季节预测

Encyclopedie environnement - prevision saisonniere - couverture

 

  本文所讲述了如何用数值天气预报的数值模式提供未来几个月的天气信息,尽管该数值模式有其局限性。它展示了科学家们是如何进行合作,从而使这一工作顺利进行的。尽管这项活动在某些方面已经有一些应用,但它仍然算是一项研究活动。

1. 为什么要做整个季节的天气预测。

  天气预报是基于计算机对大气-海洋-冰冻圈-陆面观测的流体力学方程的解析(见天气预报介绍)。因此,基于第D天的观测,我们可以部分精确地描述D+1天的状态。再通过一些简单的论证,我们可以通过同样的步骤推断出D+2天的情况,然后得到D+3天、D+4天……我们有理由相信,如果不考虑计算资源,预测将不会受到限制。但是实际上,人们很快发现随着时间的推移天气预报的准确度下降的很快。几天后(2016年的情况是大约一周),预报的天气情况与前几年同一个月随机出现的情况相比,并不更接近实际情况。Lorenz(1963)[1]表明,即使天气模式是完美的,描述初始条件时的一个非常小的误差也会使超过10-15天的天气预报变得不准确。这就是众所周知的蝴蝶效应

  因此很难在季节层面做出确定性预测。然而,自然界中有一些现象在几个月内的演化并不混乱(见第3节),它们会影响大气的活动。并且,在许多人类、农业、工业或旅游活动中,确实需要进行长期预测。如果我们有模拟这些缓慢现象演变的可靠数值模式,能够大体上重现这些现象对感知时间(即温度、风、雨等)的影响,我们就有可能提供一些未来几个月将发生什么的信息。季节预测月度预测的一个区别是月度预测的目的是描述一个粗略的年表(例如,一月内的较热的那部分)。而季节预测摒弃了任何按时间顺序的方法,是用纯粹的统计术语来描述一个季节(例如,明年冬天出现低温的可能性很大)。在2016年,月度预测通常能预测到下两个月;季节预测则通常能预测到之后7个月。在下文中,我们将描述这些季节预测是如何产生的,以及它们的作用。

2. 预测是如何做出的呢?

  季节预测使用数值工具,是天气预报的延续。然而,从第一次短期天气预报到第一次季节预测,间隔了30多年时间[2]。其实,这个过程中有三个绕不开的技术难关一一被突破:

  • 20世纪80年代初:出现了覆盖全球的大气模式和逐日观测,
  • 20世纪90年代初:在考虑到几十年平均值情况下,20世纪90年代初出现的海洋-大气耦合模式的全球模拟结果与观测结果相当相似。
  • 21世纪初:建立三维Argo全球海洋观测网络[3]

  我们有时会讨论24小时、10天、一个月、6个月、10年甚至到本世纪末的无缝隙预测。这凸显了科学技术共享的必要性。因此,法国气象局用于季节预测的模式也被IPCC(政府间气候变化专门委员会)采用进行气候情景模拟,其大气和陆面分量来自法国气象局的短期天气预测模式。

  从实用的角度来考虑,季节预测和短期预测的实施方法上存在差异

  • 海洋是气候系统中慢变的部分,其准确模拟和初始化在季节预测中有至关重要的作用。其只有在两三天之后应用才是合理的。
  • 我们做出的季节预测不是确定性的,而是统计性的。因此,有必要进行至少50次预测,以获得可靠的统计估计。我们讨论的是集成预报(参见集成预报)。
  • 在短期内,不确定性的主要来源与初始状态有关。在季节尺度上,则需要考虑模拟气候和实际气候之间的差异。

3. 预测是符合事实的!

  评估季节预测的困难长期以来阻碍了人们对科学方法有效性的共识。

  在短期预测中,可以在几个月后评估预测系统的成功率和失败率。在季节预测中,要做出可靠的判断需要几十年的时间。因此,我们通过重新预测来回顾过去。重新预测是指在多年以后做一个回顾性的预测活动,但不会使用任何预测初始时刻之后的信息来“作弊”。这些所谓的重新预测会包括长期(出于统计原因)和近期(出于观测系统的均一化原因)。在业务应用中重新预测始于1993年(卫星测高观测开始),而研究工作中重新预测始于1979年(卫星全球海洋温度观测开始)。制作一套涵盖过去30年或40年的重新预测需要6个月到1年的时间。这种等待时间使得研究人员不受偏见的影响,即根据最高的成功率从众多系统中选择最佳系统。成功概率通常由一个叫做预测评分的指数来衡量。预测评分通常是介于0和1之间的数字,例如系统地预测某个月份和区域的气候,1代表一个完美的预测。而当预测结果系统性地与观测结果相反时,预测评分可能会是一个负值。一个经常使用的评分是两个时间序列之间的相关系数。这个指数在-1和1之间,如果两个序列是独立的,则指数为0;如果一个序列可以通过线性代数关系得到另一个序列,则指数为1。其精确定义基于统计分布的二阶矩(Jolliffe和Stephenson,2012[4])。

  在我们所处的纬度地区,冬天温和或是寒冷的概念只是一个统计学意义上的事。在同一个冬天,有时温暖,有时寒冷。有时后者的数量更多或程度更剧烈。通过集合50次日报,可以重建预测概率。估计给定冬季观测的概率要困难得多,因为只有90个值可用,而且这些值不是独立的,因为从一天到下一天存在一定的持续性。因为它没有考虑到非确定性预测中的不确定性,所以用相关系数进行评估是不够的。同时,因为要预测的现象(寒潮的概率)只能粗略估计,所以概率性评估更合理。概率性评分指数有许多(JolliffeandStephenson,2012[4])。尽管研究人员对它们感兴趣,也很难在两个模型之间进行比较,因为单个指数并不一致。

环境百科全书-季节预测-预测的月温度和观测月温度之间的相关系数
图1. .赤道中太平洋预测的月温度和观测月温度之间的相关系数,作为成熟度(月)的函数。预测时间依次为11月1日(a)、2月1日(b)、5月1日(c)和8月1日(d)。红色曲线代表法国气象局模式的分数,黑色曲线代表初始条件的持续性。该系数的计算时间为1979年至2012年。

  为了获得更可靠的评估,我们将目光投向变化较为平缓的现象,这些现象在一个季节的过程中保持恒定,同时具有较强的年际变化。目前有三方面现象可以观测和模拟,并可以得到令人满意的分数。

  20世纪80年代以来的一个众所周知的现象(Shukla1981)[5],赤道太平洋的表面温度:如图1所示ENSO现象(厄尔尼诺-南方涛动)具有长达7个月的显著可预测性(红色曲线);在冬天和秋天,这种现象非常持久(黑色曲线),但该模式也表现得很好甚至更好。

  最近发现的现象:北极冰盖的范围(Chevallier et al. 2013) [6]

环境百科全书-季节预测-预测的欧洲季节温度和观测季节温度之间的相关系数
图2. 法国气象局模式预测的欧洲季节温度和观测季节温度之间的相关系数。预测日期为11月1日(a)、2月1日(b)、5月1日(c)和8月1日(d),预测结果为第2至第4个月的平均值。全球变暖的趋势已经被减去。该系数的计算时间为1979年至2012年。

  在实践中不太有用,但在科学上很有趣:赤道平流层的风(Boer and Hamilton, 2008) [7]

  中纬度地区的可预测性较低,但依然是有必要的。这类预测质量的常用衡量标准是预测的季节平均值和观测平均值之间的相关系数(如上所述),其优点是被长期使用,且普遍适用,即使它在30年序列中不是非常稳健,并且没有考虑预测集合的分散性。图2显示了欧洲各地四季温度预测(DJF、MAM、JJA、SON,其中每个大写字母是一年中一个月的首字母)的系数值。因为全球变暖趋势会人为地夸大分数,因此该趋势已被减去。

环境百科全书-季节预测-观测到的季节平均温度与预测前一个月的月温度之间的相关系数
图3. 观测到的季节平均温度与欧洲预测(持续性预测)前一个月的月温度之间的相关系数。季节平均值分别为冬季(a)、春季(b)、夏季(c)和秋季(d)。全球变暖的趋势已经被减去。该系数的计算时间为1979年至2012年。

  小于0.2的系数被认为不能提供可用信息。可以看出,可预测性因地区和季节而异。在冬季(图2a),海洋表面温度可预测性较高。在夏季(图2c),该大陆的东南部可预测性较高。图3显示了相同的分数,但针对初始情况的持续性进行了预测。可以看出,在平季,这种低成本的计算方法通常比模式更好。

4. 预测的实际执行情况

  第2节指出,模式的系统性缺陷包括误差(可以事后纠正)和不确定性(必须进行估计)。这意味着不仅要进行预测,还要进行重新预测(该术语的定义见第3节)。在研究工作中,我们只进行重新预测。

  业务性预测包括将预测情况的平均值与称为参考气候的多年平均值进行比较。为此,我们对过去几年进行了一系列重新评估。这一工作还用于估计预测的分数。重新预测和预测结果之间必须有极高的相似性。这意味着这两种预测仅在初始条件上有所不同。在每月进行的预测和回顾性预测之间,不能对模式进行任何改动。考虑到重新预测所需的时间(六个月到一年),预测中心每两年或三年才更换一次版本。为了估计与模式缺陷相关的不确定性,最常用的方法是多模式方法。实际上,每个模式都是建立在实施它的中心的选择和假设的基础上的。有时两个模式的预测会不一致。例如,欧洲(Eurosip)、北美(NMME)和亚洲(APCC)的机构会同时使用多个模式的预测。他们将这些预测以地图或公告的形式免费提供。在美国和欧洲(欧盟哥白尼气候变化服务项目),每个月都可以免费获取接下来六个月的数据。

5. 季节预测的目的是什么?

  季节预测基本上是通过网络以地图或公告的形式告知所有潜在用户(欧洲中期预测中心[8]法国气象局[9])。赤道太平洋发生热事件(厄尔尼诺)对于干旱、渔业资源、洪水风险等方面的影响有充分的说明[10]

  除此之外,还有为特定用途开发的季节预测应用。尤其是在预测结果较可靠的热带地区,预报员和用户之间通常通过统计调整进行合作。这种方法需要在长时间内重新预测,并且不经常更改版本。例如,塞内加尔河上大坝的管理包括季节预测部分(Bader等人,2006[11])。在南美洲,Eurobrisa consortium正在进行一些统计调整[12]

  在欧洲,用户往往因为预测分数低而感到气馁(Bruno Soares和Desai,2016[13])。在法国,对小麦产量(Canal,2014[14])或河流流量和土壤湿度(Singla等人,2012[15])的研究已经展开。最后一个应用也已经开始,欧洲正在开发其他应用领域的原型,正在进行的项目有[16]:冬季环境气候对英国交通的影响,西班牙可再生能源生产,瑞典水电管理。

6. 季节预测的未来发展?

  由于季节预测花费巨大,同时受到物理(蝴蝶效应)和技术(数字计算机资源)的限制,其结果乍一看往往令人失望。然而,与天气预报系统相比,它的额外成本非常低,如果你接受风险管理,它就能够提供有效的服务。如果我们将2015年的最新水平与2005年或1995年的水平进行比较,预测质量的进步是显而易见的,但从一年到下一年的进步没有那么明显,有时会给人一种停滞不前的错觉。

  最可预测的进展将来自模式水平和垂直分辨率的提高,以及热源和湿度源表示的复杂性(称为物理参数化)。事实上,这种努力与改进天气预报和气候模式的努力是共同的。但也它有更具体的研究方向,例如在方程中引入扰动来模拟模式误差(Batté和DoblasReyes,2015[17]),或者和预测的统计后处理一样,考虑预测期间的模式误差。缓慢改变参数,如积雪或土壤湿度,在系统的季节尺度存储记录中发挥作用。在预测和重新预测中优化初始化进程也是一个潜在的改进方面。

 


参考资料及说明

[1] Lorenz (1963). Deterministic nonperiodic flow. Journal of Atmospheric Sciences,20,130-141.

[2] Palmer, T.N. and Anderson, D.L.T. (1994). The prospects for seasonal forecasting – a review paper”. Quarterly Journal of theRoyal Meteorological Society, 120, 755-793.

[3] ARGO. Part of the integrated observation strategy.

[4] Jolliffe IT和Stephenson DB(2012)预测验证。第二版,奇切斯特(英国),威利·布莱克威尔

[5] Shukla J. (1981) Dynamical predictability of monthly means. Journal of Atmospheric Sciences, 38″,2547-2572.

[6] Chevallier M., Salas y Mélia D., Voldoire A., Déqué M. and Garric G. (2013). Seasonal Forecasts of the Pan-Arctic Sea IceExtent Using a GCM-Based Seasonal Prediction System. Journal of Climate, 26, 6092-6104.

[7] Boer G.J.和Hamilton K.(2008)。QBO对温带预测能力的影响。气候动力学,31987-1000。

[8] ECMWF, Forecast Charts

[9] Météo-France, Seasonal forecast(公共网站,但需要身份认证)。

[10] Météo-France, Understanding: El Nino and La Nina

[11] Bader JC, Piedelievre JP, and Lamagat JP (2006). Seasonal forecast of the Senegal River flood volume: use of the results ofthe ARPEGE Climat model. Hydrological Science Journal, 51:3, 406-417, DOI: 10.1623/hysj.51.3.406

[12] Eurobrisa; A EURO-BRazilian Initiative for improving South American seasonal forecasts.

[13]  Bruno Soares M and Dessai, S. (2016). Barriers and enablers to the use of seasonal climate forecasts amongst organisationsin Europe. Climatic Change, 1-2, 89-103. DOI: 10.1007/s10584

[14] Channel N, (2014). Application of seasonal forecasting to agriculture: assessment at the scale of France. Thesis from the University Paul Sabatier. Toulouse

[15] inla S, Ceron JP, Martin E, Regimbeau F, Déqué M, Habets F and Vidal JP (2012). Predictability of soil moisture and river ows over France for the spring season. Hydrology and Earth System Sciences, 16, 201-216.

[16] Euporias. European Provision Of Regional Impacts Assessments on Seasonal and Decadal Timescales.

[17] Batté L. and Doblas-Reyes F.J. (2015) Stochastic atmospheric perturbations in the EC-Earth3 earth system model: impact of SPPT on seasonal forecast quality. Climate Dynamics, 45, 3419-3439.


环境百科全书由环境和能源百科全书协会出版 (www.a3e.fr),该协会与格勒诺布尔阿尔卑斯大学和格勒诺布尔INP有合同关系,并由法国科学院赞助。

引用这篇文章: DEQUE Michel (2024年3月11日), 季节预测, 环境百科全书,咨询于 2024年12月9日 [在线ISSN 2555-0950]网址: https://www.encyclopedie-environnement.org/zh/air-zh/seasonal-forecast/.

环境百科全书中的文章是根据知识共享BY-NC-SA许可条款提供的,该许可授权复制的条件是:引用来源,不作商业使用,共享相同的初始条件,并且在每次重复使用或分发时复制知识共享BY-NC-SA许可声明。

The seasonal forecast

Encyclopedie environnement - prevision saisonniere - couverture

This article shows how the numerical models used for numerical weather prediction can be used to provide information on the weather over the next few months, despite their limitations. It presents how scientists have organized themselves to give an operational status to this activity, which still remains a research activity, although it has already conquered some applications.

1. Why a season-wide forecast?

Weather forecasting is based on the resolution by a computer of fluid mechanics equations from observations of the atmosphere-ocean-cryosphere system-continental surfaces (see Introduction to weather forecasting). Thus, from the observation of the D-day, we can describe with some precision the state of the D+1-day. By a simple reasoning, we deduce that the state of the day D+2 is accessible by the same process, then the day D+3… We could believe that there is no limit to the possibility of producing forecasts, except for the calculation resource. In practical terms, it was quickly realized that the quality of the forecasts decreased very quickly with time: after a few days (about a week in 2016) a forecasted weather situation does not look any more like the actual situation than a situation taken at random in the same month in previous years. Lorenz (1963) [1] showed that even if the weather models were perfect, a very small error in describing the initial conditions made it impossible to forecast beyond an estimated limit of 10-15 days for the atmosphere. This notion has been popularized by the term butterfly effect.

It is therefore not possible to make deterministic forecasts at the season level. However, there are phenomena in nature whose evolution over a few months is not chaotic (see section 3) and which influence the behaviour of the atmosphere. And, moreover, there is a real need for long-term forecasts in many human, agricultural, industrial or tourist activities. If we have numerical models that faithfully simulate the evolution of these slow phenomena, and that are able to reproduce on average the effect of these phenomena on the perceived time (i.e. temperature, wind, rain, etc.), then we can hope to provide partial information on what will happen in the coming months.

A distinction is made between the monthly forecast, which aims to describe even a rough chronology (for example, a hot episode at the end of January), and the seasonal forecast, which abandons any chronological approach to describe a season in purely statistical terms (for example, a significant probability of having a cold episode next winter). In 2016 monthly forecasts are generally produced until month M+2; seasonal forecasts are generally produced until month M+7 (see section 4). In the following, we will describe how these seasonal forecasts are produced and what can be expected from them.

2. Yes, but how do we make these forecasts?

Seasonal forecasting is the heir to weather forecasting through the numerical tools it uses. However, it took more than thirty years between the first operational short-term forecasts and the first seasonal forecasts (Palmer and Anderson, 1994) [2]. Indeed, three particularly difficult technical steps had to be taken:

  • early 1980s: emergence of atmospheric models and daily observations covering the entire globe,
  • early 1990s: emergence of coupled ocean-atmosphere models whose behaviour on the globe is reasonably similar to observations, when considering an average over several decades,
  • early 2000s: establishment of a three-dimensional Argo global ocean observation network [3].

We sometimes talk about seamless forecasting for the production of forecasts at 24h, 10 days, one month, 6 months, 6 months, 10 years, or even climate scenarios for the end of the century. This expression underlines the need to share scientific and technological developments. Thus, the model used in Météo-France for seasonal forecasting is also the one used for the IPCC (International Panel on Climate Change) climate scenarios and its component for the atmosphere and continental surfaces is the model used by Météo-France for short-term weather forecasting.

For practical reasons, there are implementation differences between the seasonal forecast and the short-term forecast:

  • Accurate modelling and initialization of the ocean, a slow component of the system, is essential for seasonal maturity. It is only justified for a few applications after 2-3 days.
  • The forecast we are trying to make at the seasonal maturity is not of a deterministic nature but of a statistical nature. It is therefore necessary to make at least 50 forecasts to obtain robust statistical estimates. We talk about ensemble forecasting (see The Ensemble forecasting).
  • In the short term, the main source of uncertainty concerns the initial state. At the seasonal scale, it is also necessary to take into account the differences between the simulated climate and the actual climate.

3. A forecast is true!

The difficulty of assessing a seasonal forecast has long delayed the consensus on the validity of a scientific approach.

In short-term forecasting, the success and failure rate of a forecasting system can be assessed after a few months. In seasonal forecasting, it would take decades to make a reliable judgment. Therefore, we look to the past by producing what are called re-forecasts. A re-forecast is a forecast produced retrospectively, years later, but without cheating, i.e. by not using any information after the initial time of the forecast. These re-forecasts should cover both a long period (for statistical reasons) and a recent period (for reasons of homogeneity of the observation system). Operational applications start this period in 1993 (satellite altimetry observation begins), while research mode evaluations begin in 1979 (satellite global ocean temperature observation begins). Producing a set of re-forecasts covering the last 30 or 40 years takes from six months to one year. This waiting time protects the researcher from the bias of choosing the best system from among a large number, based on the highest success rate. This success rate is generally measured by an index called the forecast score. A score is a number generally between 0 and 1, corresponding to a trivial forecast, for example systematically predicting the climatology of the month and place, 1 corresponding to a perfect forecast. A forecast can have a negative score, for example when it systematically announces the opposite of what is observed. A very often used score is the correlation coefficient between two time series. This index between -1 and 1 is 0 if the series are independent, and 1 if one can move from one to the other by an increasing linear algebraic relationship. Its precise definition is based on the second order moments of the statistical distribution (Jolliffe and Stephenson, 2012 [4]).

The notion of mild or cold winter has only a statistical reality at our latitudes. During the same winter, there are a succession of hot and cold episodes. Sometimes the latter are more numerous or more intense. With sets of 50 daily forecast members, a predicted probability law can be reconstructed. Estimating a probability for observations in a given winter is much more difficult, as only 90 values are available, and these values are not independent because there is some persistence from one day to the next… The evaluation by a correlation coefficient is insufficient because it does not take into account the uncertainty inherent in a non-deterministic forecast. The probabilistic assessment is more delicate because the phenomenon to be predicted (the probability of a cold wave) can only be estimated roughly. Many probabilistic scores are available (Jolliffe and Stephenson, 2012 [4]). If they are of interest to the researcher, they are difficult to compare between models because there is no consensus on a single score.

Encyclopedie environnement - prevision saisonniere - Coefficient de correlation temperature prevue et temperature observee Pacifique
Figure 1. Correlation coefficient between expected monthly temperature and observed monthly temperature in the central equatorial Pacific as a function of maturity (months). The departure date is successively November 1 (a), February 1 (b), May 1 (c) and August 1 (d). The red curve represents the scores of the Météo-France model, the black curve represents the persistence of the initial conditions. This coefficient is calculated for the period 1979-2012.

For a more reliable assessment, we turn to slow-moving phenomena that maintain a constant characteristic over the course of a season, while offering a high degree of variability from one year to the next. There are three of them that are accessible to both observation and modelling, and for which satisfactory scores are now available.

  1. A well-known phenomenon since the 1980s (Shukla 1981) [5], the surface temperature of the equatorial Pacific Ocean: as shown in Figure 1, the ENSO phenomenon (El Niño Southern Oscilation) offers remarkable predictability up to 7 months (red curve); in winter and autumn, the phenomenon is very persistent (black curve), but the model does as well or better.
  2. Phenomenon discovered quite recently: the extent of Arctic ice pack (Chevallier et al., 2013) [6].
  3. Less useful in practice, but scientifically interesting: wind in the equatorial stratosphere (Boer and Hamilton, 2008) [7].
Encyclopedie environnement - prevision saisonniere - Coefficient de correlation temperature saisonniere prevue et temperature saisonniere observee Europe
Figure 2. Correlation coefficient between expected seasonal temperature and observed seasonal temperature over Europe for the Météo-France model forecasts. The departure date is November 1 (a), February 1 (b), May 1 (c) and August 1 (d), and the forecast is for the average of the 2nd to 4th month. The trend due to global warming has been subtracted. This coefficient is calculated for the period 1979-2012.

Nevertheless, it is necessary to estimate predictability at mid-latitudes, even if it is low. The usual measure of the quality of this type of forecast is the correlation coefficient (mentioned above) between the expected seasonal average and the observed average, which has the advantage of being used for a long time and being universally used, even if it is not very robust over 30-year series and does not take into account the dispersion of the forecast set.

Figure 2 shows the values of this coefficient for the four-season temperature forecasts (DJF, MAM, JJA, SON, where each capital letter is the initial of a month of the year) for the temperature over Europe. The trend due to global warming has been subtracted because it artificially inflates the scores.

Encyclopedie environnement - prevision saisonniere - Coefficient de correlation entre temperature saisonniere observee et temperature mensuelle du mois precedant la prevision sur l'Europe
Figure 3. Correlation coefficient between observed seasonal temperature and monthly temperature in the month preceding the European forecast (persistence forecast). The seasonal averages are for winter (a) spring (b) summer (c) and autumn (d) respectively. The trend due to global warming has been subtracted. This coefficient is calculated for the period 1979-2012.

A coefficient of less than 0.2 is considered to provide no usable information. It can be seen that predictability varies from region to region and from season to season. In winter (Figure 2a) the oceanic façade is favoured. In summer (Figure 2c), the Southeast of the continent is the preferred region. Figure 3 shows the same scores, but for a forecast by persistence of the initial situation. It can be seen that in the shoulder seasons this low-cost method of calculation generally does better than the model. A lesson in modesty!

4. The operational implementation of the forecasts

In Section 2, it was written that systematic model defects were a source of both error (which can be corrected a posteriori) and uncertainty (which must be estimated). This implies producing not only forecasts, but also re-forecasts (this term has been defined in section 3). In a research activity, we only carry out re-forecasts.

Operational forecasts consist of comparing an average of expected situations with a multi-year average called the reference climate. For this purpose, we have a set of re-estimates of past years. This game is also used to estimate the expected scores for the forecast. It is essential to have perfect homogeneity between re-forecasts and forecasting. This means that the two productions differ only in the initial condition. There is no question of making any changes in the model between the re-forecasts, made once and for all, and the forecast made each month. Given the time required to carry out a re-forecast (six months to one year), the forecast centres only change versions every two or three years.

To estimate the uncertainty associated with model imperfections, the most commonly used method is the multi-model approach. Indeed, each model is built on the basis of choices and hypotheses that are specific to the centre that implements it. Sometimes the forecasts of two models disagree. For example, there are consortia in Europe (Eurosip), North America (NMME) and Asia (APCC) that combine forecasts from several models. These forecasts are available free of charge in the form of maps or bulletins. In the USA and soon in Europe (Copernicus Climate Change Service programme of the European Union) digital data are freely accessible every month for the next six months.

5. What is the purpose of seasonal forecasts?

Seasonal forecasts are essentially made to be distributed to all potential users in the form of maps or bulletins via the web (European Medium-Term Forecast Centre [8] Météo-France [9]). The consequences of a hot event in the equatorial Pacific (El Niño) are sufficiently documented [10] for a prediction of its occurrence to be interpreted in terms of droughts, fisheries resources, flood risk..

There are also seasonal forecast applications developed for specific use. It is especially in tropical regions, where scores are higher, that there is collaboration between forecasters and users, usually through statistical adaptation. This type of approach requires re-forecasts over long periods of time and infrequent version changes. For example, the management of a dam on the Senegal River includes a seasonal forecast component (Bader et al., 2006 [11]). In South America, a number of statistical adaptations are being made by the Eurobrisa consortium [12].

In Europe, users are often discouraged by low scores (Bruno-Soares and Dessai, 2016 [13]). In France, studies have been carried out on wheat yield (Canal, 2014 [14]) or on river flow and soil moisture (Singla et al., 2012 [15]). This last application is now operational. Prototypes are being developed in Europe in other fields of application, through the Euporias project [16] : impact of winter conditions on transport in the United Kingdom, renewable energy production in Spain, hydropower management in Sweden.

6. What future for seasonal forecasts?

With an ambitious objective and strongly limited by the constraints of physics (butterfly effect) and technology (digital computer resources), seasonal forecasting often disappoints at first glance. However, its additional cost compared to a weather forecasting system is very low and the services it can provide are undeniable if you accept risk management. Progress on the quality of forecasts is visible if we compare the state of the art in 2015 with that of 2005 or 1995, but very little progress is made from one year to the next, which sometimes gives a false idea of stagnation.

The most predictable advances will come from the increase in horizontal and vertical resolution of the models and the complexity of the representation of heat and humidity sources (known as physical parameterizations). Indeed, this effort is common to that on improving weather forecasting and climate modelling. But there are also more specific research axes such as the introduction of perturbations in the equations to simulate model errors (Batté and Doblas-Reyes, 2015 [17]), or as the statistical post-processing of forecasts taking into account model errors over a learning period. Slowly changing parameters such as snow cover or soil moisture play a role in the system’s seasonal scale memory. The refinement of their initialization in forecasts and re-forecasts is also a potential source of improvement.

 


References and notes

[1] Lorenz (1963). Deterministic nonperiodic flow. Journal of Atmospheric Sciences,20, 130-141.

[2] Palmer, T.N. and Anderson, D.L.T. (1994). The prospects for seasonal forecasting – a review paper”. Quarterly Journal of the Royal Meteorological Society, 120, 755-793.

[3] ARGO. Part of the integrated observation strategy.

[4] Jolliffe IT and Stephenson DB (2012) Forecast Verification. 2nd edition, Chichester (UK), Wiley-Blackwell

[5] Shukla J. (1981) Dynamical predictability of monthly means. Journal of Atmospheric Sciences, 38″,2547-2572.

[6] Chevallier M., Salas y Mélia D., Voldoire A., Déqué M. and Garric G. (2013). Seasonal Forecasts of the Pan-Arctic Sea Ice Extent Using a GCM-Based Seasonal Prediction System. Journal of Climate, 26, 6092-6104.

[7] Boer G.J. and Hamilton K. (2008). QBO influence on extratropical predictive skill. Climate Dynamics, 31, 987-1000.

[8] ECMWF, Forecast Charts

[9] Météo-France, Seasonal forecast (public site, but identification required),

[10] Météo-France, Understanding: El Nino and La Nina

[11] Bader JC, Piedelievre JP, and Lamagat JP (2006). Seasonal forecast of the Senegal River flood volume: use of the results of the ARPEGE Climat model. Hydrological Science Journal, 51:3, 406-417, DOI: 10.1623/hysj.51.3.406

[12] Eurobrisa; A EURO-BRazilian Initiative for improving South American seasonal forecasts.

[13] Bruno Soares M and Dessai, S. (2016). Barriers and enablers to the use of seasonal climate forecasts amongst organisations in Europe. Climatic Change, 1-2, 89-103. DOI: 10.1007/s10584

[14] Channel N, (2014). Application of seasonal forecasting to agriculture: assessment at the scale of France. Thesis from the University Paul Sabatier. Toulouse

[15] inla S, Ceron JP, Martin E, Regimbeau F, Déqué M, Habets F and Vidal JP (2012). Predictability of soil moisture and river ows over France for the spring season. Hydrology and Earth System Sciences, 16, 201-216.

[16] Euporias. European Provision Of Regional Impacts Assessments on Seasonal and Decadal Timescales.

[17] Batté L. and Doblas-Reyes F.J. (2015) Stochastic atmospheric perturbations in the EC-Earth3 earth system model: impact of SPPT on seasonal forecast quality. Climate Dynamics, 45, 3419-3439.

Registering

Registering

Registering


环境百科全书由环境和能源百科全书协会出版 (www.a3e.fr),该协会与格勒诺布尔阿尔卑斯大学和格勒诺布尔INP有合同关系,并由法国科学院赞助。

引用这篇文章: DEQUE Michel (2024年3月6日), The seasonal forecast, 环境百科全书,咨询于 2024年12月9日 [在线ISSN 2555-0950]网址: https://www.encyclopedie-environnement.org/en/air-en/seasonal-forecast/.

环境百科全书中的文章是根据知识共享BY-NC-SA许可条款提供的,该许可授权复制的条件是:引用来源,不作商业使用,共享相同的初始条件,并且在每次重复使用或分发时复制知识共享BY-NC-SA许可声明。