Climate and long-term weather forecasts based on self-organizing AI modeling technologies from geospatial BigData from Satellite Earth Observation (SatEO), Ocean data, Ground Truth data help making smarter decisions in many areas, because the ability to make more reliable predictions about the future is far advanced than the physical and statistical models.

Today, in short to long term weather forecasting, there is a large predictive information gap between existing short-term local weather forecasts (few days ahead) and long-term global climate projections (30+ years ahead).

Our CLIMFOR system, which is build by our original Model Self-Organizing technologies, generates daily, sub-seasonal, seasonal, or decadal spatially high-resolved forecasts from Satellite Earth Observation (SatEO), Ocean data, and Ground Truth data for smart decision making, taking into account that the planning horizon of weather dependent economy, politics and society is on the order of months to several years.

The CLIMFOR Concept - Creating DIGITAL TWIN OF THE ATMOSPHERE

The ability to use DATA for modeling and reliable predictions about the future is far advanced than the physical and statistical models.

According to the CLIMFOR concept, in order to model all the necessary atmospheric parameters for a specific region of interest over time, the atmosphere above the earth surface is mapped with a virtual 3D grid. The idea is to construct this grid that constitutes of various nodes and each single node of this grid will run our original Core Modeling technology based on the SatEO data of the geolocation which this node represents, to autonomously develop predictive models of the complex atmosphere dynamics.

For each grid node, the CLIMFOR system will constitute three levels of self-organization and knowledge extraction from data, and it generates probabilistic ex-ante forecasts of the node’s represented atmosphere column for various atmosphere parameters such as air and sea surface temperature, pressure, humidity, ozone, aerosols, water vapor and GHG concentrations, irradiation and others.

Currently, developments are underway for the full 3D grid node system covering larger parts of the earth´s atmosphere. But with the advances already present with geopredict, a location specific forecast with different time horizons is already successfully implemented.

Real-world processes are complex, interdependent systems coupled with uncertainty. In many cases, it is impossible to create analytical models using classical theoretical systems analysis or common statistical methods since there is incomplete knowledge of these processes involved. Environmental, medical and socio-economic systems are but three examples. In contrast, inductive models obtained by self-organization are derived from real physical observation data and represent the relationships implicit within the system without or with only little knowledge of the physical processes or mechanisms involved.

Model self-organization extracts the information and knowledge necessary to build up a model from the provided data, only. It therefore transforms information into knowledge.

Self-organizing modeling works on adaptive networks. Self-organization is considered in identifying connections between the network units by a learning mechanism to represent discrete items. For this approach, the objective is to estimate networks of relevant and sufficient size with a structure evolving during the estimation process. A process is said to undergo self-organization if identification emerges through the system's environment.

geopredict’s Core Modeling implements a unique and original solution of multi-level inductive self-organization to accomplish the task of modeling for various problems and purposes easily, objectively, and autonomously.

Self-organization of optimal complex models from data along with analytical descriptions in implicit and explicit format.

03 |

Interdependent Systems of Equations

Real-world systems are usually high-dimensional. They are described by a number of system parameters. The atmosphere, for example, consists of nitrogen, oxygen, ozone, aerosols, water vapor, CO2 and other substances and is influenced by the oceans and cosmic forces like sun irradiance. These system parameters of the atmosphere can be divided into internal state (endogenous) variables of the system and external (exogenous) variables such as the influence of the sun.

Endogenous and exogenous parameters may depend one from another at time t and/or at certain times t-n in history (lags) so that they build a complex interdependence structure of the system. This interdependence structure, however, is not given a priori but is self-organizing from data. Linear, time invariant, dynamic systems are described mathematically as systems of algebraic or difference equations as follows:

with x - endogenous variables, u - exogenous variables, e - error vector, t - time, and L1, L2 - max. time lags (memory).

Interdependent systems of equations obtained by multi-level model self-organization and other innovative features implemented in our Core Modeling technology run in every single grid node of the CLIMFOR system and drive our forecasting services in climate, energy, agriculture, or marine.

Example of an interdependent system of equations for multi-step ex ante forecasting adding another level of self-organization.