Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
Methods and systems for generating a multi-model ensemble of global climate simulation data from a plurality of global climate simulation models (GCMs), to be used in training a neural network (NN)-based climate forecasting model, are disclosed. The methods and systems perform steps of computing a GCM validation measure for each GCM; selecting a validated subset of the GCMs, by comparing each computed GCM validation measure to a validation threshold determined based on observational historical climate data; computing a forecast skill score for each validated GCM, based on a first forecast function; selecting a validated and skillful subset of GCMs; generating one or more candidate ensembles by combining simulation data from at least two validated and skillful GCMs; computing an ensemble forecast skill score for each candidate ensemble, based on a second forecast function; and selecting a best-scored candidate ensemble. Embodiments of the present invention enable accurate climate forecasting without the need to run new dynamical global climate simulations on supercomputers.
1. A method for generating a multi-model ensemble of global climate simulation data from a plurality of pre-existing global climate simulation model (GCM) datasets, and for training a neural network (NN)-based climate forecasting model, comprising:
pre-training the NN-based climate forecasting model using reanalysis data or pre-existing image recognition data, wherein the NN-based climate forecasting model comprises a predictive neural network;
computing a GCM dataset validation measure, for each of the plurality of pre-existing GCM datasets, based on at least one sample statistic for at least one climate variable from the pre-existing GCM dataset;
selecting a validated subset of the plurality of pre-existing GCM datasets, by comparing each computed GCM dataset validation measure to a validation threshold determined based on a set of observational historical climate data;
computing a forecast skill score for each validated GCM dataset based on the NN-based climate forecasting model;
selecting a validated and skillful subset of GCM datasets by choosing at least two best-scored GCM datasets;
generating one or more candidate ensembles of global climate simulation data, by combining at least two validated and skillful GCM datasets from the validated and skillful subset of GCM datasets;
computing an ensemble forecast skill score for each candidate ensemble of global climate simulation data based on the NN-based climate forecasting model;
generating the multi-model ensemble of global climate simulation data by selecting a best-scored candidate ensemble of global climate simulation data; and
training the NN-based climate forecasting model using the multi-model ensemble of global climate simulation data generated from the plurality of pre-existing GCM datasets.
2. The method of claim 1 , wherein the NN-based climate forecasting model is selected from the group consisting of a Convolutional Recurrent Neural Network (CRNN) having at least one Long Short-Term Memory (LSTM) layer, a Spherical Convolutional Neural Network (S 2 -CNN), and a transformer neural network.
3. The method of claim 1 , wherein the validation threshold is determined by computing the at least one sample statistic for the at least one climate variable from the set of observational historical climate data.
4. The method of claim 1 , wherein the at least one climate variable is an input climate variable to the NN-based climate forecasting model, and wherein the NN-based climate forecasting model forecasts the target output variable at a target lead time from the input climate variable.
5. The method of claim 1 , wherein the observational historical climate data comprises reanalysis data.
6. The method of claim 1 , wherein the combining the at least two validated and skillful GCM datasets comprises at least one of concatenating and interleaving the at least two validated and skillful GCM datasets.
7. A system for generating a multi-model ensemble of global climate simulation data from a plurality of pre-existing global climate simulation model (GCM) datasets, and for training a neural network (NN)-based climate forecasting model, comprising:
at least one processor; and
a non-transitory physical storage medium for storing program code and accessible by the processor, the program code when executed by the processor causes the processor to:
pre-train the NN-based climate forecasting model using reanalysis data or pre-existing image recognition data, wherein the NN-based climate forecasting model comprises a predictive neural network;
compute a GCM dataset validation measure, for each of the plurality of pre-existing GCM datasets, based on at least one sample statistic for at least one climate variable from the pre-existing GCM dataset;
select a validated subset of the plurality of pre-existing GCM datasets, by comparing each computed GCM dataset validation measure to a validation threshold determined based on a set of observational historical climate data;
compute a forecast skill score for each validated GCM dataset based on the NN-based climate forecasting model;
select a validated and skillful subset of GCM datasets by choosing at least two best-scored GCM datasets;
generate one or more candidate ensembles of global climate simulation data, by combining at least two validated and skillful GCM datasets from the validated and skillful subset of GCM datasets;
compute an ensemble forecast skill score for each candidate ensemble of global climate simulation data based on the NN-based climate forecasting model;
generate the multi-model ensemble of global climate simulation data by selecting a best-scored candidate ensemble of global climate simulation data; and
train the NN-based climate forecasting model using the multi-model ensemble of global climate simulation data generated from the plurality of pre-existing GCM datasets.
8. The system of claim 7 , wherein the NN-based climate forecasting model is selected from the group consisting of a Convolutional Recurrent Neural Network (CRNN) having at least one Long Short-Term Memory (LSTM) layer, a Spherical Convolutional Neural Network (S 2 -CNN), and a transformer neural network.
9. The system of claim 7 , wherein the validation threshold is determined by computing the at least one sample statistic for the at least one climate variable from the set of observational historical climate data.
10. The system of claim 7 , wherein the at least one climate variable is an input variable to the NN-based climate forecasting model, and wherein the NN-based climate forecasting model forecasts the target output variable at a target lead time from the input climate variable.
11. The system of claim 7 , wherein the observational historical climate data comprises reanalysis data.
12. The system of claim 7 , wherein the program code to combine the at least two validated and skillful GCM datasets comprises program code for at least one of concatenating and interleaving the at least two validated and skillful GCM datasets.
13. A non-transitory physical storage medium for generating a multi-model ensemble of global climate simulation data from a plurality of pre-existing global climate simulation model (GCM) datasets, and for training a neural network (NN)-based climate forecasting model, the storage medium comprising program code stored thereon, that when executed by a processor causes the processor to:
pre-train the NN-based climate forecasting model using reanalysis data or pre-existing image recognition data, wherein the NN-based climate forecasting model comprises a predictive neural network;
compute a GCM dataset validation measure, for each of the plurality of pre-existing GCM datasets, based on at least one sample statistic for at least one climate variable from the pre-existing GCM dataset;
select a validated subset of the plurality of pre-existing GCM datasets, by comparing each computed GCM dataset validation measure to a validation threshold determined based on a set of observational historical climate data;
compute a forecast skill score for each validated GCM dataset based on the NN-based climate forecasting model;
select a validated and skillful subset of GCM datasets by choosing at least two best-scored GCM datasets;
generate one or more candidate ensembles of global climate simulation data, by combining at least two validated and skillful GCM datasets from the validated and skillful subset of GCM datasets;
compute an ensemble forecast skill score for each candidate ensemble of global climate simulation data based on the NN-based climate forecasting model;
generate the multi-model ensemble of global climate simulation data by selecting a best-scored candidate ensemble of global climate simulation data; and
train the NN-based climate forecasting model using the multi-model ensemble of global climate simulation data generated from the plurality of pre-existing GCM datasets.
14. The non-transitory physical storage medium of claim 13 , wherein the NN-based climate forecasting model is selected from the group consisting of a Convolutional Recurrent Neural Network (CRNN) having at least one Long Short-Term Memory (LSTM) layer, a Spherical Convolutional Neural Network (S 2 -CNN), and a transformer neural network.
15. The non-transitory physical storage medium of claim 13 , wherein the validation threshold is determined by computing the at least one sample statistic for the at least one climate variable from the set of observational historical climate data.
16. The non-transitory physical storage medium of claim 13 , wherein the at least one climate variable is an input climate variable to the NN-based climate forecasting model, and wherein the NN-based climate forecasting model forecasts the target output variable at a target lead time from the input climate variable.
17. The non-transitory physical storage medium of claim 13 , wherein the observational historical climate data comprises reanalysis data.
18. The non-transitory physical storage medium of claim 13 , wherein the combining the at least two validated and skillful GCM datasets comprises at least one of concatenating and interleaving the at least two validated and skillful GCM datasets.