也都自己卖一日券。新雪谷联合（比罗夫、Village、Annupuri、HANAZONO）算成一个地方，取代表点測量。志贺高原算成九个独立经营的区域（例如烧额山、橫手山），再加上一笔涵盖蓮池周边较小雪道的项目。这是我们还没能把同一条规則套用到所有地方的一处，发现一处就改一处。
计算方式
降雪量是十二月到三月逐日降雪量的合计，再取最近十二个完整雪季的平均。南半球的雪场以六月到九月计算。我们从三季擴大到十二季，是因为单独一个大雪的冬天，会把一座雪场抬到远高于它长期平均该在的位置。
数据来源是 Open-Meteo 以 ERA5 再分析为基础的历史数据——一套计算涵盖全世界，这也是跨国比较之所以可能的前提。
雪质由四件事構成：下雪时的气温、单日超过 5 厘米的频率、下雨的频率，以及下雪日的风。下多少雪刻意不列入——列進去的話，雪质排行就只是降雪量排行的改写。
权重与门檻是我们的判断，不是測量结果。換个人会选得不一样，顺序也会跟著变。这份排行能承载的，是每一个地方都走过同一套计算。
这些数字该怎么读
单一雪场的「平年」是平均值，都道府县或国家的是中位数。 一座雪场的平年降雪量是它各季的平均。都道府县或国家則取其中雪场的中间值，这样一个极端的地方就拉不動整体。
比例和絕对值可能指向相反的方向。 北海道从一月到三月流失的雪比全国少，但三月的降雪量卻低于全国中位数。新潟剛好相反：掉得最凶，三月的雪卻仍是全国最多的之一。答案取决于看哪一个时，我们两个都給。
请读名次和比较，不要读厘米数。 推估值低于实际观測，所以这些数字不能拿去和雪场公布的积雪深度或別的网站并列。它们的用途，是比较用同一套方法量出来的地方。
与观測值的比对
降雪量来自再分析数据——那是推估，不是实际下了多少的纪录。我们拿它和日本气象厅 63 个观測点比对，看差多少。
结果大约是实际观測值的六成（平均 ×0.60）。它偏低，而且各地偏低的程度不一样：有些只读到实际值的三分之一，有些几乎吻合。
我们试过用固定倍率整体缩放来修正，几乎没有帮助，因为误差因地而异。既然确定修不好，我们就不修正数字，而是把这件事说清楚。
唯一的例外是日本页面上的地区地圖。 比对附近有日本气象厅观測点的 156 座滑雪场后发现，误差会依地区聚集：北海道约是观測值的 52%，北陆约 88%。推估是用固定比例把降水換算成雪，所以在雪更干的地方会低估得更多。单一地点的误差修不掉，但整个地区的中位数可以，所以只有那張地圖把这个差额加回去。雪场页、都道府县页与各排行，用的都是未修正的推估值。
比对用的观測点是机械选出来的：距滑雪场 15 公里以内，每个都道府县最多四个，由近到远。这条规則是为了让人能查核我们没有挑对自己有利的观測点。
我们发现并订正过的错误
本站的数字，是靠找出错误并订正而建立起来的。每次修正，我们都把「哪里错了、怎么修的」记在这里。结论因此改变的文章已经重写，并附上说明。以下纪录维持英文原文，以免改写了历史本身。
August 2026Switching to a finer weather model would have made things worse
A reader told us some resorts were ranked implausibly. We considered moving to the Japan Meteorological Agency's finer 5 km model, but first checked both against measurements at nine weather stations. The current model came to 73% of what was observed on average; the 5 km model came to 42%, and fell below half at every station. Switching would have made the figures worse, so we kept the current model. The reader was right, though: the error differs from place to place, and that is what was upsetting the ranks. The verification above began with this.
September 2026The snow-quality ranking mixed in places that had not been fully measured
Of the 454 places scored for snow quality, 81 did not have all three seasons of data — the download had hit a limit and stopped partway. Ranking places measured over different numbers of seasons in one table distorts the order, so every place was re-measured over three seasons. 393 places moved, and the tables and text of 12 articles were corrected one by one (Tengendai went from 3rd to 9th, and its headline changed with it). Any place missing a season is now always listed by name before anything is published.
September 2026Month-by-month patterns were built from three seasons
For a ranking of season totals, three seasons are enough: every place shares the same three winters, so their quirks fall on all places alike and the order barely moves. The shape of a season is different. A question like "is December or January snowier?" can flip for a whole region because of one unusual winter. Re-measured over 11–12 seasons, March snowfall at resorts above 1,200 m was 60% of January, not 91%, and "Kusatsu and Nikko Yumoto get the most snow in March" was wrong — both peak in January. One article was rewritten from its conclusion down, and two comparing Japan with the Alps and North America were withdrawn. Those two were later re-measured over the same number of seasons; neither conclusion held, so one was rewritten with its new finding and the other stays withdrawn.
September 2026Year-to-year variability was compared over three winters
The three winters we were using (2023/24 to 2025/26) did not include 2015/16 or 2019/20, the worst winters at most resorts. Over twelve winters, the swing at Kiroro — which we had called the most stable — is 70%, not 9%, almost the same as Hakuba's 74%. Iwate went from the most stable prefecture to one of the least. Bad winters arrive across the whole country at once: 240 of 377 resorts had their worst winter in 2015/16. Picking a stable region to dodge a bad winter is not something you can actually do, so the article was rewritten around how much snow is left in a bad winter. Eight other affected articles carry a correction note and are being rewritten in turn.
September 2026World ranks quoted in articles had gone stale
Each time resorts are added, world ranks shift slightly. We did check the ranks quoted in articles against the data, but it turned out the check had been passing without comparing against anything. Once it was fixed, it found 16 outdated ranks, which have been corrected. A rank in an article that no longer matches the data now gets flagged.
这些数字说不出来的事
我们不做预报。 这里公开的是过去的观測与计算，加上气象数据供应商提供的当日资讯。我们不製作自己的预报。
雪场公布的积雪深度是完全不同的东西。 每一家在自己的地点、用自己的方式測量，所以那些数字彼此之间并不可比。我们的数字之所以可比，只是因为每一个地点都走过同一套流程。
联絡
关于计算方式的问题、数据使用的洽询、刊载内容的订正要求，请由联絡页面告诉我们。