1. Introduction
Weather modification and geoengineering represent the deliberate alteration of atmospheric and climate conditions, locally or globally, by humans using the available assets and resources based on the existing theoretical understanding of weather and climate processes (see [
1,
2,
3,
4,
5,
6] and references herein). Over the years, people have sought to modify the environment, including the atmosphere, in an attempt to adapt to its ever-changing conditions. However, the scientific-based stage of modification of the environment, first of all meteorological processes, only began in the middle of the 20th century when scientists at the General Electric Research Laboratories suggested using dry ice to disperse clouds [
7]. Since the early 1960s, dozens of projects have been implemented around the world aimed at modifying various atmospheric phenomena including different types of clouds, precipitation, lighting, hail, tornadoes, thunderstorms, hurricanes, and tropical cyclones. In the current century, the study and assessment of the human ability to modify environmental conditions has taken on a new sound in connection with climate change, which is caused by anthropogenic activities posing a serious threat to all of humanity [
8]. To mitigate the impact of climate change on nature and society, scientists have proposed the use of tools and methods of so-called geoengineering (e.g., [
9,
10,
11,
12,
13]).
However, we should make a difference between weather modification and geoengineering. Weather characterizes the current atmospheric conditions in a certain area or point at a particular time, or, in other words, the state of the atmosphere at some period of time is described by such atmospheric variables as temperature, pressure, humidity, wind speed, and direction, etc. Meanwhile, climate is the ensemble of states traversed by the Earth’s climate system over a sufficiently long temporal interval (~30 years). In this regard, the atmosphere, one of the five major components of the Earth’s climate system, is the most dynamic, unstable, and fastest-responding element of the climate system. The goal of weather modification is to change weather conditions over some limited geographical area or some geographical point. In other words, weather modification technologies are used to affect processes only in the atmosphere, which, as was stated above, is the most variable element of the climate system. In turn, geoengineering, being a planetary-scale process, is aimed at neutralizing anthropogenic radiative forcing and thereby reducing or even preventing human-caused warming of the Earth. Let us note that anthropogenic radiative forcing is produced mainly by greenhouse gases (carbon dioxide CO2, methane CH4, nitrous oxide N2O, and fluorinated gases) entering the atmosphere via burning fossil fuels and industrial and agricultural activities. However, both weather modification and geoengineering have one thing in common: these two procedures are goal-oriented processes implemented by means of external human-produced effects (interventions) to achieve specific objectives under various constraints. Thus, deliberate modification of weather and climate is, in substance, a dynamic optimization problem.
By viewing the atmosphere and/or the climate as a controllable dynamical system, we can approach weather modification and geoengineering from the perspective of optimal control theory. Within this conceptual and unified theoretical framework, the purpose of weather modification or geoengineering is formulated in terms of an extremal problem, which involves finding control functions and the corresponding climate (atmospheric) system trajectory that minimize or maximize a given objective function (also referred to as performance measure or index) subject to various constraints. In this instance, the atmospheric (climate) process in question is considered a closed-loop dynamical system, the evolution of which is described by the appropriate mathematical model, commonly represented by a set of differential equations. In this context, the human control actions can be described by variations in the model parameters selected on the basis of sensitivity analysis as control variables. This multi-disciplinary approach for planning and implementation of weather modification and geoengineering projects is known as geophysical cybernetics, the theoretical foundations of which were laid by the authors of this paper in the 1980s and ‘90s [
14].
This article is theoretical and methodological in nature aiming at generalizing and, to some extent, summarizing our previous and current research related to weather and climate modification and control. In the present paper, we first consider the deliberate change in weather and climate from an optimal control and dynamical systems perspective and, second, illustrate the application of this approach using a low-order conceptual model of the Earth’s climate system. For the sake of convenient interpretation, we provide some weather and climate basics, as well as giving a brief glance at control theory and sensitivity analysis of dynamical systems relevant to weather and climate control.
3. Illustrative Example
Holding the increase in global mean surface temperature “to well below 2 °C above pre-industrial levels and pursuing efforts to limit the temperature increase to 1.5 °C above pre-industrial levels” was designated in the 2015 Paris Agreement on Climate as a priority area for combating global warming. This ambitious goal is expected to be achieved through the transition to low-carbon development, which, on the one hand, is a vital necessity, and on the other, a serious challenge. Nevertheless, many countries are already in the transition to a low-carbon economy, developing national strategies to achieve the Paris Agreement goals using, in particular, the Shared Socioeconomic Pathways scenarios of projected social and economic worldwide changes up to 2100 [
29]. However, the Earth’s climate system possesses significant inertia (e.g., [
30]): it takes several decades for the climate system to reach a new equilibrium state in response to low (even zero) emissions of greenhouse gases (GHG). Therefore, after significant reduction of atmospheric GHG concentrations, surface temperature apparently will continue to rise. This phenomenon is known as “lag (time delays) between cause and effect”. Thus, geoengineering can be considered as one of the technologically feasible options to stabilize the climate. In this paper, we leave “behind brackets” of such important aspects of geoengineering as physical side effects and environmental risks, ethical, legal and other social issues, recognizing the importance of their consideration and assessment (e.g., [
31]).
Geoengineering can be implemented via human intervention in the redistribution of solar radiation flux due, for example, to the injection into the stratosphere of fine aerosol particles, which have the properties to scatter solar radiation in the visible spectral range and weakly absorb in the infrared range. For example, sulfate aerosols have such properties. Controlled emissions of sulfur dioxide or hydrogen sulfide (precursor gases) into the stratosphere ultimately lead to the formation of sulfate aerosol particles. The stratospheric aerosols increase the Earth’s planetary albedo
, change the radiation balance, and, as a consequence, decrease the Earth’s surface temperature. Sensitivity analysis shows that an increase in
by 1% leads to a decrease in the solar radiation flux at the top of the atmosphere by about 3.4 W/m
2, which is comparable with the radiation effect of doubling the concentration of atmospheric
. To assess the effectiveness of geoengineering projects and their consequences for nature and society, numerical modeling is used for given scenarios of anthropogenic GHG emissions and atmospheric concentrations, and heuristically specified geoengineering scenarios. It is obvious that going through all possible options of intentional geoengineering manipulations is an ineffective approach. In contrast to that, we consider climate manipulation within the framework of the optimal control theory. To illustrate this approach, we will apply the two-box low-parametric climate model, taking into account the radiation effects of the sulfate aerosols artificially injected into the stratosphere. The model equations are as follows [
32]:
where
and
are temperature anomalies for the upper (the atmosphere and mixing ocean layer) and lower (the deep ocean) boxes,
and
are the effective heat capacities for the upper and lower boxes,
is a climate feedback parameter,
is a coupling strength parameter describing the rate of heat loss by the upper box,
is the radiative forcing produced by GHG, and
is the radiative forcing generated by stratospheric aerosols.
The two-box model, in spite of its simplicity, is capable of simulating globally averaged climate change caused by human-induced radiative forcing with a reasonable accuracy (e.g., [
33,
34]). The following values of model parameters have been used in calculations [
31]:
,
,
and
. Radiative forcing
is approximated by a linear function
, where the parameter
η is determined from the Representative Concentration Pathway (RCP) data [
35]. For the worst-case emission scenario (RCP8.5 [
35]), the annual radiative forcing rate is
. Radiative forcing produced by aerosols is calculated by the formula:
, where
is the albedo of the aerosol layer (note that
),
Q0 = 342 W/m
2 is the mean insolation on the top of the Earth atmosphere. This allows albedo
to be considered as a control variable. However, in reality, we have the ability to control the rate of aerosol emissions
, which is included in the aerosol mass balance equation:
where
is the total mass of stratospheric aerosols,
is the aerosol emission rate and
is the residence time of stratospheric aerosol particles. The mass of aerosols
is linearly related to the albedo
:
, where
is the empirical coefficient,
is the mass extinction coefficient of aerosol particles,
is the Earth’s surface area.
In practice, precursor gases are injected into the stratosphere, therefore the mass of sulfate aerosols and their emission rate are expressed in sulfur units and denoted as
(TgS/year) and
(TgS), respectively, taking into account that 1 Tg of sulfur is equivalent to 4 Tg of aerosol particles. Then, Equation (17) can be rewritten as
where
.
If the optimal albedo of aerosol layer is determined, then the optimal emission rate , which provides the formation of an aerosol layer of the optimal mass , is calculated using Equation (18).
The optimal control problem will be considered with regard to the finite time interval
on which the dynamics of the control object are described by Equation (16) with given boundary conditions:
Thus, the left end of the phase trajectory is fixed, and the right end is fixed only for the variable
T, while the variable
is free. These boundary conditions are chosen since the surface temperature anomaly
T is of primary interest. The optimal control problem is formulated as follows:
on the finite time interval find the control variable belonging to an admissible value domain, so that when the dynamic constraints (16) and boundary conditions (19) imposed on the system are satisfied, the given functional characterizing the mass flow rate of aerosolshas reached its minimum value.
The terminal condition represents a target change in the global mean surface temperature at . As an example, we assume that °C. In essence, the performance index characterizes the consumption of aerosols for geoengineering manipulations since the albedo of the aerosol layer is a linear function of . So, we aim at minimizing the mass of aerosols required to achieve the target surface temperature change at final time which is fixed. The amount of aerosols that can be annually delivered to the stratosphere can be limited by the available technical capabilities. Therefore, we will formally assume that the domain of admissible controls is an interval , where is the maximum value of technically feasible albedo .
To solve the optimal control problem, we use the Pontryagin’s maximum principle (PMP), which is the major tool in optimal control. The Hamiltonian function used to solve the problem of optimal control for dynamical system (16) is given by:
where
and
are the time-varying Lagrange multipliers that satisfy the adjoint equations:
The optimal control
at each fixed time
must be such that
In other words, the optimal control results in a maximum value of
at any
The corresponding stationarity conditions for Hamiltonian yields:
To find the optimal control
and the optimal surface temperature anomaly
generated by
, we must solve the system of four ordinary Equations (16) and (21) in four unknown variables
, and
, with given initial and terminal conditions (18). Since no terminal condition is specified for the variable
, the following transversality condition
is used in calculations. We have derived the following analytical expressions for the optimal albedo of aerosol layer
and the corresponding optimal surface temperature anomaly
:
where
and
are the eigenvalues of the coefficient matrix of the adjoint system (22),
and
are the components of the corresponding eigenvectors,
,
,
, and
, are arbitrary integration constants, while
,
Considering geoengineering as a state-constrained optimal control problem
additional necessary conditions for optimality, the complementary slackness condition, must be specified. The meaning of the path constraint condition (28) is that the global mean surface temperature anomaly should not exceed the value of threshold parameter
, which is set a priori.
As an example, we consider the results of calculations for the RCP8.5 emission scenario (as we mentioned earlier, this is the most conservative scenario in relation to the growth of atmospheric GHG concentrations). The optimal control problem is examined on the finite time interval 2020–2100. In other words, and . Since the temperature anomalies are calculated relative to 2020, the initial conditions for variables and are as follows: and , where the numerical subscript is referred to the year 2020. We assume that:
- -
By 2020, the surface temperature anomaly would exceed the pre-industrial level by 1.1 °C, i.e., ΔT2020 = 1.1 °C
- -
By 2100, the surface temperature anomaly would exceed the pre-industrial level by 1.5 °C;
- -
For the 2020 to 2100 period of time, the increase in T should not exceed 2 °C above the pre-industrial level.
Then the allowable temperature growth by year 2100 relative to 2020 would be . This value is taken as a terminal condition for variable at . The threshold parameter, which defines a path constraint, is .
Calculations show that in the absence of deliberate interventions in the Earth’s climate system, the growth of global average surface temperature over the 80-year (2020–2100) interval is about 3.8 °C which is significantly higher than the level established by the Paris Agreement.
Figure 1 presents two curves, one for the optimal albedo of the aerosol cloud and the other for the corresponding surface temperature anomaly, as functions of time calculated for the case (a) when no constraints are imposed on the control variable and the surface temperature anomaly, and (b) with constraint imposed on the surface temperature anomaly. As can be seen from
Figure 1b, in the absence of limitations on the surface temperature increase, beginning from 2060, some overshoot beyond the threshold
is observed. However, stratospheric aerosols, artificially injected into the stratosphere, guarantee compliance with the specified terminal condition. Recall that this condition is as follows:
at
.
The optimal control problem with state constraint requires the consideration of an additional necessary condition termed as a condition of complementary slackness:
, where
is the Lagrange multiplier. For the case of
(see above), the optimal albedo and the corresponding surface temperature anomaly are shown in
Figure 1 in red color. As can be seen from this figure, by using state constraint, it is possible to prevent an undesirable growth of surface temperature within a given time interval. The amount of aerosols required over an 80-year period to maintain geoengineering is 36.5 TgS for the unconstraint case and 73.6 TgS for the case with state constraint. For RCP6.0 scenario (this scenario is described by the Intergovernmental Panel on Climate Change as an intermediate scenario in which emissions’ peak is around 2080, then decline) we obtained that the total amount of aerosols is 17.0 Tg for the unconstrained optimal control problem, and 23.3 Tg for the problem with state constraint.
4. Discussion
In response to human-induced climate change, which is one of the main threats to natural and anthropogenic systems in the 21st century, the scientific meteorological community has proposed several technologies known as geoengineering to decelerate global warming and stabilize the climate. However, all of these technologies, including solar radiation management (SRM) considered in this paper, are still theoretical in nature since “Playing with the Earth’s climate is a dangerous game with unclear rules” [
36]. Nevertheless, scientists from different countries continue geoengineering research using mainly computer simulations expecting that geoengineering techniques could potentially be helpful in the future to enhance ongoing efforts to mitigate global warming by reducing anthropogenic GHG emissions. Meanwhile, in the vast majority of studies, a heuristic approach is used to developing geoengineering strategies and scenarios (e.g., [
11,
37,
38,
39,
40] and references therein). This is because global climate models used to project climate change are extremely complex. This important circumstance is a serious obstacle to the application of optimization methods, and, in particular, optimal control in geoengineering problems. However, a heuristic technique is not guaranteed to be optimal and rational, but this approach is sufficient for obtaining approximate (satisfactory) solutions in the case when finding an optimal solution is impossible. As a result, theoretical studies of geoengineering operations are usually performed outside the framework of optimization theory. Note that the optimal (best) solution is a solution that is preferable for one reason or another. More specifically, the optimal decision (option, choice, etc.) is the best decision among admissible alternatives if there is a rule of preference for one over the other, known as the optimality criterion. One can make a comparative assessment of possible decisions (alternatives) and choose from the best based on such criterion. In our case, optimality criterion is the total mass of aerosols injected into the stratosphere.
In relation to the problem considered in this paper, the commonly used procedure for determining the albedo of aerosol layer (and, consequently, the total mass of aerosol particles injected into the stratosphere) required to counteract temperature rise is, in a simplified manner, as follows. First, we select the GHG emissions scenario of interest (e.g., RCP8.5) and then calculate the corresponding radiative forcing due to changes in concentrations of the relatively well-mixed GHG, using, for example, the first-order approximation expression [
41]:
, where
is the empirical parameter,
is the
-equivalent concentration in parts per million by volume at time
t,
is the reference concentration. Second, we assume that stratospheric aerosols, partially reflecting solar radiation back to space, produce radiative forcing
, where the parameter
defines the portion of anthropogenic radiative forcing that should be neutralized by stratospheric aerosols. Note that if the parameter
is set to 1, then anthropogenic radiative forcing will be completely neutralized. Next, we calculate the albedo of aerosol layer as follows (see
Section 3):
. According to this equation, the albedo
is linearly related to the radiation forcing
. The corresponding aerosol emission rate at time
t can be found using the Equation (18). The solution found in such a manner is in a certain sense non-optimal, although the SRM problem is essentially mathematically solved. With the parameter values used in this paper (see
Section 3), we found that for the RCP8.5 scenario, assuming that the anthropogenic radiative forcing will be completely compensated by stratospheric aerosols (
), and also assuming that
, the albedo
should increase linearly from zero in 2020 to ~0.02 in 2100.
The study of geoengineering operations within the framework of the optimal control theory, which is a branch of mathematical optimization, allows for obtaining, in a certain sense, the optimal solution, considering various constrains, which, of course, must be formalized mathematically in the form of equalities and/or inequalities. We emphasize that the dynamics of the system are defined as optimal only with respect to the selected performance index. Note that the goal-setting problem (formulation of the performance index) is non-trivial and requires special consideration.
In the present study, a very simple climate model, the two-box energy balance model (EBM) is used, allowing for predicting the globally averaged surface temperature from the analysis of the planetary energy balance. The model is linear and does not describe the climate system dynamics. A primary motivation of using such a model is that similar models have been examined in a number of research papers studying the essential features of climate system response to natural and human-induced radiative perturbations that affect the Earth’s climate system. Despite its simplicity, the two-box EBM was able to reproduce the evolution of global mean surface temperature over time in response to time-dependent radiative forcing with reasonable accuracy [
33]. To demonstrate the applicability of new mathematical approaches (in our case, the optimal control theory) to “real world problems”, the use of simple modeling tools is a common and useful practice. However, caution should be exercised in evaluating and interpreting the results obtained using this approach. We considered the applicability of optimal control theory to hypothetical deliberate weather and climate modification and, in particular, to the SRM approach of geoengineering. An algorithm for solving the problem of optimal control of the Earth’s climate using classical Pontryagin’s maximum principle was presented, both with and without constraints imposed on the state variable. In fact, the Earth’s climate is a highly complex nonlinear dynamical system with feedbacks and cycles that affect the climate response to forcing generated by GHG [
4]. However, nonlinear problems of the Earth’s climate system optimal control can be solved mainly numerically using highly complicated coupled general circulation models of the atmosphere and ocean. Undoubtedly, such problems are of extreme complexity. Therefore, we believe that the application of optimal control theory in combination with simple atmospheric and climate models will be very useful in the design of weather and climate control systems, as well as in the development of scenarios for intentional modifications of climate and weather.
It is important to note that two classical mathematical tools for studying optimally control systems are Pontryagin’s maximum principle and Bellman’s Dynamic programming. However, the application of these methods is very difficult or even impossible if nonlinear models of high complexity are considered. In order to overcome the difficulty, some asymptotic and approximate approaches can be used (e.g., [
42,
43,
44]).
It should be underlined that the results obtained in this paper serve only to illustrate the applicability of optimal control theory to climate modification problems since the use of very simple models of climate systems allows one to solve the optimal control problem analytically. In the next studies, we intend to apply the considered approach to solving the problems of modifying various atmospheric processes and phenomena, as well as to examine a number of geoengineering problems using more complex climate models.