Robustness of Process-Based versus Data-Driven Modeling in Changing Climatic Conditions

Abstract. In hydrological climate-change impact studies, regional climate models (RCMs) are commonly used to transfer large-scale global climate model (GCM) data to smaller scales and to provide more detailed regional information. Due to systematic and random model errors, however, RCM simulations often show considerable deviations from observations. This has led to the development of a number of correction approaches that rely on the assumption that RCM errors do not change over time. It is in principle not possible to test whether this underlying assumption of error stationarity is actually fulfilled for future climate conditions. In this study, however, we demonstrate that it is possible to evaluate how well correction methods perform for conditions different from those used for calibration with the relatively simple differential split-sample test. For five Swedish catchments, precipitation and temperature simulations from 15 different RCMs driven by ERA40 (the 40 yr reanalysis product of the European Centre for Medium-Range Weather Forecasts (ECMWF)) were corrected with different commonly used bias correction methods. We then performed differential split-sample tests by dividing the data series into cold and warm respective dry and wet years. This enabled us to cross-evaluate the performance of different correction procedures under systematically varying climate conditions. The differential split-sample test identified major differences in the ability of the applied correction methods to reduce model errors and to cope with non-stationary biases. More advanced correction methods performed better, whereas large deviations remained for climate model simulations corrected with simpler approaches. Therefore, we question the use of simple correction methods such as the widely used delta-change approach and linear transformation for RCM-based climate-change impact studies. Instead, we recommend using higher-skill correction methods such as distribution mapping.

Benchmarking data-driven rainfall-runoff modeling across 54 catchments in the Yellow River Basin: Overfitting, calibration length, dry frequency

The relationship of δD and δ18O in soil water and its implications for soil evaporation across distinct rainfall years in winter wheat field in the North China Plain

The present study investigates the year-to-year variations of September–October rainfall in Hainan, China, for the period 1965–2010. The dominant circulation anomalies feature a cyclone (an anticyclone) over the Indochina Peninsula and northern South China Sea, an anticyclone (a cyclone) over subtropical western North Pacific and lower-level convergence (divergence) over the Maritime Continent in the wet (dry) years. These circulation anomalies are responses to an east–west sea surface temperature (SST) anomaly pattern with negative (positive) SST anomalies in the equatorial central Pacific and positive (negative) SST anomalies around the Maritime Continent in the wet (dry) years. Although the SST anomaly pattern is similar (but with opposite anomaly), the SST signal in the equatorial central Pacific is more significant in the dry years than in the wet years. This difference indicates a larger case-to-case variability in the wet years than in the dry years. The large variability in the wet years is attributed to contributions of tropical cyclones (TCs) and intraseasonal oscillations (ISOs). There are more TCs impinging on Hainan and the TC tracks are closer to the island in the wet years than in the dry years. The rainfall shows large intraseasonal variations with periods of 10–20 and 30–60 days during September–October in the wet years. The 10–20-day ISO originates from the Maritime Continent, whereas the 30–60-day ISO develops over tropical Indian Ocean and propagates northeastward to northern South China Sea. In contrast, the ISO signal is much weaker in the dry years.

Hydrological models often perform poorly in simulating dry years in regions with large inter-annual variability in rainfall. We calibrated the Soil and Water Assessment Tool (SWAT) model to dry and wet years separately, using the semi-arid Barrett watershed on the west coast of USA as an example. We used hydrological and meteorological data from 1980–2010 to calibrate the SWAT model parameters, compared the monthly runoff results simulated by SWAT using a traditional calibration for the entire runoff series with results using a calibration with the wet and dry year series, and analyzed differences in the most sensitive parameters between the wet and dry year series. The results showed that (1) the SWAT model calibrated to the entire runoff series produced significant differences in simulation efficiency between the wet years and dry years, with lower efficiency during the dry years; (2) the calibration with separate wet and dry years greatly enhanced the SWAT model’s simulation efficiency for both wet and dry years; (3) differences in hydrological conditions between wet and dry years were represented by changes in the values of the six most sensitive parameters, including baseflow recession rates, channel infiltration rates, Soil Conservation Service (SCS) curve number, soil evaporation, shallow aquifer flow, and soil water holding capacity. Future work can attempt to determine the physical processes that underlie these parameter changes and their impact on the hydrological response of the semi-arid watersheds.

Green infrastructures are decentralized semi-natural solutions for managing stormwater runoff. To assess their effectiveness in urban drainage, many process-based hydrological models have been developed. However, these models sometimes fail to deliver satisfactory results due to inadequate representations of the involved hydrological processes or due to limited field measurements for model setup. Data-driven modelling simulates directly the connections between the state variables of the system (such as the input variable and the output variable), and thus reduces the need for hydrological process characterization. Yet the usefulness of this modelling approach in green infrastructure related studies has not been fully explored. To demonstrate its effectiveness, two bioretention systems in the U.S. are studied. The designs of these two systems are unique, making them difficult to be modelled directly using the standard process-based models. In one site, the observations are also censored, i.e., only the overflow/no overflow binary outcomes are being monitored, with which the model calibration is inhibited. We showed that using the state-of-the-art data-driven modelling framework (e.g., CARET and MXNET), the occurrence or the outflow discharge rate of the bioretention systems at a given time can be accurately simulated only using the rainfall, evapotranspiration and the inflow time series. We conclude that data-driven modelling can be useful for simulating the hydrologic performance of green infrastructures, especially when the existing process-based models are inadequate in representing the design variabilities or the data for model setup is limited.

This study utilizes a linear, primitive equation spherical model to study the development and propagation of easterly wave disturbances over West Africa. Perturbations are started from an initial disturbance consisting of a barotropic vortex and the governing equations are integrated forward in time. The perturbations are introduced into basic states corresponding to the observed dynamical and thermodynamical characteristics of two wet years in the Sahel and two dry years. The model simulations show consistent contrasts in wave activity between the wet and dry years. The waves are markedly stronger in the wet years and show a barotropic structure throughout the troposphere. The waves tend to extend throughout the troposphere to the level of the tropical easterly jet (TEJ) in the wet years, but not in the dry years. The upper-tropospheric shear, which is stronger in wet years, appears to be a key factor in wave development. This shear is dependent on the intensity of the TEJ, suggesting that the TEJ is an important factor in interannual variability in the Sahel. When the overall shear is weak, vertical development is suppressed. Another contrast is that in the dry years the growth rates show a single maximum around 3000–4000 km, but in the wet years there is a second, around 6000–7000 km. This suggests that both synoptic-scale and planetary-scale waves are active in the rainy season of some wet years. Imposing considerations of potential vorticity, the generation of planetary-scale waves implies a strong link between the surface and the TEJ in wet years. Such a link is absent in the dry years. This is likely a major factor in the interannual variability of rainfall in the Sahel.

The impacts on ecosystems of global and local shifts in the environment due to e.g. climate change, land use change, and urbanisation present us with complex and interacting societal challenges. Solving a single problem often aggravates another. Scientific insights needed to face these challenges are based on interdisciplinary multi-scale research. It requires that we work together across (sub)disciplinary boundaries and combine our insights and knowledge to understand the complexities and interrelatedness within ecosystems. Making this combination is currently hampered by the availability and scatteredness of datasets and the lack of methods and tools to combine knowledge, data and models. Developing digital twins of ecosystems could provide a paradigm to achieve this combination of knowledge, data and models, and generate the insight and knowledge that we require to face the challenges of global and local change. A digital twin of an ecosystem is a digital representation of an ecosystem tailored to the user’s research objectives and perspective of that ecosystem. They bring together (long-term) data collected of multiple ecosystem facets and across multiple scales, process-based models, and data science and computational modelling tools, making them a versatile instrument. Digital twins can be used to understand ecosystem functioning, by aiding the discovery of intricate ecological relationships not yet understood or measured. Furthermore, they are ideal for exploring the effects of in situ management strategies, or for the detection of early-warning signals of global change impacts on ecosystem dynamics. LTER-LIFE is a scientific infrastructure, embedded in LTER-NL (Long-Term Ecosystem Research Netherlands) and LifeWatch ERIC, aimed at supporting shared and integrated ecological research, providing scientists the platform and tools to build digital twins of ecosystems. We do this by supporting the sharing of data and models and bringing these together into a site-focused virtual lab where people can work together. LTER sites provide an ideal opportunity for this as they often come with a long history of extensive data collection and research across multiple spheres. Drawing on experience of building digital twins in the Veluwe LTSER platform, this keynote will take you through the journey of digital twinning parts of an ecosystem with a long legacy of ecological and environmental research and data collection. As these data exist in many different formats, we built data mobilisation and FAIRification pipelines to augment the focal experimental sites in the Veluwe with relevant historical, privately-owned or national datasets. Subsequently, we explored the potential of the Notebook as a Virtual Research Environment (NaaVRE), which provided us with a virtual lab in which we developed workflows for processing and linking FAIRified data with process-based and data-driven models. Ultimately, we show that this way of working can bring together researchers to generate insights on the impacts of climate change, land use change, and urbanisation on the functioning of the ecosystem, providing scientific support for management and policy decisions.

Interannual rainfall extremes over southwest Western Australia (SWWA) are examined using observations, reanalysis data, and a long-term natural integration of the global coupled climate system. The authors reveal a characteristic dipole pattern of Indian Ocean sea surface temperature (SST) anomalies during extreme rainfall years, remarkably consistent between the reanalysis fields and the coupled climate model but different from most previous definitions of SST dipoles in the region. In particular, the dipole exhibits peak amplitudes in the eastern Indian Ocean adjacent to the west coast of Australia. During dry years, anomalously cool waters appear in the tropical/subtropical eastern Indian Ocean, adjacent to a region of unusually warm water in the subtropics off SWWA. This dipole of anomalous SST seesaws in sign between dry and wet years and appears to occur in phase with a large-scale reorganization of winds over the tropical/subtropical Indian Ocean. The wind field alters SST via anomalous Ekman transport in the tropical Indian Ocean and via anomalous air–sea heat fluxes in the subtropics. The winds also change the large-scale advection of moisture onto the SWWA coast. At the basin scale, the anomalous wind field can be interpreted as an acceleration (deceleration) of the Indian Ocean climatological mean anticyclone during dry (wet) years. In addition, dry (wet) years see a strengthening (weakening) and coinciding southward (northward) shift of the subpolar westerlies, which results in a similar southward (northward) shift of the rain-bearing fronts associated with the subpolar front. A link is also noted between extreme rainfall years and the Indian Ocean Dipole (IOD). Namely, in some years the IOD acts to reinforce the eastern tropical pole of SST described above, and to strengthen wind anomalies along the northern flank of the Indian Ocean anticyclone. In this manner, both tropical and extratropical processes in the Indian Ocean generate SST and wind anomalies off SWWA, which lead to moisture transport and rainfall extremes in the region. An analysis of the seasonal evolution of the climate extremes reveals a progressive amplification of anomalies in SST and atmospheric circulation toward a wintertime maximum, coinciding with the season of highest SWWA rainfall. The anomalies in SST can appear as early as the summertime months, however, which may have important implications for predictability of SWWA rainfall extremes.

Rice field has been traditionally considered as a nonpoint source of reactive nitrogen (N) for the environment, while it, with surrounding ditches and ponds, also contributes to receiving N inputs from atmosphere and waterbodies and intercepting N outputs from rice field. However, due to the paucity of multi-site, long-term observation and control experiment data, as well as robust process-based model for nitrogen budget, a comprehensive assessment of the N source (i.e., outputs > inputs) or sink (i.e., inputs > outputs) of rice field for the environment, from site to regional scale, is lacking. Here, a 2-year systematic observation and process-based simulations of N budget across China, covering typical annual rice cropping systems including single rice, double rice and rotation, were conducted to identify the roles of rice field in nitrogen cycling in China. The cost-benefit analysis for shifting China’s rice fields from nitrogen source to sink in different climatic scenarios (wet, normal and dry year) by innovative agricultural management, without compromising crop yield or soil fertility, were further evaluated. Rice fields, with surrounding ditches and ponds, perform as a nitrogen sink for atmosphere but a source for waterbody, which was confirmed by both the observational data and regional assessment by process-based model, regardless of rice types or climatic scenarios. With the adoption of sustainable N and water regulation measures, single rice field across China could be shifted to N sink (12.9-29.9 kg N ha-1) while the N source of double rice would be reduced by 51%-66%. For middle rice field (rotation with other crops), N sink would be achieved in dry year, while 31%-55% would shift to sink in wet and normal year. However, with such great benefits, the costs only accounted for 41%-49% of the expenditure for waste water treatment. Furthermore, through sorting all the measures and adopting economic ones (lower cost with higher benefit) in priority, we found that rice fields across China have great space for source-to-sink regulation with zero cost, which means the benefits would exceed the regulation costs, and even achieve overall N sink in dry year. Together these findings help us to update scientific knowledge to the role of rice fields in ecosystems, as well as highlight the significance and possibility for achieving environmental-friendly rice field, by improving agricultural management technologies.

Much of the prairie region in North America is characterized by relatively flat terrain with many depressions on the landscape. The hydrological response (runoff) is a combination of the conventional runoff from the contributing areas and the occasional overflow from the non-contributing areas (depressions). In this study, we promote the use of a hybrid modelling structure to predict runoff generation from prairie landscapes. More specifically, the Soil and Water Assessment Tool (SWAT) is fused with artificial neural networks (ANNs), so that SWAT and the ANN module deal with the contributing and non-contributing areas, respectively. A detailed experimental study is performed to select the best set of inputs, training algorithms and hidden neurons. The results obtained in this study suggest that the fusion of process-based and data-driven models can provide improved modelling capabilities for representing the highly nonlinear nature of the hydrological processes in prairie landscapes.Editor D. Koutsoyiannis; Associate editor L. See

<p>The recent boom in machine learning and data science has led to a number of new opportunities in the environmental sciences. In particular, climate models represent the best tools we have to predict, understand and potentially mitigate climate change, however these process-based models are incredibly complex and require huge amounts of high-performance computing resources. Machine learning offers opportunities to greatly improve the computational efficiency of these models.</p><p>Here we discuss our recent efforts to reduce the computational cost associated with running a process-based model of the physical ocean by developing an analogous data-driven model. We train statistical and machine learning algorithms using the outputs from a highly idealised sector configuration of general circulation model (MITgcm). Our aim is to develop an algorithm which is able to predict the future state of the general circulation model to a similar level of accuracy in a more computationally efficient manner.</p><p>We first develop a linear regression model to investigate the sensitivity of data-driven approaches to various inputs, e.g. temperature on different spatial and temporal scales, and meta-variables such as location information. Following this, we develop a neural network model to replicate the general circulation model, as in the work of Dueben and Bauer 2018, and Scher 2018.</p><p>We present a discussion on the sensitivity of data-driven models and preliminary results from the neural network based model.</p><p> </p><p><em>Dueben, P. D., & Bauer, P. (2018). Challenges and design choices for global weather and climate models based on machine learning. Geoscientific Model Development, 11(10), 3999-4009.</em></p><p><em>Scher, S. (2018). Toward Data‐Driven Weather and Climate Forecasting: Approximating a Simple General Circulation Model With Deep Learning. Geophysical Research Letters, 45(22), 12-616.</em></p>

Carbon dioxide (CO2) efflux from soil (or soil respiration, SR) is one of the most important yet variable characteristics of soil. When evaluating large areas, CO2 efflux modeling serves as a viable alternative to direct measurements. This research aims to identify site-specific differences and their effects on empirical CO2 efflux modeling. The experimental data from 25 years of field observations were utilized to identify the optimal site- and weather-specific models, parameterized for normal, wet, and dry years, for the forest and grassland ecosystems located on similar Entic Podzols (Arenic) in the same bioclimatic coniferous–deciduous forest zone. The following parameters were considered in the examined models: mean monthly soil or air temperatures (Tsoil and Tair), amount of precipitation during the current (P) and the previous (PP) months, and the storage of soil organic carbon (SOC) in the top 20 cm of soil. The weighted non-linear regression method was employed to estimate the model parameters for the normal, wet, and dry years. To increase the magnitude of the model resolutions, we controlled the slope and intercept of the linear model comparison between the measured and modeled data through the change in R0—CO2 efflux at Tsoil = 0 °C. The mean bias error (MBE), root-mean-square error (RMSE), and determination coefficient (R2) were employed to assess the quality of the model’s performance. The measured Tsoil, Tair, and P, as well as the litter (for forest) or sod (for grassland) horizon (modeled by the Soil SCLmate Statistical Simulator (SCLISS)), and soil temperatures (Tlit_m, Tsoil_m) and moistures (Mlit_m, Msoil_m), were used for SR simulation. For the CO2 efflux in the forest ecosystem with the lower SOC availability for mineralization, the direct Tsoil and Tair measurements in combination with SOC storage provided better parameterization for the empirical TPPC model. For the CO2 efflux in the grassland ecosystem with the high SOC availability for mineralization, the temperature became the governing factor, and the TPPrh model provided better performance over all the considered models. The model’s performance was the best for the wet years, and the worst for the dry years for both ecosystems. For forest ecosystems, the model performance for average precipitation years was equivalent to that in wet years. For grassland ecosystems, however, the model performance was equivalent to that in dry years due to differing exposure and hydrothermal regimes. The wet-year R0 obtained for both forest and grassland ecosystems differed from the normal- and dry-year values. The measured SR values relevant for the R0 estimations distribute along the precipitation range for the forest and along the temperature range for the grassland. The SCLISS-modeled Tlit_m and Mlit_m provide good alternatives to direct atmospheric measurements, and can be used as initial temperature and moisture data for CO2 efflux modeling when direct soil and moisture observations are not available on site.

Drainage in tropical peatlands increases CO2 emissions, the rate of subsidence, and the risk of forest fires, among other negative environmental impacts. These effects can be mitigated by raising the water table depth (WTD) using canal or ditch blocks. The performance of canal blocks in raising WTD is, however, poorly understood, because the WTD monitoring data is limited and spatially concentrated around canals and canal blocks. This raises the following question: how effective are canal blocks in raising the WTD over large areas? In this work we composed a process-based hydrological model to assess the rewetting performance of 168 canal blocks in a 22000 ha peatland area in Sumatra, Indonesia. We simulated daily WTD over one year using an existing canal block setup and compared it to the situation without blocks. The study was performed across two El-Niño Southern Oscillation (ENSO) scenarios, and four different peat hydraulic properties. Our simulations revealed that while canal blocks had a net positive impact on WTD rise, they lowered WTD in some areas, and the extent of their effect over one year was limited to a distance of about 600 m around the canals. We also show that canal blocks are most effective during dry periods and in peatlands with high hydraulic conductivity. Averaging over all modelled scenarios, blocks raised the annual mean WTD by only 0.9 cm. This value was 2.78 times larger in the dry year than in the wet year (1.39cm versus 0.50 cm), and there was a 2.76 fold difference between the scenarios with the maximum and minimum hydraulic conductivity (1.50 cm versus 0.54 cm). Using a linear relationship between WTD and CO2 emissions, we estimated that, averaging over peat hydraulic properties, canal blocks prevented the emission of 1.03 Mg ha-1 CO2 in the dry year and 0.37 Mg ha-1 CO2 in the wet year.

Saguling Reservoir has a potential to be used as a raw water supply for Bandung Metropolitan Area (BMA), with the discharge of 1.622 dm·s−1. However, further studies are needed to ensure that the water quality is in accordance with the government regulations. This study shows that the reservoir’s P concentration was 315.0 mg·m−3 on average in 1999–2013. This value only meets the class III of government standards of water quality for the cultivation of freshwater fish, livestock, and to irrigate landscaping, but does not belong to the class I standard of 200 mg·m−3 for drinking water. The total-P concentration in wet, normal, and dry years was 796.3, 643.8, and 674.8 mg·m−3, respectively. The pollution load was highest in wet years due to the high levels of sediment. The pollution load of the reservoir did not exceed the class III classification of 29 405.01 kg·year−1. The pollution load in wet, normal, and dry years was 38 790.1, 25 991.9, and 23 929.0 kg·year−1, respectively. The phosphorus pollution is caused by the use of floating net cages in the reservoir, which makes it difficult to meet the standards. In wet years, the pollution load was higher than in normal and dry years. The P load could be higher in the wet season due to dilution and could probably decrease the pollutant concentration in the reservoir.