Final Report Summary - CLIMITS (Performance and usefulness of CLImate predictions: Beyond current liMITationS)
Over the past 25 years, the modeling community achieved steady progress in dynamical climate predictions, with skill level that is now considered useful for some societal applications at the seasonal time-scale. However, major factors such as coupled models errors, initialization strategies and unconstrained physical parameterizations are still substantially limiting predictability, particularly over land areas. Long-term improvements in climate predictions must necessarily come by (i) improved understanding and description of the physical processes through dedicated process studies and observations. In the meanwhile, (ii) the multi-model approach can be used combining the imperfect models available to enhance predictions. Progresses in both objectives (i) and (ii) need more international collaborative efforts.
By implementing collaboration between European and Asian-Pacific climate prediction communities, this project will largely contribute to objective (ii). In fact, five European and eleven Asian Pacific climate prediction systems will be collected to form a grand multi-model. The maximum level of multi-model prediction performance currently attainable will be assessed. Innovative techniques will be developed in order to evaluate achievable and attained forecast skill over land and the related multi-model capability to enhance usefulness.
The project will contribute to objective (i) by improving land surface-vegetation representation in at least one of the multi-model components (i.e: IPRC global climate model). In particular, efforts will be made on assessing the impact of improved land surface on simulation and prediction of the Indian summer monsoon (ISM) and then, in turn, on those of Euro-Mediterranean climate during boreal summer, which is related to global teleconnection patterns emanating from the ISM.
The CLIMITS project requires a preliminary phase in which a global array of relevant up-to-date state-of-the-art datasets is acquired and analyzed (WP1). Concurrently, the grand ENSEMBLES/APCC multi-model needs to be collected and analyzed (WP2) and the improved land modeling implemented in the IPRC coupled model (WP3) so that to evaluate the improvements and the potential benefit gained by users. WP4 studies the simulation and predictability of the relationship between Indian summer monsoon and Euro-Mediterranean region. Finally, WP5 focuses on the evaluation of forecast improvements in terms of forecast probabilistic quality and potential economical value to end-users.
Work performed and achievements
The one-year return phase has been completed successfully and the unprecedented model developments and knowledge acquired from the international collaborative effort established during the first year between ENEA (Italian Institution representing the European climate prediction community) and IPRC (U.S. Institution representing the Asian-Pacific prediction community), has been greatly exploited. By gathering independent and complementary tools and competences we obtained significant progresses in climate-prediction research and with improved usefulness of the forecasts to end-users community.
Work Package 1 (WP1): A global array of relevant up-to-date state-of-the-art datasets has been acquired, pre-processed and analyzed. The comprehensive dataset has been analyzed to characterize the land variability as a function of the space and time scales leading to improved understanding of the relationship and feedbacks between land and climate. Using the coupled manifold technique (CMT), the relationship and the coupling between the acquired surface variables has been assessed with particular focus to the Boreal Summer (June-July-August-September, JJAS) interannual anomalies.
Furthermore, by taking advantage of the new global array of relevant up to date state-of-the-art datasets, the analysis of the reciprocal forcing between precipitation and LAI seasonal-mean anomalies previously performed by Alessandri and Navarra (2008) has been substantially extended.
Work Package 2 (WP2): In collaboration with Asia-Pacific Economic Community Climate Center (APCC), the hindcasts from the Asian Pacific (CliPAS/APCC) and European (ENSEMBLES) communities has been collected into a grand Multi-Model Ensemble (MME) covering the period 1983-2005. The latest update includes 11 Seasonal Prediction Systems (SPS) coming from CliPAS/APCC MME and the 5 SPS from ENSEMBLES MME.
Probabilistic performance measures have been implemented and applied to the retrospective forecasts collected from APCC-CliPAS and ENSEMBLES. To assess the potential usefulness of probabilistic forecasts we applied the Potential Economic Value metric by assuming a simple cost/loss model related to binary events. The results show that the grand APCC-ENSEMBLES MME improves the overall probabilistic accuracy and the potential economical value compared to the contributing MMEs, both over sea and over land. The analysis of all possible MME combinations shows that the maximum achievable skill is always obtained by mixing ENSEMBLES and CliPAS/APCC models.
Considering the six ENSEMBLES/APCC seasonal prediction systems that provided an extended retrospective forecast period (1960-2001), the interdecadal changes in the Northern Hemisphere seasonal climate predictability and skill has been investigated. Two decadal transitions of the climate predictability have been identified in response to the changes in the ENSO phenomenon occurring (i) in late 1970’s and (ii) in the winter 1988/1989. Both transitions lead to increased predictability and forecasting skill towards the more recent decades. A relationship between biases in the western pacific warm pool and performance over South and East Asia of the grand ENSEMBLES-CliPAS/APCC MME has been identified. Based on this relationship it is shown that the MME built from the less biased models is able to enhance significantly the seasonal prediction skill.
A generalized multivariate regression method based on the CMT is applied to the grand ENSEMBLES-CliPAS/APCC MME to decompose climate anomalies over land into a predictable component that is a remotely-forced from the ocean and a predictable component that is locally forced. The predictable component of the observed temperature variability appears to be strongly forced by remote teleconnections from the ocean. Overall, it is estimated that, globally, 38% of the land variability can be predicted by using surface temperature over ocean as the predictor. In particular, the predictable component exceeds 50% of the total variance over the Sahel and central-eastern Africa, South-East Asia, Indian Peninsula, Southern North America, Amazon basin and eastern Brazil. On the other hand, 14% of the global temperature predictability originates locally.
Noteworthy, the application of the CMT leads to the exploitation of the performance over land, which represents a considerable improvement compared with the relatively poor original seasonal APCC-ENSEMBLES forecasts there. The CMT is therefore proposed as a promising tool for the statistical calibration of the prediction signal over land and with expected impact on the end-users community.
Work Package 3 (WP3): The atmospheric component (Echam5) of the IPRC global climate model is improved to include the Land Surface Model (LSM) ORCHIDEE, which replaces the simpler land surface scheme included in Echam5. A procedure to suitably initialize the ORCHIDEE model from the initial state taken from observationally based data has been implemented. To replace the fields simulated by ORCHIDEE at the start date of the forecasts, we rescaled and calibrated the soil moisture and snow fields obtained from Era-Interim Reanalysis following the method suggested by Randall Koster in a recently published paper.
New sets of enhanced climate simulations and retrospective forecasts with improved land representation and initialization. For each starting date of the retrospective forecasts, an ocean IC and an ensemble of five atmospheric initial states were created. Starting from these ICs, the four sets of retrospective forecasts have been integrated, both for coupled mode (real predictions) and for prescribed SST case (potential predictions), for six months, producing eight sets of five-member ensemble forecasts covering the period 1983–2012. The four sets of hindcasts are as follows: i) Echam5 not coupled with improved land model (no-Land experiment), (ii) Echam5 coupled with the improved land model ORCHIDEE (Krinner et al., 2005; Land experiment), (iii) ORCHIDEE included and initialized using ERA-Interim soil moisture and snow depth (Land-ini experiment) and (iv) ORCHIDEE included but using, for each year, land IC selected randomly (Land-random experiment) from the pool in (iii).
Work Package 4 (WP4): The observed and simulated Mediterranean (MED) climate in both historical and 21st century climate projection has been characterized. A probabilistic approach has been developed to quantitatively address how and why the geographic distribution of MED will change based on the latest-available climate projections for the 21st century. Our analysis provides, for the first time, a robust assessment of significant northward and eastward future expansions of MED over both the Euro-Mediterranean and western North America. The unprecedented probabilistic information and the robust quantitative assessment of MED changes provided by this study can be profitably used by decision makers and end-users as well as researchers from a broad spectrum of disciplines when evaluating the future consequences on ecosystems and human activities as well as the possible adaptation policies.
Dry summers over the eastern Mediterranean are characterized by strong descent anchored by long Rossby waves, which are forced by diabatic heating associated with summer monsoon rainfall over South Asia (so called monsoon-desert mechanism). The ability of the available models in representing the physical processes involved in this mechanism for both historical and 21st century projections is evaluated and a subset of models that are able to capture the monsoon-desert mechanism for correct reasons is identified. In these models, the projected changes in subsidence over the Mediterranean can be related to the projected changes in rainfall over South Asia. The identified models project an increase of summer precipitation over South Asia in the 21st century. Contemporarily, over the Mediterranean the maximum of subsidence is projected to move westward, toward the center of the basin. The identification of these processes and the models' performance are favorable for the advances in climate sciences, including the understanding of these dynamical processes and their predictability. The Grand ENSEMBLES/APCC MME is exploited to evalutate the relationship between the skill over the Euro-Mediterranean sector and the capability of the forecasts to represent the Indian Summer Monsoon interannual variability. The forecasts skill in Precipitation over the Euro-Mediterranean region for all the possible combinations of models coming from APCC and ENSEMBLES MMEs is evaluated as a function of the skill in representing the Indian Monsoon. The results show that the capability of the models to forecast the interannual anomalies in the Indian monsoon circulation and convective activity is very important to have a good skill over the Euro-Mediterranean domain. For both temperature and precipitation we found a clear relationship between skill in Indian Monsoon activity and the probabilistic performance over the Euro-Mediterranean domain. This suggests that the monsoon-desert mechanism is indeed an important source of predictability for the Euro-Mediterranean domain during boreal summer.
Work Package 5 (WP5):
The grand multi-model approach has led to a considerable improvement of the probabilistic performance for the prediction of above/below normal temperature. Compared with the maximum performance of the contributing ENSEMBLES and APCC multi-model ensembles (MMEs), the grand APCC/ENSEMBLES MME shows a significant gain in skill, with enhancements exceeding 40% over some mid to high latitude areas in both winter and summer. This result confirms that by combining independent Seasonal Prediction Systems such as the ones coming from European and Asian-Pacific communities can lead to considerable skill amplification. It is of particular relevance to note that the combination of models from both ENSEMBLES and CliPAS/APCC improves the maximum performance over the Euro-Mediterranean by 50% (25%) compared to the best combinations obtained from ENSEMBLES (CliPAS/APCC) models only.
The capability of the grand APCC/ENSEMBLES MME to provide valuable information to end-users is evaluated by computing the Potential Economical Value (PEV). It is shown that the regions with higher BSS indeed tend to correspond with regions with the higher PEV. Even if most of the positive PEV are located over the sea, useful PEV for the end users are revealed over some land areas over Euro-Mediterranean, Middle East, South East Asia and western U.S. in Boreal Summer and Africa, South America and Maritime Continent in Boreal Winter.
The capability of the available boreal-summer climate forecasts to provide valuable information for the prediction of the electricity load in Italy is evaluated with details for the period 1990-2007. The results show that, especially in the Center-South of Italy, seasonal forecasts of temperature issued in May lead to a significant prediction skill of electricity demand for the subsequent season. The use of Multi-Model seasonal climate predictions can improve significantly the Potential Economic Value of the forecasts of electricity demand. In this respect, it is shown that the PEV of the Grand APCC/ENSEMBLES MME are largely positive, therefore revealing the potential for the useful application for the electricity management in Italy in collaboration with the Italian TSO, Transmission System Operator (TERNA SpA).
The inclusion of the improved land representation and initialization in the IPRC seasonal forecasting system enhances the probabilistic performance and PEV for the prediction of above/below normal temperature and precipitation over Amazon basin, North America, Western Africa, South East Asia, Mediterranean and Middle East. The analysis shows that the initialization of the land improves significantly the seasonal forecast skill in several regions covering North America, eastern Europe and Western Russia, Middle East and South East Asia. Interestingly, compared with the effect of improving the land surface model, which showed much of the improvement over Tropics, the land initialization improves the probabilistic forecasting skill mostly over mid-to-high latitudes. This indicates an increasingly important effect of land initialization and local memory at higher latitudes.
By implementing collaboration between European and Asian-Pacific climate prediction communities, this project will largely contribute to objective (ii). In fact, five European and eleven Asian Pacific climate prediction systems will be collected to form a grand multi-model. The maximum level of multi-model prediction performance currently attainable will be assessed. Innovative techniques will be developed in order to evaluate achievable and attained forecast skill over land and the related multi-model capability to enhance usefulness.
The project will contribute to objective (i) by improving land surface-vegetation representation in at least one of the multi-model components (i.e: IPRC global climate model). In particular, efforts will be made on assessing the impact of improved land surface on simulation and prediction of the Indian summer monsoon (ISM) and then, in turn, on those of Euro-Mediterranean climate during boreal summer, which is related to global teleconnection patterns emanating from the ISM.
The CLIMITS project requires a preliminary phase in which a global array of relevant up-to-date state-of-the-art datasets is acquired and analyzed (WP1). Concurrently, the grand ENSEMBLES/APCC multi-model needs to be collected and analyzed (WP2) and the improved land modeling implemented in the IPRC coupled model (WP3) so that to evaluate the improvements and the potential benefit gained by users. WP4 studies the simulation and predictability of the relationship between Indian summer monsoon and Euro-Mediterranean region. Finally, WP5 focuses on the evaluation of forecast improvements in terms of forecast probabilistic quality and potential economical value to end-users.
Work performed and achievements
The one-year return phase has been completed successfully and the unprecedented model developments and knowledge acquired from the international collaborative effort established during the first year between ENEA (Italian Institution representing the European climate prediction community) and IPRC (U.S. Institution representing the Asian-Pacific prediction community), has been greatly exploited. By gathering independent and complementary tools and competences we obtained significant progresses in climate-prediction research and with improved usefulness of the forecasts to end-users community.
Work Package 1 (WP1): A global array of relevant up-to-date state-of-the-art datasets has been acquired, pre-processed and analyzed. The comprehensive dataset has been analyzed to characterize the land variability as a function of the space and time scales leading to improved understanding of the relationship and feedbacks between land and climate. Using the coupled manifold technique (CMT), the relationship and the coupling between the acquired surface variables has been assessed with particular focus to the Boreal Summer (June-July-August-September, JJAS) interannual anomalies.
Furthermore, by taking advantage of the new global array of relevant up to date state-of-the-art datasets, the analysis of the reciprocal forcing between precipitation and LAI seasonal-mean anomalies previously performed by Alessandri and Navarra (2008) has been substantially extended.
Work Package 2 (WP2): In collaboration with Asia-Pacific Economic Community Climate Center (APCC), the hindcasts from the Asian Pacific (CliPAS/APCC) and European (ENSEMBLES) communities has been collected into a grand Multi-Model Ensemble (MME) covering the period 1983-2005. The latest update includes 11 Seasonal Prediction Systems (SPS) coming from CliPAS/APCC MME and the 5 SPS from ENSEMBLES MME.
Probabilistic performance measures have been implemented and applied to the retrospective forecasts collected from APCC-CliPAS and ENSEMBLES. To assess the potential usefulness of probabilistic forecasts we applied the Potential Economic Value metric by assuming a simple cost/loss model related to binary events. The results show that the grand APCC-ENSEMBLES MME improves the overall probabilistic accuracy and the potential economical value compared to the contributing MMEs, both over sea and over land. The analysis of all possible MME combinations shows that the maximum achievable skill is always obtained by mixing ENSEMBLES and CliPAS/APCC models.
Considering the six ENSEMBLES/APCC seasonal prediction systems that provided an extended retrospective forecast period (1960-2001), the interdecadal changes in the Northern Hemisphere seasonal climate predictability and skill has been investigated. Two decadal transitions of the climate predictability have been identified in response to the changes in the ENSO phenomenon occurring (i) in late 1970’s and (ii) in the winter 1988/1989. Both transitions lead to increased predictability and forecasting skill towards the more recent decades. A relationship between biases in the western pacific warm pool and performance over South and East Asia of the grand ENSEMBLES-CliPAS/APCC MME has been identified. Based on this relationship it is shown that the MME built from the less biased models is able to enhance significantly the seasonal prediction skill.
A generalized multivariate regression method based on the CMT is applied to the grand ENSEMBLES-CliPAS/APCC MME to decompose climate anomalies over land into a predictable component that is a remotely-forced from the ocean and a predictable component that is locally forced. The predictable component of the observed temperature variability appears to be strongly forced by remote teleconnections from the ocean. Overall, it is estimated that, globally, 38% of the land variability can be predicted by using surface temperature over ocean as the predictor. In particular, the predictable component exceeds 50% of the total variance over the Sahel and central-eastern Africa, South-East Asia, Indian Peninsula, Southern North America, Amazon basin and eastern Brazil. On the other hand, 14% of the global temperature predictability originates locally.
Noteworthy, the application of the CMT leads to the exploitation of the performance over land, which represents a considerable improvement compared with the relatively poor original seasonal APCC-ENSEMBLES forecasts there. The CMT is therefore proposed as a promising tool for the statistical calibration of the prediction signal over land and with expected impact on the end-users community.
Work Package 3 (WP3): The atmospheric component (Echam5) of the IPRC global climate model is improved to include the Land Surface Model (LSM) ORCHIDEE, which replaces the simpler land surface scheme included in Echam5. A procedure to suitably initialize the ORCHIDEE model from the initial state taken from observationally based data has been implemented. To replace the fields simulated by ORCHIDEE at the start date of the forecasts, we rescaled and calibrated the soil moisture and snow fields obtained from Era-Interim Reanalysis following the method suggested by Randall Koster in a recently published paper.
New sets of enhanced climate simulations and retrospective forecasts with improved land representation and initialization. For each starting date of the retrospective forecasts, an ocean IC and an ensemble of five atmospheric initial states were created. Starting from these ICs, the four sets of retrospective forecasts have been integrated, both for coupled mode (real predictions) and for prescribed SST case (potential predictions), for six months, producing eight sets of five-member ensemble forecasts covering the period 1983–2012. The four sets of hindcasts are as follows: i) Echam5 not coupled with improved land model (no-Land experiment), (ii) Echam5 coupled with the improved land model ORCHIDEE (Krinner et al., 2005; Land experiment), (iii) ORCHIDEE included and initialized using ERA-Interim soil moisture and snow depth (Land-ini experiment) and (iv) ORCHIDEE included but using, for each year, land IC selected randomly (Land-random experiment) from the pool in (iii).
Work Package 4 (WP4): The observed and simulated Mediterranean (MED) climate in both historical and 21st century climate projection has been characterized. A probabilistic approach has been developed to quantitatively address how and why the geographic distribution of MED will change based on the latest-available climate projections for the 21st century. Our analysis provides, for the first time, a robust assessment of significant northward and eastward future expansions of MED over both the Euro-Mediterranean and western North America. The unprecedented probabilistic information and the robust quantitative assessment of MED changes provided by this study can be profitably used by decision makers and end-users as well as researchers from a broad spectrum of disciplines when evaluating the future consequences on ecosystems and human activities as well as the possible adaptation policies.
Dry summers over the eastern Mediterranean are characterized by strong descent anchored by long Rossby waves, which are forced by diabatic heating associated with summer monsoon rainfall over South Asia (so called monsoon-desert mechanism). The ability of the available models in representing the physical processes involved in this mechanism for both historical and 21st century projections is evaluated and a subset of models that are able to capture the monsoon-desert mechanism for correct reasons is identified. In these models, the projected changes in subsidence over the Mediterranean can be related to the projected changes in rainfall over South Asia. The identified models project an increase of summer precipitation over South Asia in the 21st century. Contemporarily, over the Mediterranean the maximum of subsidence is projected to move westward, toward the center of the basin. The identification of these processes and the models' performance are favorable for the advances in climate sciences, including the understanding of these dynamical processes and their predictability. The Grand ENSEMBLES/APCC MME is exploited to evalutate the relationship between the skill over the Euro-Mediterranean sector and the capability of the forecasts to represent the Indian Summer Monsoon interannual variability. The forecasts skill in Precipitation over the Euro-Mediterranean region for all the possible combinations of models coming from APCC and ENSEMBLES MMEs is evaluated as a function of the skill in representing the Indian Monsoon. The results show that the capability of the models to forecast the interannual anomalies in the Indian monsoon circulation and convective activity is very important to have a good skill over the Euro-Mediterranean domain. For both temperature and precipitation we found a clear relationship between skill in Indian Monsoon activity and the probabilistic performance over the Euro-Mediterranean domain. This suggests that the monsoon-desert mechanism is indeed an important source of predictability for the Euro-Mediterranean domain during boreal summer.
Work Package 5 (WP5):
The grand multi-model approach has led to a considerable improvement of the probabilistic performance for the prediction of above/below normal temperature. Compared with the maximum performance of the contributing ENSEMBLES and APCC multi-model ensembles (MMEs), the grand APCC/ENSEMBLES MME shows a significant gain in skill, with enhancements exceeding 40% over some mid to high latitude areas in both winter and summer. This result confirms that by combining independent Seasonal Prediction Systems such as the ones coming from European and Asian-Pacific communities can lead to considerable skill amplification. It is of particular relevance to note that the combination of models from both ENSEMBLES and CliPAS/APCC improves the maximum performance over the Euro-Mediterranean by 50% (25%) compared to the best combinations obtained from ENSEMBLES (CliPAS/APCC) models only.
The capability of the grand APCC/ENSEMBLES MME to provide valuable information to end-users is evaluated by computing the Potential Economical Value (PEV). It is shown that the regions with higher BSS indeed tend to correspond with regions with the higher PEV. Even if most of the positive PEV are located over the sea, useful PEV for the end users are revealed over some land areas over Euro-Mediterranean, Middle East, South East Asia and western U.S. in Boreal Summer and Africa, South America and Maritime Continent in Boreal Winter.
The capability of the available boreal-summer climate forecasts to provide valuable information for the prediction of the electricity load in Italy is evaluated with details for the period 1990-2007. The results show that, especially in the Center-South of Italy, seasonal forecasts of temperature issued in May lead to a significant prediction skill of electricity demand for the subsequent season. The use of Multi-Model seasonal climate predictions can improve significantly the Potential Economic Value of the forecasts of electricity demand. In this respect, it is shown that the PEV of the Grand APCC/ENSEMBLES MME are largely positive, therefore revealing the potential for the useful application for the electricity management in Italy in collaboration with the Italian TSO, Transmission System Operator (TERNA SpA).
The inclusion of the improved land representation and initialization in the IPRC seasonal forecasting system enhances the probabilistic performance and PEV for the prediction of above/below normal temperature and precipitation over Amazon basin, North America, Western Africa, South East Asia, Mediterranean and Middle East. The analysis shows that the initialization of the land improves significantly the seasonal forecast skill in several regions covering North America, eastern Europe and Western Russia, Middle East and South East Asia. Interestingly, compared with the effect of improving the land surface model, which showed much of the improvement over Tropics, the land initialization improves the probabilistic forecasting skill mostly over mid-to-high latitudes. This indicates an increasingly important effect of land initialization and local memory at higher latitudes.