skip to main content

Title: The Historical Development of Large‐Scale Paleoclimate Field Reconstructions Over the Common Era

Climate field reconstructions (CFRs) combine modern observational data with paleoclimatic proxies to estimate climate variables over spatiotemporal grids during time periods when widespread observations of climatic conditions do not exist. The Common Era (CE) has been a period over which many seasonally‐ and annually‐resolved CFRs have been produced on regional to global scales. CFRs over the CE were first produced in the 1970s using dendroclimatic records and linear regression‐based approaches. Since that time, many new CFRs have been produced using a wide range of proxy data sets and reconstruction techniques. We assess the early history of research on CFRs for the CE, which provides context for our review of advances in CFR research over the last two decades. We review efforts to derive gridded hydroclimatic CFRs over continental regions using networks of tree‐ring proxies. We subsequently explore work to produce hemispheric‐ and global‐scale CFRs of surface temperature using multi‐proxy data sets, before specifically reviewing recently‐developed data assimilation techniques and how they have been used to produce simultaneous reconstructions of multiple climatic fields globally. We then review efforts to develop standardized and digitized databases of proxy networks for use in CFR research, before concluding with some thoughts on important next steps for CFR development.

more » « less
Author(s) / Creator(s):
 ;  ;  
Publisher / Repository:
DOI PREFIX: 10.1029
Date Published:
Journal Name:
Reviews of Geophysics
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Climate field reconstructions (CFRs) attempt to estimate spatiotemporal fields of climate variables in the past using climate proxies such as tree rings, ice cores, and corals. Data assimilation (DA) methods are a recent and promising new means of deriving CFRs that optimally fuse climate proxies with climate model output. Despite the growing application of DA-based CFRs, little is understood about how much the assimilated proxies change the statistical properties of the climate model data. To address this question, we propose a robust and computationally efficient method, based on functional data depth, to evaluate differences in the distributions of two spatiotemporal processes. We apply our test to study global and regional proxy influence in DA-based CFRs by comparing the background and analysis states, which are treated as two samples of spatiotemporal fields.We find that the analysis states are significantly altered from the climate-model-based background states due to the assimilation of proxies. Moreover, the difference between the analysis and background states increases with the number of proxies, even in regions far beyond proxy collection sites. Our approach allows us to characterize the added value of proxies, indicating where and when the analysis states are distinct from the background states. Supplementary materials for this article are available online. 
    more » « less
  2. Over the late Holocene, a variety of hydroclimate-sensitive proxies have identified substantial, multidecadal changes in Indian summer monsoon (ISM) precipitation, the most prominent of which is the “4.2 ka event”. This interval, dated to ~4.2-3.9 ka, is associated with severe droughts across South Asia that are linked to societal change. Given the absence of the 4.2 ka event in polar records, the 4.2 ka event is generally associated with low latitude forcings, but no clear consensus on its origins has been reached. We investigated the ISM response to the 4.2 ka event through analysis of aragonite stalagmites from Siddha cave, formed in the lower Paleozoic Dhading dolomite in the Pokhara Valley of central Nepal (28.0˚N, 84.1˚E; ~850 m.a.s.l.). The climate of this region is dominated by small monthly variations in air temperature (21±5˚C) but strong precipitation seasonality associated with the ISM: ~80% of the annual 3900 mm of rainfall occurs between June and September. High uranium and low detrital thorium abundances in these stalagmites yield precise U/Th ages that all fall within stratigraphic order. These dates reveal continuous growth from 4.30-2.26 ka, interrupted only by a hiatus from 3.27-3.10 ka. Overlap with coeval aragonite stalagmites reveals generally consistent trends in carbon and oxygen isotope ratios, suggesting that these stalagmites reflect environmental variability and not secondary (e.g., kinetic) effects. Many stalagmite-based paleomonsoon reconstructions rely on oxygen isotope ratios, which track amount effects in regional rainfall. However, our on-going rainwater collection and analysis program, as well as a previous study conducted in Kathmandu, 120 km the east of Siddha cave, reveals that amount effects in precipitation are weak in this region, particularly during the monsoon season, and thus we rely instead on carbon isotope ratios, which have been demonstrated to track site-specific effective precipitation. Siddha cave stalagmite carbon isotopes, in contrast to other South Asian proxy records, indicate that ISM rainfall increased at Siddha cave from 4.13-3.91 ka. As a further test of this result, we analyzed uranium abundances in the section spanning 4.3-3.4 ka. Uranium serves as an indicator of prior aragonite precipitation and thus of hydroclimate, and like carbon isotopes, suggests increased ISM rainfall coincident with the 4.2 ka event. This precipitation anomaly is nearly identical in timing and structure but anti-phased with stalagmites from Mawmluh cave, northeastern India. We investigated the climatic origins of this precipitation dipole using observational data from the Global Precipitation Climatology Centre (GPCC) and Hadley Center Sea Ice and Sea Surface Temperature (HadISST) products. Preliminary spatial composites suggest that large precipitation differences between Mawmluh and Siddha caves are associated with SST anomalies in the equatorial Pacific. Additionally, superposed Epoch Analysis shows relatively rapid eastern Indian Ocean cooling during the summer monsoon season coeval with large precipitation differences between these sites. Our findings lend support to a tropical Indo-Pacific origin of the 4.2 ka event. 
    more » « less
  3. null (Ed.)
    Abstract We use theNorthern Hemisphere Tree-RingNetwork Development (NTREND) tree-ring database to examine the effects of using a small, highly-sensitive proxy network for paleotemperature data assimilation over the last millennium. We first evaluate our methods using pseudo-proxy experiments. These indicate that spatial assimilations using this network are skillful in the extratropical Northern Hemisphere and improve on previous NTREND reconstructions based on Point-by-Point regression. We also find our method is sensitive to climate model biases when the number of sites becomes small. Based on these experiments, we then assimilate the real NTREND network. To quantify model prior uncertainty, we produce 10 separate reconstructions, each assimilating a different climate model. These reconstructions are most dissimilar prior to 1100 CE, when the network becomes sparse, but show greater consistency as the network grows. Temporal variability is also underestimated before 1100 CE. Our assimilation method produces spatial uncertainty estimates and these identify treeline North America and eastern Siberia as regions that would most benefit from development of new millennial-length temperature-sensitive tree-ring records. We compare our multi-model mean reconstruction to five existing paleo-temperature products to examine the range of reconstructed responses to radiative forcing. We find substantial differences in the spatial patterns and magnitudes of reconstructed responses to volcanic eruptions and in the transition between the Medieval epoch and Little Ice Age. These extant uncertainties call for the development of a paleoclimate reconstruction intercomparison framework for systematically examining the consequences of proxy network composition and reconstruction methodology and for continued expansion of tree-ring proxy networks. 
    more » « less
  4. Abstract

    Paleoclimate reconstructions are now integral to climate assessments, yet the consequences of using different methodologies and proxy data require rigorous benchmarking. Pseudoproxy experiments (PPEs) provide a tractable and transparent test bed for evaluating climate reconstruction methods and their sensitivity to aspects of real-world proxy networks. Here we develop a dataset that leverages proxy system models (PSMs) for this purpose, which emulates the essential physical, chemical, biological, and geological processes that translate climate signals into proxy records, making these synthetic proxies more relevant to the real world. We apply a suite of PSMs to emulate the widely-used PAGES 2k dataset, including realistic spatiotemporal sampling and error structure. A hierarchical approach allows us to produce many variants of this base dataset, isolating the impact of sampling bias in time and space, representation error, sampling error, and other assumptions. Combining these various experiments produces a rich dataset (“pseudoPAGES2k”) for many applications. As an illustration, we show how to conduct a PPE with this dataset based on emerging climate field reconstruction techniques.

    more » « less
  5. Obeid, I. ; Selesnik, I. ; Picone, J. (Ed.)
    The Neuronix high-performance computing cluster allows us to conduct extensive machine learning experiments on big data [1]. This heterogeneous cluster uses innovative scheduling technology, Slurm [2], that manages a network of CPUs and graphics processing units (GPUs). The GPU farm consists of a variety of processors ranging from low-end consumer grade devices such as the Nvidia GTX 970 to higher-end devices such as the GeForce RTX 2080. These GPUs are essential to our research since they allow extremely compute-intensive deep learning tasks to be executed on massive data resources such as the TUH EEG Corpus [2]. We use TensorFlow [3] as the core machine learning library for our deep learning systems, and routinely employ multiple GPUs to accelerate the training process. Reproducible results are essential to machine learning research. Reproducibility in this context means the ability to replicate an existing experiment – performance metrics such as error rates should be identical and floating-point calculations should match closely. Three examples of ways we typically expect an experiment to be replicable are: (1) The same job run on the same processor should produce the same results each time it is run. (2) A job run on a CPU and GPU should produce identical results. (3) A job should produce comparable results if the data is presented in a different order. System optimization requires an ability to directly compare error rates for algorithms evaluated under comparable operating conditions. However, it is a difficult task to exactly reproduce the results for large, complex deep learning systems that often require more than a trillion calculations per experiment [5]. This is a fairly well-known issue and one we will explore in this poster. Researchers must be able to replicate results on a specific data set to establish the integrity of an implementation. They can then use that implementation as a baseline for comparison purposes. A lack of reproducibility makes it very difficult to debug algorithms and validate changes to the system. Equally important, since many results in deep learning research are dependent on the order in which the system is exposed to the data, the specific processors used, and even the order in which those processors are accessed, it becomes a challenging problem to compare two algorithms since each system must be individually optimized for a specific data set or processor. This is extremely time-consuming for algorithm research in which a single run often taxes a computing environment to its limits. Well-known techniques such as cross-validation [5,6] can be used to mitigate these effects, but this is also computationally expensive. These issues are further compounded by the fact that most deep learning algorithms are susceptible to the way computational noise propagates through the system. GPUs are particularly notorious for this because, in a clustered environment, it becomes more difficult to control which processors are used at various points in time. Another equally frustrating issue is that upgrades to the deep learning package, such as the transition from TensorFlow v1.9 to v1.13, can also result in large fluctuations in error rates when re-running the same experiment. Since TensorFlow is constantly updating functions to support GPU use, maintaining an historical archive of experimental results that can be used to calibrate algorithm research is quite a challenge. This makes it very difficult to optimize the system or select the best configurations. The overall impact of all of these issues described above is significant as error rates can fluctuate by as much as 25% due to these types of computational issues. Cross-validation is one technique used to mitigate this, but that is expensive since you need to do multiple runs over the data, which further taxes a computing infrastructure already running at max capacity. GPUs are preferred when training a large network since these systems train at least two orders of magnitude faster than CPUs [7]. Large-scale experiments are simply not feasible without using GPUs. However, there is a tradeoff to gain this performance. Since all our GPUs use the NVIDIA CUDA® Deep Neural Network library (cuDNN) [8], a GPU-accelerated library of primitives for deep neural networks, it adds an element of randomness into the experiment. When a GPU is used to train a network in TensorFlow, it automatically searches for a cuDNN implementation. NVIDIA’s cuDNN implementation provides algorithms that increase the performance and help the model train quicker, but they are non-deterministic algorithms [9,10]. Since our networks have many complex layers, there is no easy way to avoid this randomness. Instead of comparing each epoch, we compare the average performance of the experiment because it gives us a hint of how our model is performing per experiment, and if the changes we make are efficient. In this poster, we will discuss a variety of issues related to reproducibility and introduce ways we mitigate these effects. For example, TensorFlow uses a random number generator (RNG) which is not seeded by default. TensorFlow determines the initialization point and how certain functions execute using the RNG. The solution for this is seeding all the necessary components before training the model. This forces TensorFlow to use the same initialization point and sets how certain layers work (e.g., dropout layers). However, seeding all the RNGs will not guarantee a controlled experiment. Other variables can affect the outcome of the experiment such as training using GPUs, allowing multi-threading on CPUs, using certain layers, etc. To mitigate our problems with reproducibility, we first make sure that the data is processed in the same order during training. Therefore, we save the data from the last experiment and to make sure the newer experiment follows the same order. If we allow the data to be shuffled, it can affect the performance due to how the model was exposed to the data. We also specify the float data type to be 32-bit since Python defaults to 64-bit. We try to avoid using 64-bit precision because the numbers produced by a GPU can vary significantly depending on the GPU architecture [11-13]. Controlling precision somewhat reduces differences due to computational noise even though technically it increases the amount of computational noise. We are currently developing more advanced techniques for preserving the efficiency of our training process while also maintaining the ability to reproduce models. In our poster presentation we will demonstrate these issues using some novel visualization tools, present several examples of the extent to which these issues influence research results on electroencephalography (EEG) and digital pathology experiments and introduce new ways to manage such computational issues. 
    more » « less