 is Paul Diemeier and he is a professor at the George Mason University. Paul is an expert on the role of the land surface in the climate system and he's interested in the development and application of land surface model as he studies the impact of land surface variability on the predictability of climate and interactions between the terrestrial and atmospheric branches of the hydrological cycle. Paul we're looking very forward to your talk and it is so nice to not only look only at the ocean and the atmosphere but also the other components which do play a role in estrous predictability. Go ahead Paul. Thank you Judith and Anish again giving me a second chance to speak hopefully you see the full screen presentation I trust it's working all right. I'm going to end up in very much the same place as the questions that Andy just answered so well so but the starting point just as a reminder to the students and it's been four weeks now there's been so much coming past you just to kind of reorient that for the land surface what we're really looking for here is when and where is the land surface important to atmosphere weather and climate because it's not important in all places and all locations the in all times it's very similar to to Enso to the El Nino phenomenon you can think people even the general public now knows to sort of pay attention to ssts in the tropical Pacific that somehow that has some connection to weather and climate all over the globe over the US especially in the winter time right we that's something that we really focus on and maybe not so much in the summer it's kind of a similar situation with the land surface but the the pathways are a bit different in s2s forecasting sub seasonal to seasonal climate anomalies are driven by as you've been hearing for weeks these sort of persistent large-scale circulation features that's what we really hang our hat on it's really these long atmospheric ways that have remote sources from the the mjo or from the tropics or even coming down from the stratosphere communicated as it were to where we live which is basically most of us living on continents we're not living in the middle of the tropical pacific ocean where we live on continents it's rosby wave propagation right that's carrying the signal to us and so predictable s2s phenomenon are sort of delivered by the atmosphere to where we live in much the same way that you might say you know afraid companies delivering your amazon order or that your internet service provider is delivering you data to your house so if we think of this analogy then we can think of the atmosphere as being that sort of large-scale shipping but then like communications or shipping or things often break down is that last mile that last kilometer to your house right once the package is in your neighborhood 30% of the cost and most of the failures in this kind of shipping or in isps occur in your neighborhood at that final step to the customer's door and so this is a very similar analogy that we would post for forecast models if the land surface in your neighborhood in your country in your state in your continent is poorly initialized or if the couple land atmosphere processes are not well represented then the delivery breaks down or as lost or as garbled and so we've kind of transitioned from this dynamics fluid dynamics picture with the atmosphere down to something local that's a lot more thermodynamics we're talking about with the land surface and the atmosphere above it and it becomes much more of a regional thermodynamic problem so a quick review a quick reminder kind of where we stand right now about 20 years ago the the Glossier experiment revealed this notion of hot spots of land atmosphere coupling or feedbacks that tend to be in these transition regions between arid and humid locations and so these these hot spots were found initially through modeling studies only without observational data because we didn't have the observational data at the time and so the way this was determined in models was basically through a very clever design of a model experiment where we run a large ensemble of simulations and in one case all the ensemble members are initialized identically but the same land surface state and when the other one they're all sort of jumbled up they're all sort of randomized or or something is done differently and so when we do that we can look at the coherence within an ensemble as a measure of the feedback this omega parameter shown here so the idea is basically that when you have an ensemble and you specify the land surface to be identical and all the ensemble members and specify it you know potentially throughout the entire integration then if the land surface is actually controlling the atmosphere then the atmosphere should be kind of more constrained more coherent less spread in the ensemble members they should track each other more closely if the land surface states have no bearing on it then it won't change that spread at all and the ensemble will spread just as it did when you just did an initialization or just you know randomized surface so coherence is a hint of predictability and an increase in coherence say when you specify the same land surface state says ah this could be a source of predictability and its coherence then can potentially carry the signal in time to these sub seasonal to seasonal distances so in our gloss case this omega increased the spread when you increase the spread in land surface states it increases the spread in your ensemble forecast and that was our clue that's basically what this map is telling us a quick reminder just on the processes we see this as links in a chain so for instance the little pseudo equation at the bottom there if you have an increase in rainfall then you're going to have an increase in soil moisture that has the potential to lead to increase in evaporation more moisture in the atmosphere might lead to more increase in rainfall or sustaining rainfall that's the positive feedback idea and what we talked about before was when you have a high positive correlation between soil motion evaporation that means you're in a moisture limited regime variations in soil moisture are controlling the evaporation when you have an energy limited regime where you have a lot of water in the soil but maybe not enough sunlight it's not warm enough then it's actually the energy availability that's the limiting factor and that's what controls evaporation and then soil moisture is kind of along for the ride and it becomes negatively correlated that's what we see in that middle panel but for that to get communicated the atmosphere the atmosphere has to be ready and it has to play along and so the atmosphere has to be in a state where it's responsive to a change in surface fluxes and if all the pieces of the puzzle are in place then that's when soil moisture can control things like boundary layer depth cloud formation and the amount of rainfall we saw that that sensitivity does not occur in all ranges of soil moisture there's a particular range where we have a lot of sensitivity that's kind of outlined in red here kind of moderate amounts of soil moisture in the soil not too wet not too dry goldilocks sort of zone above some amount the moisture in the soil is so high that again evaporation levels off all the variation and evaporation is about the energy availability not the water availability and likewise at the low ends we can shut down to a point where there's basically no evaporation and again this is all thermodynamics it's not dynamics how these feedbacks occur is is through this pathway and as soon as the sun comes up you have energy potentially to evaporate moisture so if you're in that red box then you're sort of ready to go the slope of that line is the sensitivity the stronger the slope then the more surface fluxes change with the change in soil witness and we pointed out before that the atmospheric part also has different ranges of response and again kind of the medium range of evaporation is where precipitation seems to be most sensitive again you know not too hard not too soft the bed is just right for goldilocks glossay 2 was our prediction version of glossay not a predictability experiment but actually doing hind casts it was kind of like s2s before s2s prediction project and so what we saw there was a realistic soil moisture initialization improves the forecast and so what you see on the right here is temperature prediction skill the dots are significant areas yellow is good lead time is the different rows but here I'm also showing that the extremes really contribute to prediction skill so when we look at the extreme terciles or the quintiles are on the far right the desiles the top 10% wettest and driest soils are contributing almost all of the skill that we see in the whole range just like a strong El Nino or a strong line Nino is where we get most of our seasonal predictability when El Nino is weak and it's hard to hard to get a forecast same thing applies for soil moisture so the skill impacts are also longer where there's that memory where there's that persistence of a soil moisture anomaly it just happens to turn out over North America that the hot spot where the coupling is the strongest is also a place where the memory is weak so we see in these maps that the great planes kind of fades away after 30 days 45 days and is no longer an area of predictable temperature the coupling is strong there but the memory of the soil moisture anomalies just isn't there okay so this all affects prediction skill in Glossy 2 and as we said before a quick reminder if you're in kind of a dry area or it's typically kind of too dry to have much evaporation and you have a wet spell oh suddenly you've moved into that sensitive range likewise if you're in kind of a wet area where it's usually too wet for the variations in soil moisture to affect surface fluxes and have no response but you go into kind of a dry phase oh now suddenly that area is in a sensitive regime and so in each case if you're in that area with a strong slope of surface fluxes as a function of soil wetness then you can have land atmosphere interactions and potential predictability and potential prediction skill to harvest a great example of that wet area getting dry and suddenly being able to give a feedback is what we saw in northern europe in 2018 during the drought and heat wave in that region and we saw this plot before too a couple weeks ago the temperature extremes in this case it's the fraction of days in may june and july august of 2018 that were among the five percent of the warmest days for the whole uh 40 year period leading up to that time and in the bottom it's it's for soil moisture although we're taking the top quarter the top quartile instead of the top five percent and what we said was that there is kind of on the bottom of that scale there's the place where the the evaporation basically shuts down and any energy any any net radiation at the surface goes into sensible heat flux and just gets pumped right back in the atmosphere as heat that's this critical threshold this break point where we can really get an amplification in the strength of a heat wave because the land atmosphere feedbacks have now sort of cranked it up to 11 and so that's kind of shown in the bottom here uh bottom left figure but to kind of expand on that intensification idea so effectively when you drop below if you know any soil hydrology when you drop below the wilting point uh when when plants become very stressed and basically shut down and stop trying to do um uh any sort of photosynthesis and thus are also not transpiring any moisture to the atmosphere that's when we get this really strong feedback and that threshold depends on the type of soil that you have it depends on the type of vegetation the map on the left shows this climatological threshold that we've produced over europe for that soil moisture shutdown and there's a lot of spatial variability it's not just a function of latitude or or or anything and all that fine structure is because of different vegetation different soils in those different locations even orography can affect this and the hypersensitivity of the maximum temperatures on declining soil moisture is shown on the right the darker the colors the more quickly temperatures will ramp up as you cross down and dry out the soil by some increment of of wetness and so you see dark colors in places like in in iberia and in uh in turkey over ukraine even in sort of northern france around paris there's that band in southern britain for instance that you see that has the darker colors that's the area where they have chalk soils or chalk underneath the soils it's kind of a karst formation which has a very different interaction with the vegetation and the soils above it then further north in britain where you kind of have a typical bedrock substrate underneath so all these things have an effect even the geology we talked about that the time scale where this really emerges is in the s2s range i showed this picture before and compared it to our um idealized schematic it shows the land sort of peaking in the sub-seasonal range there's not a lot of impact of land or ocean at very short time scales because it's all an initial condition problem to the atmosphere all the skill comes from the atmosphere there's not much room left to improve the skill with the land or the ocean but once you get out five days a week two weeks three weeks suddenly it becomes from the atmospheric perspective a boundary value problem not an initial value problem but from the land or the ocean no what do you mean boundary it's us there's the atmosphere is our boundary to them it's still an initial value problem it was what was the initial soil moisture what's the initial ssts if you think about it as a system which is the way we should think about it as you know andy mentioned uh then you have to look at the whole picture and realize that it's it's a couple of feedback and you just have different components that have different memories and different uh scales to to persist their initial states so this land impact on prediction skill here we're showing it actually quantified um a realistic initial initialization initialization of the land surface can extend significant skill and prediction of air temperatures in this case we're looking at pentads what we see here is that how many pentads is significant skill extended beyond what we had before by using the right land initialization instead of say random or climatological initialization about 40 of the land area of the globe has skill extended by at least two pentads 10 days or more about 80 at least five days or more extended by doing these pentad averages which suggests neglecting initialization is going to be a big problem for forecast skill and we can also look at it kind of the way the cpc always looks at you know monthly chunks if we look at month one month two month three then we see a similar picture that 30 to 50 percent of the globe has skill extended by at least one month looking at temperature and here also humidity and dry season subtropics three months or more and this is with the old cfs version two which is not a state-of-the-art model it's a really kind of a kind of an old beater of a model at this point it's kind of been in use for nearly 15 years now the new ufs model is being developed has a potential to improve upon this by improving the models as well not just the initialization let me skip this one i want to get to the initialization question assimilating soil moisture this came up and the last couple talks about soil moisture so this is actually showing information content this is from information theory in of soil moisture in these different products the x axis is called metric entropy if it's really high then you basically got noise random noise if it's really low on the left side zero means it's periodic and somewhere in between is kind of a mix of the two the y axis is the complexity in the signal how much information content is there and what we see is that in situ measurements of soil moisture outlined here in blue what you have is is the our kernel density are high our high blob here is really it's not too noisy it's not too periodic it's got a lot of information when we look at the models they kind of replicate that to some extent they're a little too periodic a little too deterministic our models you know they tend to have trouble expanding and getting the whole envelope but what we see here is that the the satellites the old satellites at least ever soil moisture which were not purpose built to measure soil moisture mostly are just noise there's not a whole lot of information in them the SMAP satellite from NASA soil moisture it's still got a lot of noise but man it looks so much better and in fact the information content looks very useful and much more comparable to in situ measurements which is really giving us a lot of promise for initializing the land globally a lot of forecast centers already do land data simulation of satellite soil moisture data for initialization ECMWF did this starting with era two and it does it operationally so does medial France does it KMA does it the Environment Canada does it some places don't do it and so it's going to be a couple years before they get their system going but eventually that will come in okay so to wrap up land models and atmosphere models have been developed separately and sort of plugged together it's a system we need to consider model development as a system the coupling is very important it's important for predictability and prediction until recently we didn't have a lot of data either the in situ or the satellite data of good enough quality of good enough coverage to really do this sort of validation but now we can and now we're starting to do that and so this is a picture like to show nature land and atmosphere are involved in this in this beautiful elegant dance going back and forth and our models historically have not represented this in a very elegant realistic fashion but now we have a chance to model nature much better and improve our S2S predictions which is really where the land could have a lot of impact because it is a coupled system and I will stop there oh one thing here's the references that were were cited in the previous slides thank you thank you very much very comprehensive very educational thank you so much I I have a question so please post your questions or raise your hands but I wanted to ask you about uncertainty so yes we often have that problem in in the atmosphere that we're under dispersive and that problem gets probably worse on the S2S scale and I was wondering if you could say about the uncertainty sort of in the land parameters and if you have enough observation to constrain them do you have that same problem of under dispersion and in that context can you talk a little bit about um separate scale heterogeneity yes uh wow that is a lot of that question uh yes absolutely uh yeah you under dispersive is the right term and we have exactly that same situation with the land surface um parameter choosing is a big problem um as is the case with say you know rainfall data and so many other things we have a lot of information in the developed world uh so you know Europe North America East Asia Australia we have a lot of good data on soil parameters you know in situ that have been gathered and so forth and vegetation we don't in a lot of parts of the world we have to rely on satellites to give us the global coverage and then there's the usual problems of interpreting the satellite data although some clever people have worked out ways to try to back out the kind of information we need from the variability they see in long satellite records but it's still a little bit klugey in that regard one thing I didn't point out but is relevant to the heterogeneity is that when we do the in-situ validation that's a site and even if you have a flux tower that flux tower kind of has a footprint of at most a square kilometer usually on the areas of you know thousands of square meters which is much smaller than any of our model grids so there's also the concern about scale and heterogeneity is always an issue it doesn't matter how how fine you go with your model there are always scales that are finer than your highest resolution and for the land surface in particular surface heterogeneity can drive mesoscale circulations which can contribute upscale to a lot of what we see at the grid scale and are simply not resolved in our climate models especially getting to these convection permitting or cloud permitting scales of models has it has a lot of promise is going to help quite a bit but there's still heterogeneity below those scales potentially that could be important and we have actually right now a CPT climate process team project with NOAA to look at heterogeneity in the land surface and how that affects the atmosphere actually two projects there's another one that looks at heterogeneity and atmospheric radiation and how that affects the land surface and so we're actually trying to address that to create a sort of a subgrid land surface parameterization kind of on the scale of what we have in atmospheric models now for convection like club for instance that can be used and coupled to these subgrid unresolved convection models so that we can get the land surface drivers more correctly done in our in our core scale models. Yeah and if I my comment I've been involved in a few collaborations with NOAA where we looked at perturbing the parameters in the in the land model in the hurry at the time and unfortunately we introduced huge biases when we perturbed things like hydraulic conductivity and so in our experience it was beneficial to perturb the initial condition but not to I mean the idea was representing subgrid scale heterogeneity the bias was just too big to to to be more beneficial than the representation of uncertainty Jan has a question please go ahead. Yeah thank you very much for this very interesting talk my question was a little bit in the direction of the one plot where you showed the predictability on S2S timescales that comes initially from the atmosphere and then the land surface kind of on the second probably to fifth week timescale and in the context of this high sensitivity of the soil moisture to heating to changing atmospheric conditions and that you showed over Europe for example and my question is a little bit okay I'm trying to face it well so if we can predict for example blocking over over Europe better so we increase our prediction of the atmosphere on those timescales up to two weeks does that would that have an impact then on on the land processes I hope certainly yeah and increase predictability potentially I hope that came across as yeah no yes no certainly and because it's a coupled system so um in order to get the soil moisture forecast correct three four or five weeks out in advance you would like to have the right rainfall you would like to have the right temperature and that radiation at the surface and that would come from having a better atmospheric prediction you'd like the clouds to be correct so and then in so doing then if you have the correct soil moisture that could then feedback and and help improve the atmospheric states so yes very much so that you know the places where we do have the land controlling the atmosphere you would like the land states to be correct we talked about there being that sort of break point where you have the higher sensitivity in a heat wave or a drought you would like your model to be on the left or right side of that break point when reality is on the right or left side of that break point or you like it to cross that threshold the same time your forecast cross the threshold the same time the real world is crossing the threshold so these are the kind of details that we're that we're looking at so yeah i mean it's i have this you know cynical thing i say that a forecast model is just a carefully balanced set of errors and when you correct and remove one error then all the other errors are upset and start manifesting so um you really have to look at all the pieces of the puzzle you can't just put all the blame on one component or all the the attempt to to correct a bad forecast on one component it's all connected thank you thanks so much i should memorize that one yeah i have a question back related to your coupling work and first of all a very nice presentation it's always interesting to see this even parts i've seen before so enjoyable so in the coupling aspect you mentioned that one of the key pieces is that memory you know where you have strong coupling and where you have strong memory right you need to predictability but do you think that our models are evaluated to see if their memory is correct or not i mean do we know that they tend to have good memory due to whatever parameters and incineration yeah now that that's something that we we're starting to look at i should say starting we've been doing this for some time where we have in situ soil moisture measurements um we have been evaluating soil moisture memory and comparing that to models again where are most of the measurements it's it's in the u.s it's in europe it's in australia it's in china um places like uh sub-saharan africa where there's where this where soil moisture is such a critical component we know from modeling studies we just we don't have data in those locations we've got some satellite data now um snap has only been up for five six years now so it's kind of hard we're just sort of barely getting to the threshold where we might start to have enough data to hang some significance on that i think as time goes forward we'll be collecting more and more satellite data and we can make a better assessment of memory but yeah persistence of anomalies is a problem oh and one problem with the satellite i didn't mention uh these microwave-based uh satellites is that they're only seeing a couple of centimeters of the soil and if you got much vegetation at all they're really just seeing the water and the vegetation they're not even seeing the soil so what's going on subsurface is much more of a mystery and not easily addressed by satellite so as you well know Anish is that a new hand or the old hand it's a new hand so yeah my question is i was recalling the study from GFDL i think Gabe Becky and group had the study on the ENSO teleconnection in the 2015-16 case and how the teleconnection was not the canonical teleconnection on the western u.s that the southwest u.s was not wet and the opposite in the pacific northwest and they showed that the land initialization on the seasonal timescale played a big role in in getting the teleconnection right at least with the GFDL model right i was wondering so how much is the land surface land atmosphere coupling a function of the regime dependence especially over western u.s and do we understand that fully and how much it modulates the ENSO teleconnections into u.s yeah i can i don't know about that that's the specific case of the western u.s but certainly yes that that has to do with the thermodynamic state of the atmosphere and whether it's responsive to to variations in the land surface and that can be controlled to some extent by whether you're in a ridge or a trough for instance they're going to have different different responses different stabilities different profiles of moistatic energy or dry static energy so yes it's it it is all connected and and furthermore some of these large land anomalies large in magnitude as well as extent particularly when they are over elevated terrain such as well in the u.s would be over the western u.s over the rocky mountains or the the canonical example is the tibetan plateau when you have a an anomaly in land surface states on elevated terrain it's kind of effectively putting any heating anomalies up into the middle troposphere right away so it's kind of similar to having an alnino signal and there can be a lot more response to the atmosphere and teleconnection patterns from the land surface anomalies in those locations the bolivian altiplano is is the third area in the world that kind of has this large region where you could have that when you have a similar say temperature anomaly or humidity anomaly at land it's near sea level it it just doesn't get up into the atmosphere enough to affect anything but sort of the local and the regional areas it doesn't create a planetary scale wave but when you have the anomalies on elevated terrain it can so that's another thing that's actually relatively new just in the last a few years have people started looking at that and there's an ls4p project at yung kong shui at ucla is leading that's really looking into that and doing some some really interesting modeling studies thanks thank you and the questions i don't see any heads um so um this concludes our day for today thank you so much pal for giving us a few and um i see everyone tomorrow at 9 a.m for the last day of the workshop thanks to all the speakers and all the good questions from the audience see you tomorrow