History of Regional Climate Modelling – How it started?

The first general circulation model was developed by Norman Phillips in 1956. After that within a decade or two around late 1970s there was reasonable community of climate modellers around the world¹ . But these early climate models were of very coarse resolutions (~500-1000km ) and they were used, in general, for producing planetary scale climate projections. Then in late 1980s there was a special situation that needed climate projections at higher resolutions than the resolution of then climate models.

The story goes like this – US Congress selected the Yucca Mountain as a long term nuclear waste deposite site to bury nuclear waste hundreads of meteres below ground. It is one of the very dry regions of US and hence was supposed to have lower risk of ground water contamination. Because nuclear waste lasts for thousands of years, people wanted to make sure there is no possibility of changes in ground water table over the site in coming centuries as well. Ground water table would possibly increase either by onsite considerable precipitation increase or increase in snowpacks over nearby mountains. The dryness of region was attributed to its location – leeside of the Sierra Nevada mountain range. However, the climate models of the day were so coarse that they would represent the site on upslope of the Rocky mountains and hence no proper precipitation projections. As a solution, Dickinson et al. (1989) proposed to use limited area modelling system to get higher resolution projections over the project site². Limited area modelling was already in use for weather forecasting but was never used for climate projections. They used PSU/NCAR Mesocale Model version 4 (precursor of WRF) nested within NCAR Community Climate Model (precursor of CESM) to get first dynamical downscaling of climate projections.

Its been over thirty years since first regional dynamical downscaling was produced, a lots of progress has been made in climate models as well as limited area models. Given that resolution of climate models has increased many folds, do we still need regional downscaling? Answer is yes – Ever increasing complexity of Earth Syatem Models (ESMs) and need to get more ensembles (to reduce uncertainity) make it impossible to run them at very high resolutions so regional climate downscaling will remain indispensible tool in coming years. Convection permiting regional climate models, regional ESMs and attribution studies using regional climate models are few of the future directions in progress of regional dynamical downscaling along with standards use to get regional reliable high resolution climate projections^3,4.

References –

1. Gates, W. Lawrence. “Report of the JOC Study Conference on Climate Models: Performance, Intercomparison and Sensitivity Studies, Volume I.” Global Atmospheric Research Programme (1979).
2. Dickinson, Robert E., et al. “A regional climate model for the western United States.” Climatic change 15 (1989): 383-422.
3. Giorgi, Filippo. “Thirty years of regional climate modeling: where are we and where are we going next?.” Journal of Geophysical Research: Atmospheres 124.11 (2019): 5696-5723.
4. Gutowski, William J., et al. “The ongoing need for high-resolution regional climate models: Process understanding and stakeholder information.” Bulletin of the American Meteorological Society 101.5 (2020): E664-E683.

Regional Climate Projections – Need

The complex interactions and feedbacks between the components of earth system operate over a wide spectrum of spatial and temporal scales. Anthropognic climate change is hemispheric to global scale warming due to increase in greenhouse gase concentrations, mainly CO₂ . However regional manifestations of global climate change are not uniform and not simulated realiably by coarse resolution global climate models/Earth system Models (ESMs). Therefore, it is important to have physically robust, realiable high-resolution regional climate projections to guide policymakers and other stakeholders.

ESMs reliably simulate the large scale phenomenon, however fine-scale features such as mountains, coastlines, lakes, land use and other surface heterogenity are under-resolved by ESMs. Regional climates are outcomes of complex interactions of local physical processes, governed by fine-scale features listed above, with the large-scale phenomena. Many regional climate processes are not simulated by ESMs therefore for reliable regional climate projections dynamical downscaling is employed. Dynamical downscaling refers to obtaining regional projection from limited area high resolution numerical models, called as Regional Climate Models (RCM), forced by ESM outputs at boundaries.

Key regional features that are unresolved in ESMs –

1. Regional Monsoons
2. Mesoscale Convective Systems
3. Tornadoes
4. Atmospheric Rivers
5. Hurricanes
6. Tropical Cyclones
7. Dust Storms
8. Orographic precipitations
9. Extreme heat and precipitation events

The recent increase in computational capacities made it posible to have high resolutions global ESMs but they are still coastly given the increased complexity of ESMs and requirement to produce high number of ensembles. Therefore RCMs remain indespensible tool for regional climate studies. Regional climate projections provide value addition to ESM projections to better understand regional climate proceses, variability and projections that ultimately help the local policy decisions.

References –

Christensen, Jens Hesselbjerg, et al. “Climate phenomena and their relevance for future regional climate change.” Climate change 2013 the physical science basis: Working group I contribution to the fifth assessment report of the intergovernmental panel on climate change. Cambridge University Press, 2013. 1217-1308.

Doblas-Reyes, F. J., et al. “Linking global to regional climate change.” (2021).

Gutowski, William J., et al. “The ongoing need for high-resolution regional climate models: Process understanding and stakeholder information.” Bulletin of the American Meteorological Society 101.5 (2020): E664-E683.

Kumar, Pankaj, et al. “Downscaled climate change projections with uncertainty assessment over India using a high resolution multi-model approach.” Science of the Total Environment 468 (2013): S18-S30.

Self Organizing Maps

Self Organizing Maps (SOM) is a non-parametric regression of input data to discrete, ordered reference vectors. In a way, it is dimensionality reduction and topology preserving clustering technique. It is a non-linear projection of the probability density function p(x) of the high-dimensional input data vector x onto the two-dimensional display (maps). It is an unsupervised neural network based on competitive learning.

Like any clustering algorithm, SOM algorithm groups the input data such a way that within group differences are minimized and inter-group differences are maximized. The speciality of SOM lies in that, while clustering, it preserves the topological relationships that exist within the primary data, i.e. similar(nearby) inputs are placed in nearby nodes/groups in the output maps. Thus the inputs are grouped in ordered ways.

SOM tool has been successfully used to study various meteorological and oceanographic problems. It is used in synoptic climatology, extreme weather & rainfall pattern analysis, cloud classification, as well as climate change analysis over various regions of the world. Why is SOM so useful for meteorological applications? To understand atmospheric processes, we need to extract characteristic patterns of variability of the interested meteorological variable from a large set of observations of that variable.  SOM not only extracts characteristic patterns but also arranges those patterns in order. In classical synoptic climatology, daily large-scale circulation charts are classified into discrete circulation patterns and these patterns are further studied to understand typical atmospheric processes that impact surface variables/weather. Sometimes different types of circulations can lead to same surface condition/weather with a different mechanism. In such cases, it is important to understand different circulation types and associated mechanisms for a comprehensive understanding of that particular weather phenomenon.  SOM serves as an important tool for synoptic classification or circulation type classification and further process studies. Changes (of frequency, intensity, duration etc.) in identified circulation patterns over time can be used for climate change analysis.

How does SOM algorithm work? SOM results in mapping from input data space ℜⁿ  onto a two-dimensional array of nodes. It associates each input data vector x = [ξ₁,ξ₂,ξ₃,…..,ξ_n] ∈ ℜⁿ  to a node. A reference vector m_i= [μ_i1,μ_i2,μ_i3,…μ_in] ∈  ℜⁿ is associated with each node i. The lattice type of array of nodes can be rectangular, hexagonal or of any shape.

1. Initially reference vectors associated with all nodes are initiated. One can initiate reference vectors by random numbers. During long process of self organization reference vectors attain ordered values. However initialization with regular values can result in faster convergence to ordered values.
2. Next step is to select an input vector randomly  and present it to array of nodes.
3. Input vector is then compared with each node to find out best matching node/unit. Best  matching node is the one which will have the smallest distance from input vector.
4. Once best matching node is selected then the neighboring nodes i.e. nodes within certain geometric distance from best matching node, are also activated to learn something from input vector. This process of local relaxation, during long learning process results into global ordering.
5. During learning process best matching node and nodes withing neighborhood radius are updated to look like input vector. The closer the node to best matching unit , more it gets altered to look like input vector. 

   m_i(t+1)=m_i(t)+h_ci(t)[x(t)-m_i(t)]

   where t = 0,1,2,3,…. , is discrete time step

   h_ci(t) is neighborhood function and such that  h_ci(t) ->0 when t -> infinity and h_ci(t)=h(||r_c – r_i||,t),  where r_c ∈  ℜⁿ and r_i ∈  ℜⁿ are location vectors of node c (best matching node) and i (any other node) respectively also as ||r_c – r_i|| increases  h_ci(t)->0.

   Generally following neighborhood function is used.

h_ci(t)=α(t) .exp {-(||r_c – r_i||²)/(2σ²(t))}

where α(t) is scalar learning rate factor such that (0 < α(t) < 1) and as t increases   α(t) decreases. Usually following learning rate function is used.

α(t)=exp(-t/total no of iterations)

σ(t)  is neighborhood radius, which also decreases with time.

Step 2 to 5 are iterated large number of times to obtain desired result.

 

 

References –

Kohonen, T., 2001. Self-Organizing Maps. Self-organizing maps. 3rd ed. Berlin: Springer, 2001, xx, 501 p. Springer series in information sciences, ISBN 3540679219.

Liu, Yonggang, and Robert H. Weisberg. “A review of self-organizing map applications in meteorology and oceanography.” Self Organizing Maps-Applications and Novel Algorithm Design. InTech, 2011.

http://www.ai-junkie.com/ann/som/som1.html

http://mlexplore.org/2017/01/13/self-organizing-maps-in-go/

https://www.pitt.edu/~is2470pb/Spring05/FinalProjects/Group1a/tutorial/som.html

Click to access l16.pdf

Empirical Orthogonal Functions

Empirical Orthogonal Functions (EOF) or Principal Component Analysis (PCA) is very widely used technique in atmospheric science. Technically, EOFs are nothing but eigen vectors of the co-variance matrix formed by time series of anomalies of any variable (e.g. geopotential, temperature) over the area.  This explanation is useful for understanding the procedure to get EOF, but it cannot tell what is the meaning of it, why we use EOF and how to interpret the results of EOF analysis. Today let us try to understand it intuitively.

EOF helps to understand patterns of simultaneous variations or leading modes of variability. We will try to understand this with the help of  an example. Let us take example of equatorial Pacific SST. We have time series of grided SST data for equatorial Pacific. We want to understand variability of SST there. If we take time mean at each grid point, we will get climatology. To understand variability we can use Standard Deviation (SD) at each grid point give the range . SD at each point wont tell us whether two points vary simultaneously or not but EOF analysis tells us the same. When we do EOF of equatorial Pacific SST we get canonical ELNINO pattern as leading (explaining maximum variability) EOF  pattern. What does this mean? Whatever changes occur in SST of equatorial pacific, majority of the time it follows pattern shown by first EOF. Another intuitive explanation is EOFs are the bases like for 2D vector K the unit vector x  and y along X and Y axis respectively.

 

Coriolis Force

Earth rotates around its own axis with great speed  (angular speed : 7.2921150 ×10⁻⁵  radians / S   or equatorial speed of 465.1 m/s i.e. 1,674.4 km/h). Besides causing day night cycle, there is one major effect of Earth’s rotation termed  as “Coriolis Effect“. It plays important role in large scale motions of atmosphere and ocean. So let us try to understand it.

I always studied it, as the pseudo force arising due to rotation of the earth and it turns the moving objects to right in northern hemisphere and to left in southern hemisphere. I also know mathematical derivation of it. But it always created some confusion in my mind. Being pseudo force, the work done by it, is zero. But still it helps to create cyclones. How?

To understand this let us start with effect of Earth’s rotation on moving objects.

1. the horizontal deflection of vertical motion : Consider we drop vertically an object from great height and we are looking at it from a point on ground. While the object is coming down, the Earth rotates through some distance i.e. the point where we are standing, is moving. But we wont feel the rotation, instead we see that object has been deflected from its vertical path.
2. the vertical deflection of horizontal motion (Eotvos effect) : To understand this effect consider a ball is rotating around its vertical axis. An object is placed on ball and is attached loosely to it, so that it can slide over the ball. Now If the object moves in same sense as that of ball but with more speed, then the object will travel more distance than the ball, i.e. object will appear to be thrown away from ball. In other words, the eastward moving objects are deflected upward and westward moving objects are deflected downward on the earth.
3. the horizontal deflection of horizontal motion (Coriolis effect) :  Suppose we roll a ball from north pole towards equator in straight line. By the time ball reach the equator, it has turned through some distance hence for observer it looks like ball has turned to its right, instead of following straight line. This animation can make it more clear.

Till now we understood that Coriolis force is the fictitious force arising due to movement of frame of reference (an all above cases observer) itself, and no work is done. But, how it helps to form cyclones, is not yet understood. Its not that we just observe the deflections, but circular winds are formed and sustained around lows and highs. If the Earth were not rotating, then once any low pressure area is formed, then wind would blow from high pressure area and kill the pressure gradient and no circular motion is observed. But now on rotating earth, when there is low pressure area and high pressure area, wind will start blowing from high towards low under pressure gradient but by the time it reaches the low, it will move to new position and circular motion around low forms. In scientific words, around lows and highs we get geostrophic flow i.e. flow under the balance between Coriolis and pressure gradient force.

Note – My intention here is to understand concept so not going through the mathematics of it. It can found in any Geophysical Fluid Dynamics book.

Understanding extreme rainfall events in the vicinity of Himalaya

Orography always have effect on spatial distribution of rainfall. Presence of Himalaya (extending upto midtroposhpere) makes the Indian monsoon very intense compared to other monsoons of the world.  Himalaya is east-west oriented huge landmass, occasionally we get extreme rainfall events in this region. These extreme events are very complex events resulting from monsoon-extratropical circulations interactions, owing to latitudinal location of the region, and the orography. Interestingly the dynamical characteristics of extreme rainfall events of western Himalaya (WEH) are different from that of extreme events over Central and Eastern Himalaya (CEH) and north-east India. The dynamics behind these events is unraveled by Ramesh et. al. in two papers separately. 1.”On the anamalous precipitation enhancement over the Himalayan foothills during monsoon breaks”  2. “Monsoon-extratropical circulation interactions in Himalayan extreme rainfall”

Enhanced rainfall over CEH foothills during break period is well known from several years. Large scale circulation aspects of break period are well studied by many people but the dynamics of anomalous rainfall over foothills always remained elusive. During break phase the southward intruding midlatitude westerly trough interacts with monsoon circulation. This interaction causes the mid-level convergence, strong midtropospheric ascent and vorticity stretching over the region and results into the deep convection and concentrated heavy rainfall event. The orography also play important role in enhancing the vertical ascent and rainfall.

Contrary to CEH extreme rainfall events, the enhanced precipitation occur over WEH during the active monsoon period. The extremes (e.g. Northwest Pakistan flood in July 2010 and Uttarakhand flood in June 2013) over WEH are results of vigorous interaction between active monsoon and extratropical circulations. The key steps for this interaction can be given as 1. midlatitude Rossby wave breaking 2. west-northwestward propagation of monsoon low pressure system. 3. eddy shedding of Tibetian anticyclone 4. ageostrophic motion and transverse circulation across Himalaya 5. strong moist convection over the region.

 

 

 

Reasons for decreasing Indian Summer Monsoon Rainfall

The Indian Summer Monsoon Rainfall (ISMR) shows decreasing trend since 1950. The recent paper by Krishnan et. al., titled “Deciphering the desiccation trend of the South Asian monsoon hydroclimate in a warming world”, attribute this observed trend to regional land-use changes, anthropogenic-aerosol forcing and the rapid warming signal of the equatorial Indian Ocean. (Link to paperhttp://dx.doi.org/10.1007/s00382-015-2886-5)

Key points of the paper –

1. June – September (JJAS) seasonal rainfall averaged over the Indian land region (70°–90°E, 10°–28°N) shows the decreasing trend of 7% with respect to mean rainfall value for years 1950 to 2005. Along with this, clear increase of drought incidences is observed in recent decades.

2. Historical experiment (natural plus anthropogenic forcing) shows the observed decreasing trend but the trend is absent in historical natural experiment (natural forcing only). This clearly indicates that decrease is due to anthropogenic changes. Similar facts are observed for Standard Precipitation Evapotranspiration Index (SPEI). SPEI is measure of aridity.

3. There is increase in frequency of precipitation extremes (rainfall >= 100 mm/day) along with weakening monsoon, specially observed over central India. Increasing frequency of precipitation extremes is also attributed to anthropogenic changes. Historical natural experiment shows absence of increase in frequency of extreme events contrary to historical experiment.

4. Further to distinguish between effect of Green House Gas (GHG) forcing and other forcing (aerosol, land use change and warming equatorial Indian ocean SST) two sets of experiment carried out. GHG forcing experiment shows the intensification of ISMR while as other forcing shows the observed decreasing trend in ISMR along with weakening monsoon circulation.

Conclusion –
Weakening of ISMR in recent decades attributed to changes in regional forcing i.e. land use changes, aerosols along with warming equatorial Indian ocean SST in addition to GHG forcing. Land use changes and  aerosols increases the planetary
albedo. This increase in albedo weakens the monsoon overturning circulation which in turn results in decline of ISMR.

Air pollution and Delhi

After many days there is bright news on front page of newspaper. Yes, I am talking about Odd-even vehicle rule in Delhi.

I want to congratulate Delhi government for taking such an important decision. I know there are many problems in implementing this. But I welcome governments step towards reducing the emissions . I hope Delhiites understand and care their health and so successfully implement this decision.  Delhi will set milestone again as it set before for Sulfur Dioxide(SO2) and Carbon Monoxide(CO) using CNG. After use of CNG, the SO2 and CO concentrations  in Delhi are within limits.

Why this rule was needed so much –

1. Delhi is one of the most polluted cities in the world.
2. Most of the air pollutants in Delhi consistently exceed 2-4 times the standard guideline limits set by government. They are particulate matter (both PM10 and PM2.5), Nitrogen oxides, Ozone and Benzene.
3. The situation worsens during winter when vertical dispersion of pollutants is arrested due cool and calm air. Such pollutant laden air causes very serious health hazards.

Even other major cities of India are not clean for breathing. I welcome any such decision for Pune as well.

PS – System of Air Quality and Weather Forecasting And Research (SAFAR) monitors the city air quality for metropolitan cities of India. It provides the location specific near real time information and 1-3 day forecast of air quality. Mobile app of it ‘SAFAR-air’ is also available.

A good article in Marathi on same subject.

Reference – DELHI CLEAN-AIR ACTION PLAN Published by Centre for Science and Environment

An Update – This experiment failed to control pollution, as higher traffic emissions observed during the period when the odd-even rule was enforced. Following are conclusions from the paper (Ref 1)  that studied this experiment -“This suggests that  many four -wheeler users chose to commute earlier, to beat the 8:00 AM–8:00PM restrictions , and/or there was an increase in the number of exempted public transport vehicles. Thus, the odd –even rule did not result in anticipated traffic emission reductions in January 2016, likely due to the changed temporal and fleet emission behaviour triggered in response to the regulation.”

Reference – 1. Chandra, B. P., et al. “Odd–even traffic rule implementation during winter 2016 in Delhi did not reduce traffic emissions of VOCs, carbon dioxide, methane and carbon monoxide.” Current Science (00113891) 114.6 (2018).

Potential Temperature

Potential temperature is one of the very important concept in atmospheric sciences.

Definition : it is the temperature of an air parcel when it is brought adiabatically from given pressure to a reference pressure.

Important property of it is that for adiabatic processes it is conserved. I heard above two points so many times while studying atmospheric sciences. But I never understood it. Now after hearing Vinu sir’s class I understood its meaning and significance. In the atmosphere, temperature of a parcel can change due to various reasons. As it rises up pressure decreases and parcel expands and due to this temperature of it changes according to ideal gas equation PV=nRT. So to differentiate between temperature change due to expansion and  due to energy release or gain, potential temperature is used. For adiabatic process in which parcel does not exchange energy with surrounding, the potential temperature of the parcel is conserved i.e. temperature changes occur but that are due to expansion or compression. Hence for adiabatic process potential temperature change means phase change has occurred within parcel and latent heat of energy is absorbed or gained.

One more confusion I had, in stably stratified atmosphere, the potential temperature increases with height. I used to think that something is wrong with this statement. On one hand people say potential temperature is conserved and again they say it increases with height in atmosphere. But when I thought on it for a while I understood, both are independent statements and are true. First one talks about the adiabatic rise of parcel and its potential temperature. Other one talks about the potential temperature of layers of stably stratified atmosphere. In stably stratified atmosphere the pressure decreases with height and to bring it to a standard pressure more and more compression is required. This leads to increase in temperature due to compression hence potential temperature of a layer increases as we go up.

P.S. : While writing this post I first time read Wikipedia article for Potential Temperature and found that it explained the concept very well specially comments part.

Atypical Monsoon 2015

The 2015 Monsoon season yet be completed but so far it showed very atypical characteristics throughout the season.

1. South Peninsula including western ghats remained void of rainfall i.e. very less rain, more than 20% less of its climatological normal value .

2. The absence of widespread light rain associated with stratiform clouds typical characteristic of monsoon. Presence of convective type of precipitation with very heavy rain confined to small region.

3. Very weak Somali jet, negative anomaly of Mascarene high(weaker high), presence of positive anomaly over Arabian sea creating anticyclonic condition near Arabian seas. The strong offshore trough along west coast of India was almost absent. This caused absence of strong monsoon flow that gives rain to South Peninsula, indeed flow passed around the South Peninsula without hitting it.

4. This season is characterized by development of deep depressions and cyclones over bay as well as over land. These are formed due to interaction of maritime air mass from ocean and mid latitude dry air masses.

5. The Tibetian anticyclone remained weaker and north, northwestward of its normal position. This caused the weak easterly shear and allowed development of deep depressions and cyclone.

4. Another important feature of this season is that continuous development of cyclones over Pacific and their movement to north and northeastward. This caused the subsidence over Indian region. Instead of recurving to north, if cyclones move westward, they create vertical motion over Indian region and rain.

5. Till now the ElNino conditions prevail in Pacific ocean (significant warming is observed in Nino3.4 region). We can attribute this years deficit rain to ElNino. Studies show during ElNino years the Northwestern part of India gets deficient rain contributing to overall deficiency. But surprisingly, despite being ElNino year(so far), We received excess rain north and northwest India and other regions remained deficit.