Dynamic Task and Resource Scheduling Towards Green Space-Air-Ground-Sea Integrated Network

Dynamic Task and Resource Scheduling Towards Green Space-Air-Ground-Sea Integrated Network Yufei Ye1 , Shijian Gao2 , Xinhu Zheng1,2 , and Liuqing Yang1,2

arXiv:2605.01414v1 [cs.NI] 2 May 2026

1

Intelligent Transportation Thrust, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China 2 Internet of Things Thrust, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China Email: [email protected], {shijiangao, xinhuzheng, lqyang}@hkust-gz.edu.cn

Abstract—In the context of 6G ubiquitous connectivity, the space–air–ground–sea integrated network (SAGSIN) emerges as a new paradigm to provide critical services for resource-limited ocean environments. To realize this paradigm efficiently, we propose an innovative dynamic task and resource scheduling approach for green SAGSIN that delivers computing support for vessels while minimizing overall task execution delay. To address the challenge of multi-layer task scheduling, a layer-wise task offloading algorithm is developed specifically for SAGSIN. It adapts to real-time, multi-dimensional system dynamics and integrates an anticipatory handover strategy that adaptively controls the amount of data offloaded to the satellite, thereby preventing post-handover congestion while improving satellite resource utilization. Furthermore, the bandwidth allocation of uncrewed aerial vehicles and base station, UAV trajectories, and computing resource allocation are jointly optimized to enhance connectivity among low-altitude devices and facilitate demand-driven resource allocation for green network development. Simulation results verify that the proposed method better adapts to dynamic system resources and achieves at least a 23% reduction in average task delay compared with benchmarks. Index Terms—Space-air-ground-sea integrated network, task offloading, efficient scheduling, edge computing, satellite handover.

I. I NTRODUCTION Driven by the vision of 6G ubiquitous connectivity, expanding maritime activities, such as environmental monitoring and resource exploration, create growing demands for communication and computing resources [1]. However, the limited computing power of vessels and the scarce maritime infrastructure hinder the support for maritime applications. Leveraging the flexible mobility and deployment, uncrewed aerial vehicles (UAVs), which are key enablers of the emerging low-altitude economy (LAE), have attracted attention as a means of providing computing support to vessels [2] and vehicles [3]. However, due to the compact design, UAVs have limited computing capabilities and struggle to support intensive computational tasks. In this context, the space-air-ground-sea integrated network (SAGSIN) has emerged as a promising paradigm for maritime edge computing [4]. It leverages the complementary advantages of heterogeneous networks and integrates their resources to provide computing services for vessels, thereby alleviating the energy and computational burden of onboard devices. When UAVs are assisted by coastal base stations (BSs), maritime task execution delay can be minimized by optimizing task offloading ratios [5], [6] and caching decisions [7]. Leveraging the wide coverage and strong computing power of low earth orbit (LEO) satellites, UAV-satellite collaborative computing frameworks have been designed to save energy for the entire system [8] or for UAVs alone [9]. A scheme for tasks with different delay sensitivities was proposed in [10], and a satellite-assisted harvesting-and-offloading method was developed to support UAVs in [11]. Moreover, [12] designed a three-tier computing system that augments satellites with base

stations. Serving as relay nodes between the space and sea layers, high-altitude platforms (HAPs) can effectively mitigate the high path loss of satellites. They were adopted in [13] and [14] to further reduce task costs. Meanwhile, task offloading methods for improving energy efficiency and communication security in SAGSIN were studied in [15] and [16], respectively. Despite the potential of SAGSIN, most existing studies simplify the network architecture to cope with the high complexity of multi-layer task offloading. This simplification may lead to inefficient resource utilization and load imbalance. Furthermore, fixed offloading and resource allocation strategies fail to adapt to rapid changes in system states during data transmission, such as fluctuations in available computing and bandwidth resources, which further degrades system performance. Additionally, the limited service time of fast-moving satellites and the critical issue of inter-satellite handover remain underexplored. To overcome these limitations, this paper proposes a novel dynamic task and resource scheduling scheme (DASH) for green SAGSIN. Unlike static or heuristic offloading methods, DASH introduces a low-complexity, layer-wise dynamic task offloading algorithm based on backpressure routing theory. This algorithm adaptively adjusts fine-grained task scheduling decisions according to real-time system conditions, including available computing resources of heterogeneous servers, task congestion levels, link capacities, and network topology. To address the underexplored handover problem, DASH further incorporates an anticipatory strategy that proactively regulates satellite offloading traffic based on the joint states of both the current and incoming satellites, thereby preventing post-handover congestion while improving satellite resource utilization. Beyond task offloading, we jointly optimize UAV-BS bandwidth allocation, UAV trajectory planning, and computing resource allocation. This strengthens connectivity between vessels and low-altitude UAVs and boosts task completion through the co-design of resource scheduling and physical network architecture, a dimension largely overlooked in prior SAGSIN research. Through adaptive multi-layer load balancing and demand-driven resource allocation, DASH provides efficient computing services while significantly improving resource utilization efficiency and the ecological stability of the system, laying a crucial foundation for green and sustainable SAGSIN development. Simulation results demonstrate that DASH outperforms benchmarks in adapting to variations in computing and bandwidth resources, achieving at least a 23% reduction in average task delay while effectively alleviating post-handover satellite congestion. II. S YSTEM M ODEL AND P ROBLEM F ORMULATION In this section, we first outline the proposed SAGSIN system model, which encompasses the network modeling, communi-

Inter-Satellite Link H2S Link U2H Links V2U Links V2B Links

Task data of different vessels

Current Satellite

HAP

Incoming Satellite

Buffers on each server

UAVs

Vessel Users Base Station

Fig. 1. Illustration of the multi-layer network architecture for SAGSIN.

cation modeling, and computation and task queuing modeling. Then we present the problem formulation and detail the overarching decomposition framework to facilitate problem-solving. A. Network Model Fig. 1 depicts the proposed system, comprising V vessels and U UAVs, denoted by set V and U , respectively, along with a coastal BS, an HAP, a current and an incoming LEO satellite. We discretize time into slots, each indexed by n with length of τ . The horizontal locations of vessel v, UAV u, BS, and HAP are denoted by Wv (n), Wu (n), Wb , and Wh (n), with the height of 0, Hu , Hb , and Hh , respectively. The displacement of UAV u between adjacent time slots is ∆du (n) = ∥Wu (n)−Wu (n−1)∥. The size and computational density of vessel v’s task data are Dv bits and Cv cycles/bit, respectively. In each time slot, each vessel selects an UAV or BS to offload a portion of task data. Each UAV and HAP offload part of the selected vessel’s data to HAP and satellite, respectively. Meanwhile, the BS, UAVs, HAP, and satellite compute buffered data from different vessels. B. Communication Model Following [8], [10], the vessel v-to-UAV u (V2U) and vessel v-to-BS (V2B) links are modeled as the combination L0 L0 of path loss hL v,i (n) = dv,i (n)2 = Hi 2 +∥Wi (n)−Wv (n)∥2 and q q K0 1 Rician fading hR v,i (n) = 1+K0 + 1+K0 ov,i (n), i ∈ {u, b}, where dv,i (n) is the distance, L0 is reference path loss, ov,i (n) ∈ CN (0, 1), and K0 is Rician factor. The channel gain R 2 is given by hv,i (n) = hL v,i (n)|hv,i (n)| . For the link from UAV u-to-HAP (U2H), we denote the total path loss defined in [14] by Lu,h (n). The channel gain from HAP-to-satellite ηh,s (n)

(H2S) is modeled as hh,s (n) = K0 10− 10 Gh Gs , where ηh,s (n) signifies H2S large-scale fading [14], Gh and Gs are their respective antenna gains. Employing orthogonal frequency division multiple access (OFDMA) [5], [14], the data rates of the V2U/B, U2H, and H2S links  are respectively  Pv (n)hv,i (n) expressed as Rv,i (n) = Bv,i (n) log2 1 + N , 0 Bv,i (n)   −Lu,h (n)/10 , and Ru,h (n) = Bh log2 1 + Pu (n)Gu GNh010Bh   Ph (n)hh,s (n) Rh,s (n) = Bs log2 1 + . Pv (n), Pu (n), and N 0 Bs Ph (n) are the transmit power of vessel v, UAV u, and HAP. Bv,i (n), Bh , Bs denote the bandwidth of each V2U/B, U2H,

and H2S link. Gu is UAV u’s antenna gain. With ultra-high-rate laser communications, transmission delay of the inter-satellite links (ISLs) is negligible and propagation delay is assumed as Tp [17]. We focus on adaptive backlog control at handover and leave the study on ISL to future works. C. Computation and Task Queue Models At time slot n, the available computing resource (i.e., maximum processable CPU cycles) of server i is Fi (n), i ∈ I = {u, b, h, s}. The task data P backlog on vessel v is given by tx Qv,0 (n) = Qv,0 (n − 1) − u∈U av,u (n − 1)Dv,u (n − 1) − tx tx tx av,b (n − 1)Dv,b (n − 1), where Dv,u and Dv,b are data amount offloaded to UAV u and BS, av,u and av,b are binary indicators of vessel v’s offloading decision. The backlog of vessel v’s tx (n − 1) − data on server i is Qv,i (n) = Qv,i (n − 1) + Dv,i comp tx ′ Dv,i (n − 1) − 1{i∈{u,h}} · Dv,i,i′ (n − 1), i ∈ {h, s}, where tx denotes the total data volume of vessel v received by server Dv,i comp i, Dv,i is the data volume of vessel v computed by server tx tx ′ i, and Dv,i,i ′ (n) = (1) av,u,h (n)Dv,u,h (n), (i, i ) = (u, h), tx ′ (2) av,h,s (n)Dv,h,s (n), (i, i ) = (h, s), signifying the data of vessel v offloaded from UAV u to HAP or HAP to satellite. av,u,h = 1 indicates UAV u chooses to offload vessel v’s data to HAP and av,h,s = 1 means HAP offloads vessel v’s data to satellite. The actual amount of transmitted data is bounded by link capacity and data backlog at each device. D. Problem Formulation We represent the total execution delay of vessel v’s task as P Tv = (Nv + 1) · τ , where Nv Psatisfies Qv,0 (Nv ) + i∈I Qv,i (Nv ) > 0 and Qv,0 (Nv +1)+ i∈I Qv,i (Nv +1) = 0 simultaneously, indicating that its task data is just cleared from all queues at time slot Nv + 1. Our objective is to minimize the total task execution delay of all vessels by jointly optimizing the multi-layer task offloading decisions a, UAV-BS bandwidth allocation B, computing resource allocation D comp , and UAV trajectory WU . The overall problem can be formulated as P0 : Overall Problem X min Tv a,B, D comp ,WU v∈V

s.t.

av,u (n), av,b (n), av,u,h (n), av,h,s (n) ∈ {0, 1}, ∀v, u, n, (1a) X av,u (n) + av,b (n) = 1, ∀v, n, (1b) u∈U

X

av,i (n)Bv,i (n) ≤ Bi (n), ∀i ∈ {u, b}, n, (1c)

v∈V

Bv,i (n) ≥ 0, ∀i ∈ {u, b}, v, n, X comp Dv,i (n)Cv ≤ Fi (n), ∀i, n,

(1d) (1e)

v∈V comp tx Dv,i (n) ≤ Qv,i (n) + Dv,i (n),

∀v, i ∈ {b, s}, n, (1f) comp tx tx Dv,i (n) ≤ Qv,i (n) + Dv,i (n) − Dv,i,i′ (n), ∀v, (i, i′ ) ∈ {(u, h), (h, s)}, n, comp Dv,i (n) ≥ 0, ∀v, i, n,

(1g) (1h)

∆du (n)/τ ≤ Su,max , ∀u, n, ∥Wu (n) − Wu′ (n)∥ ≥ dsaf e , ∀u, u′ ∈ U, u ̸= u′ , ∀n.

(1i) (1j)

Regarding the constraints, (1a) and (1b) ensure the binary indicators and each vessel offloads data to one UAV or BS in each slot, respectively. (1c) and (1d) guarantee the bandwidth allocated to vessels is non-negative and bounded by total available spectrum of each UAV and BS. (1e) ensures the total allocated computing resources of each server do not exceed its available capacity. (1f)-(1h) guarantee the computed data size is bounded by available backlog and non-negative. (1i) enforces that the motion of UAV u is bounded by its maximum speed Su,max . (1j) is UAV collision-avoidance constraint, where dsaf e denotes minimum safe distance among UAVs. To tackle this mixed-integer nonlinear programming (MINLP) problem in dynamic SAGSIN, we first develop an efficient multi-layer task offloading scheme adaptive to real-time system conditions, and then iteratively optimize multidimensional system resources via block coordinate descent (BCD) framework. III. S ATELLITE H ANDOVER -AWARE DYNAMIC TASK O FFLOADING A LGORITHM FOR M ULTI -L AYER SAGSIN Task offloading in SAGSIN involves twofold challenges: the abrupt shift in satellite resources caused by handover and the high complexity of task scheduling in dynamic multi-layer networks. To address these issues, this section first introduces the anticipatory strategy for satellite handover and then elaborates on the dynamic task offloading algorithm for SAGSIN. A. Anticipatory Satellite Handover Strategy The satellite offloading strategy based solely on the current satellite state can be myopic. If incoming satellite has significantly lower computing capacity, aggressive offloading before handover will lead to congestion afterwards. Conversely, if it offers stronger computing power, such strategies may conservative. An adaptive strategy accounting for both satellites is essential to smooth such abrupt changes. To this end, we design the real-time joint states of current and incoming satellites to guide offloading decision-making, adaptively regulating the data offloaded to satellites in advance of handover. A pre-handover period τhand is set prior to handover time Ts , as illustrated in Fig. 2 (a). Once system time T (n) enters this period, the joint states of available computing resources and backlog for satellites are respectively defined as   Ts − T (n) Ts − T (n) e Fs (n) = Fs (n) + 1 − Fs′ (n), (2a) τhand τhand   e v,s (n)= Ts − T (n) Qv,s (n)+ 1− Ts − T (n) Qv,s′ (n), (2b) Q τhand τhand where Fs′ (n) and Qv,s′ (n) are available computing resources and vessel v’s backlog at time slot n on the incoming satellite. They are dynamic weighted sums of states for current and incoming satellites, where the weight of incoming one increases as handover approaches. These real-time joint states are employed to guide the task offloading strategy, thereby smoothing the transition and enhancing satellite resource utilization. B. Dynamic Task Scheduling in Multi-Layer SAGSIN The multi-layer architecture of SAGSIN leads to prohibitive complexity in optimizing task offloading decisions. To address this challenge, we leverage backpressure (BP) routing theory to decompose the combinatorial problem into elegant layer-wise task scheduling, thereby reducing complexity while adapting to multidimensional system dynamics.

Anticipatory Handover Strategy

Graph Model of SAGSIN Real Communication Links Virtual Sink Nodes (VSN) Virtual Computation Links

Service Period of Current Satellite: �� Pre-Handover Period: ����� Weight for State of Incoming LEO

0

Starting Time System Starting of Pre-Handover Period Time

V1

Weight for State of Current LEO

Current Time: �(�) = ��

��

Satellite Handover Time

(a)

�

V1 V2

�1,� �1,1

�1,2 �2,�

� �2,2 2,1

B U1 U2

��1,�

��2,� � �1,1

��2,1

�1,ℎ �2,ℎ ��1,2 ��2,2

��1,ℎ

H

V2

��2,ℎ

(VSN V1)

��1,�

�ℎ,�

��2,�

S

(VSN V2)

(b)

Fig. 2. Satellite handover-aware multi-layer task offloading scheme: (a) Illustration of real-time joint satellite state for anticipatory handover strategy. (b) An example of graph model containing 2 vessel users, 2 UAVs, 1 BS, 1 HAP, and 1 LEO satellite, represented by node V, U, B, H, and S, respectively.

To align with routing theory, the network is remodeled into the graph shown in Fig. 2 (b). A virtual sink node (VSN) is established for each vessel v. We first homogenize the communication and computing capabilities, laying the foundation for designing the informative task scheduling indicators. We define the rate of each real communication link (i.e., rv,u , rv,b , ru,h , and rh,s ) as the corresponding channel capacity R times τ , and the rate of virtual computation link between server i and VSN c v as rv,i = Fi /Cv . In this way, the rate re of each link e is unified as the number of transmitable/computable bits per time slot. The distance of each link e is we = r · rmax /re , where r and rmax are average and maximum link rates in the system. To guide data towards servers with low congestion and abundant resources, we define the real-time Pressure Index (PI) of vessel v and server i regarding vessel v’s data as ( min Jv,0 (n) = Qv,0 (n) + wv,0 (n), (3) min Jv,i (n) = Qv,i (n) + wv,i (n), min min where wv,0 (n) and wv,i (n) are the shortest distances from that node to VSN v. The PI serves as an informative indicator that integrates task congestion level, computing capacity, rate of links to high-capacity servers, and network topology. As the driving force for offloading, the Pressure Differential (PD) of vessel v’s data on links between vessel v and UAV u/BS, UAV u and HAP, HAP and satellite are respectively given by   ∆Jv→i (n) = Jv,0 (n) − Jv,i (n), ∀i ∈ {u, b}, ∆Jv,u→h (n) = max{Jv,u (n) − Jv,h (n), 0}, (4)  ∆J v,h→s (n) = max{Jv,h (n) − Jv,s (n), 0}.

Then, the offloading decision of each vessel v is given by av,i (n)∗ = argmaxav,i (n)T [∆Jv→i (n) ⊙ rv,i (n)] ,

(5)

i∈{u,b}

where av,i (n) = [av,1 (n), . . . , av,U (n), av,b (n)]T subject to (1b), ∆Jv→i (n) = [∆Jv→1 (n), . . . , ∆Jv→U (n), ∆Jv→b (n)]T , and rv,i (n) = [rv,1 (n), . . . , rv,U (n), rv,b (n)]T . This mechanism leads each vessel to select the UAV or BS offering the best combination of low load and high throughput, thereby avoiding server overload while fully exploiting high-rate channels. The offloading decision of each UAV u and HAP is expressed as av,i,i′ (n)∗ = argmaxav,i,i′ (n)T ∆Jv,i→i′ (n),

(6)

v∈V

where ∆Jv,i→i′ (n) = [∆J1,i→i′ (n), . . . , ∆JV,i→i′ (n)]T and av,i,i′ (n) = [a1,i,i′ (n), . . . , aV,i,i′ (n)]T subject to P ′ ′ v∈V av,i,i (n) ≤ 1, (i, i ) ∈ {(u, h), (h, s)}. Formula (6)

indicates that each UAV or HAP selects a vessel with the most urgent offloading demand (e.g., severe backlog or the possibility of quickly reaching upper-layer server with more resources) and transmits its data to HAP or satellite. By shifting from global coordination to layer-wise pressuredriven forwarding, our scheme reduces the complexity from exponential level to approximate O(V U ). Furthermore, since the PI is updated at each time slot based on instantaneous system states, it possesses strong adaptability to the high dynamics of SAGSIN environments. IV. J OINT UAV-BS BANDWIDTH , UAV T RAJECTORY, AND C OMPUTING R ESOURCE O PTIMIZATION Building upon the layer-wise task scheduling strategy derived from a macroscopic perspective of network topology and system states, this section focuses on the fine-grained orchestration of underlying resources to further enhance the system performance. Specifically, we jointly optimize the inter-coupled UAV-BS bandwidth allocation, UAV trajectories, and computing resource allocation via a BCD framework. A. Bandwidth Allocation Policy of UAVs and BS This subproblem aims to maximize the total transmission rate of communication links between each UAV/BS and the vessels it serves by optimizing bandwidth allocation, thereby boosting the volume of offloaded vessel data and reducing overall task delay. This subproblem can be formulated as P1 : UAV and BS Bandwidth Allocation   X X Pv (n)hv,i (n) av,i (n)Bv,i (n) log2 1 + max Bv,i (n)N0 Bv,i (n) i∈{u,b} v∈V

s.t.

(1c) and (1d),

With determined vessel offloading decisions, the objective function and constraints are strictly concave w.r.t. Bv,i (n). Thus, P1 is a convex optimization problem. The optimal bandwidth allocation of BS and UAVs are expressed as Pv (n)hv,i (n) Bi (n), v ∈ Vi (n), i ∈ {b, u}, v∈Vi (n) Pv (n)hv,i (n) (7) where Vi (n) signifies set of vessels associated with server i. Proofs and derivations are omitted due to space limitation. ∗ Bv,i (n) = P

B. UAV Trajectory Optimization Beyond resource dimension, the mobility of UAVs provides a crucial degree of freedom to dynamically reconstruct the network topology. By optimizing their trajectories, we directly modulate the vessel-to-UAV channel states. Specifically, the trajectory planning aims to strategically position UAVs to maximize the aggregate data rate and throughput for associated vessels while strictly adhering to kinematic and safety constraints. Accordingly, this subproblem is formulated as P2 : UAV Trajectory Optimization   XX Pv (n)hv,u (n) max av,u (n)Bv,u (n) log2 1 + Bv,u (n)N0 Wu (n) u∈U v∈V

s.t. (1i) and (1j). The non-convexity of the objective function and limit (1j) poses the primary challenge. To tackle this, we employ the successive convex approximation (SCA) technique. By applying

Algorithm 1 Computing Resource Allocation Algorithm. c Input: V, Qv,i (n), Fi (n), rv,i (n), Dv , and Cv . Output: Computing resource allocation strategy Fv,i (n) = {Fv,i (n)|∀v} for server i. 1: Initialize the pending vessel set Vp = V, the assigned vessel set Va = ∅, and the remaining computing resources Fire (n) = Fi (n). 2: while Vp ̸= ∅ do 3: Allocate computing resource to each vessel v ∈ Vp based on c Qv,i (n)/rv,i (n) · Fire (n). Fv,i (n) = P c Qv,i (n)/r (n) v∈Vp

v,i

max 4: if Fv,i (n) ≤ Fv,i (n), ∀v ∈ Vp then 5: Obtain the final solution Fv,i (n) = {Fv,i (n)|∀v}. 6: break 7: else max 8: for v ∈ Vp such that Fv,i (n) > Fv,i (n) do max 9: Set its Fv,i (n) ← Fv,i (n). 10: Move this vessel v from Vp to Va . 11: end for P 12: Update Fire (n) ← Fi (n) − v∈Va Fv,i (n). 13: end if 14: end while

the first-order Taylor expansion of data rate Rv,u (n) at the given (k) point ϕv,u (n) = ∥Wu(k) (n) − Wv (n)∥2 in the k-th iteration, we can obtain its global lower-bound as (k) (k) ev,u (n) = Rv,u R (n) + ∇Rv,u (n)(ϕv,u (n) − ϕ(k) v,u (n)),

(8)

(k)

(k)

where Rv,u (n) and ∇Rv,u (n) are the data rate and first-order derivative of Rv,u (n) w.r.t. ϕv,u (n) at k-th iteration, respectively. Similarly, the square of inter-UAV distance ∥Wu (n) − Wu′ (n)∥2 is globally lower bounded by first-order Taylor (k) (k) expansion at given point ψu,u′ = Wu(k) (n) − Wu′ (n) as 2 (k) (k) T (k) deu,u′ (n) = ∥ψu,u′ ∥2 +2ψu,u′ ((Wu (n) − Wu′ (n)) − ψu,u′ ). (9) Then, we convert P2 into a series of approximated convex subproblems expressed as P2′ : Convex Approximation Problem of P2 X X ev,u (n) max R Wu (n)

u∈U v∈Vu (n)

s.t. (1i), 2 deu,u′ (n) ≥ dsaf e 2 , ∀u, u′ ∈ U, u ̸= u′ ,

(10)

which can be solved by CVX solver iteratively until converging to a near-optimal solution. This ensures that the positions of UAVs evolve synergistically with vessel mobility and dynamic task demands, effectively bridging the gap between topology optimization and physical-layer connectivity. C. Computing Resource Allocation As the final stage of task execution pipeline in SAGSIN, dynamic allocation of computing resources governs the terminal processing rate for each vessel’s data. Denoting the computing resource that server i allocates to vessel v at time slot n as Fv,i (n) (CPU cycles), constraints (1e)-(1h) can be rewritten as X Fv,i (n) ≤ Fi (n), ∀i, (11a) v∈V max 0 ≤ Fv,i (n) ≤ Fv,i (n), ∀v, i,

(11b)

where we denote maximum required resources of vessel v at max tx server i as Fv,i (n) = Cv · (Qv,i (n) + Dv,i (n) − 1{i∈{u,h}} ·

Algorithm 2 Overall Dynamic Task and Resource Scheduling Method for Green SAGSIN. Input: The optimized UAV positions Wu (n − 1) in previous time slot and all remaining parameters in current time slot. Output: Optimization results a(n), B(n), D comp (n), and WU (n). 1: Update backlog of all devices or set Qv,0 (0) = Dv and Qv,i (0) = 0 at initial time slot, and initialize iteration number j = 0. 2: Obtain a(n) through the method detailed in Section III. 3: repeat 4: Solve P1 to obtain B(n)j+1 with given WU (n)j and D comp (n)j . 5: Solve P2′ to obtain WU (n)j+1 with given B(n)j+1 and D comp (n)j . 6: Execute Algorithm 1 to obtain D comp (n)j+1 with given B(n)j+1 and WU (n)j+1 . 7: Update the value of objective function. 8: Update j = j + 1. 9: until The objective value converges or j > jmax .

tx Dv,i,i ′ (n)), ∀i. Each server should allocate resources based on backlog of vessels in its buffer while meeting above limits. To achieve this, we design a demand-driven computing resource allocation method as detailed in Algorithm 1. Only the key parameters of this subproblem are listed in the input for brevity. Computing resources are allocated in proportion to completion time of each vessel’s data on this server, thereby prioritizing vessels with severe backlog while avoiding resource waste. comp ∗ After gaining the solution, Dv,i (n) = Fv,i (n)/Cv , ∀v, i. The steps of overall approach is outlined in Algorithm 2. At the beginning, we initialize system parameters and update backlog queues of all vessels and servers. Then the multilayer task offloading decisions are obtained through the method detailed in Section III. According to dependencies among subproblems, we iteratively optimize the UAV-BS bandwidth allocation, UAV trajectory planning, and computing resource allocation in sequence, until convergence or the maximum number of iterations is reached. The superscript j denotes the solution at the start of the j-th iteration.

V. S IMULATION R ESULTS In this section, we conduct simulation experiments to evaluate the performance of the proposed method. We consider a 2 km×2 km ocean area and uniformly deploy six UAVs with maximum speed of 15 m/s and minimum safe distance of 5 m [10]. Vessels are randomly distributed, with the velocity ranging from 5 to 15 m/s [14]. A coastal BS is deployed at the position of (0, 1km) and HAP maintains a quasi-stationary hover at center of the region [13]. Similar to [8], [14], the key parameter setup is summarized in Table I. To validate the effectiveness of the proposed DASH method, we compare it with the following benchmark schemes: • DASH w/o HO: The proposed DASH method without anticipatory satellite handover strategy. • HACO: A multi-HAP-assisted computation offloading algorithm for SAGSIN proposed in [14], targeting the minimization of overall task execution delay. • FLEC: A four-layer edge computing scheme that aims to minimize system task costs proposed in [18] with a similar hierarchical architecture to this study. Fig. 3 illustrates the trend of average task completion delay for different methods with the percentage variation range of available computing resources for servers across time slots.

TABLE I S IMULATION PARAMETERS Parameter V Dv Cv L0

Definition Number of vessels Data size of vessels Computational density Noise power spectral density Reference path loss

Hu , H b , H h

Height of UAV, BS, HAP

dh,s Fu , Fb , Fh , Fs

H2S distance Computing resources of each server

Bu , Bb , Bh , Bs

Bandwidth of each server

Pv , P u , P h Gu , G h , G s τ

Transmit power Antenna gains Duration of time slot

N0

Value 10–30 2–10 Mb 100–2000 -174 dBm/Hz -30 dB [0.1, 0.03, 20] km 784 km [1, 3, 3, 10] ×108 cycles/slot [10, 20, 20, 50] MHz [1, 1.5, 2.5] W [25, 30, 35] dBi 0.1 s

Fig. 3. The comparison of average task completion delay among different methods with varying fluctuation range of available computing resources.

DASH adaptively adjusts the fine-grained task and resource scheduling strategy based on real-time available computing resources of all servers as well as the incoming satellite, achieving the lowest average task completion delay. When the incoming satellite has limited computing capacity, DASH w/o HO cannot proactively constrain the offloaded task volume to satellite, leading to post-handover congestion and increased satellite computation delay. The one-shot task and resource scheduling of HACO and FLEC leads to inconsistency in the available computing resources of servers between the decisionmaking time and the time when tasks arrive at destination servers. Upon task arrival, a reduction in available computing capacity increases computation delay, whereas an increased resource cannot be exploited. Consequently, their average task delays grow monotonically with intensifying computing resource fluctuations. DASH adaptively routes task data based on instantaneous server states, which effectively absorbs the impact of resource fluctuations, reducing the average delay by around 28% and 25% compared to HACO and FLEC, respectively. Fig. 4 depicts the average task completion delay among different methods with respect to the percentage variation range of available bandwidth for wireless links in the system across time slots. For the proposed DASH method and its ablation variant DASH w/o HO, task offloading and bandwidth resource allocation are dynamically adjusted based on real-time link states, including available bandwidth, large and small-scale fading, and channel capacities, enabling strong adaptability to communication dynamics and achieving low and stable average

for vessel users. To address the complexity and high dynamics of multi-layer task offloading, we developed a layer-wise offloading scheme that adapts to real-time system states, along with an anticipatory satellite handover policy that prevents post-handover congestion. Additionally, we jointly optimized UAV-BS bandwidth allocation, UAV trajectories, and computing resources to accelerate task completion and enhance resource utilization. Experimental results show that the proposed method significantly reduces task delay under fluctuating system resources and adaptively controls satellite backlog during handover compared with benchmarks. R EFERENCES Fig. 4. The comparison of average task completion delay among different methods with varying fluctuation range of available bandwidth resources.

Fig. 5. The comparison of satellite backlogged data at handover among different methods with varying computing resource gap of incoming satellite.

task delay with bandwidth variation. DASH reduces the average task delay respectively by around 26% and 23% compared to HACO and FLEC, since their one-shot scheduling strategies fail to adaptively reroute task when channel conditions deteriorate or reallocate resources when bandwidth becomes abundant. Compared with more dominant computing resources, which govern the terminal processing rates, the intensification of bandwidth fluctuations has a relatively smaller impact on the increase of their average task delay. Fig. 5 shows the amount of backlogged task data on the current satellite at handover (i.e., data to be transferred to incoming satellite) versus the deficit percentage in the average available computing resource of the incoming satellite relative to the current one. Due to the anticipatory handover strategy, DASH markedly reduces the amount of backlogged data at handover as the incoming satellite’s available computing capacity continuously decreases below that of the current satellite, thereby effectively alleviating post-handover congestion. Benefiting from fine-grained and more balanced task and resource scheduling, DASH w/o HO achieves lower satellite backlog at handover than HACO and FLEC. However, since all those three methods determine strategies solely based on the current satellite state, they lack adaptability to changes in the computing capability of incoming satellite. Consequently, their handover backlogs remain insensitive to the increasing gaps in the next satellite’s computing resources. VI. C ONCLUSION This paper proposed a dynamic task and resource scheduling method for green SAGSIN to minimize task execution delay

[1] J. You, Z. Jia, C. Dong, Q. Wu, and Z. Han, “Joint computation offloading and resource allocation for uncertain maritime MEC via cooperation of AAVs and vessels,” IEEE Trans. Veh. Technol., vol. 74, no. 11, pp. 1808118095, Nov. 2025. [2] Y. Bai and Y. Zhang, “Dynamic offloading based on Lyapunov optimization for UAV-assisted maritime IoT-MEC networks,” IEEE Trans. Veh. Technol., vol. 74, no. 11, pp. 17894-17906, Nov. 2025. [3] Y. Ye, S. Gao, X. Zheng, and L. Yang, “Multi-tier UAV edge computing for low altitude networks towards long-term energy stability,” in Proc. Int. Conf. Wireless Commun. Signal Process. (WCSP), Oct. 2025, pp. 1-6. [4] S. Gao et al., “Integrated sensing, communication, and computation for low-altitude networks towards seamless connectivity and connected intelligence,” IEEE Internet Things Mag., doi: 10.1109/MIOT.2026.3660848. [5] M. Dai et al., “Latency minimization oriented hybrid offshore and aerialbased multi-access computation offloading for marine communication networks,” IEEE Trans. Commun., vol. 71, no. 11, pp. 6482-6498, Nov. 2023. [6] M. Li, L. P. Qian, F. Fang, and X. Wang, “Online hierarchical computation offloading for marine IoT networks: A delay minimization approach,” IEEE Trans. Wireless Commun., vol. 25, pp. 3422-3436, 2026. [7] Y. Zhang, Z. Na, S. Li, B. Lin, Y. Lin, and A. Nallanathan, “Joint service caching and task offloading for multi-UAV-assisted offshore edge computing networks,” IEEE Trans. Veh. Technol., vol. 74, no. 12, pp. 19667-19680, Dec. 2025. [8] Z. Wang, B. Lin, and Q. Ye, “Double-edge-assisted computation offloading and resource allocation for space-air-marine integrated networks,” IEEE Trans. Veh. Technol., vol. 74, no. 9, pp. 14501-14514, Sept. 2025. [9] S. Jung, S. Jeong, J. Kang, and J. Kang, “Marine IoT systems with space–air–sea integrated networks: Hybrid LEO and UAV edge computing,” IEEE Internet Things J., vol. 10, no. 23, pp. 20498-20510, 1 Dec.1, 2023. [10] S. Qi, B. Lin, Y. Deng, H. Pan, and X. Hu, “Joint computation offloading and resource management for cooperative satellite–aerial–marine internet of things networks,” IEEE Internet Things J., vol. 12, no. 24, pp. 5316453176, 15 Dec.15, 2025. [11] M. Dai, S. Chang, Y. Wang, and Z. Su, “Energy-efficient multi-access edge computing for heterogeneous satellite-maritime networks: A hybrid harvesting-and-offloading design,” IEEE Trans. Mobile Comput., vol. 24, no. 11, pp. 12001-12018, Nov. 2025. [12] H. Zhang, S. Xi, B. Shang, P. Zhang, S. Wu, and C. Jiang, “Energy oriented three-tier computation offloading scheme in maritime edge computing network,” IEEE Trans. Veh. Technol., vol. 74, no. 5, pp. 8126-8140, May 2025. [13] W. Li, S. Li, J. Hao, Q. Wu, and R. Wang, ”Efficient task offloading and resource allocation in HAPS-assisted LEO satellite networks: A MAPPO with exact potential game approach,” IEEE Internet Things J., vol. 13, no. 6, pp. 10179-10195, 15 March15, 2026. [14] W. Wu, W. Feng, Y. Fang, Z. Lin, and X. Lu, “Multi-HAP-assisted computation offloading in space–air–ground–sea integrated network,” IEEE Internet Things J., vol. 12, no. 12, pp. 21806-21818, 15 June15, 2025. [15] Z. Lin, J. Yang, Y. Chen, C. Xu, and X. Zhang, “Maritime distributed computation offloading in space-air-ground-sea integrated networks,” IEEE Commun. Lett., vol. 28, no. 7, pp. 1614-1618, July 2024. [16] D. Wang, T. He, Y. Lou, L. Pang, Y. He, and H. -H. Chen, “Double-edge computation offloading for secure integrated space–air–aqua networks,” IEEE Internet Things J., vol. 10, no. 17, pp. 15581-15593, 1 Sept.1, 2023. [17] R. Esswein, Q. Bayer, S. Mergendahl, J. Ruffley, M. Abdelhakim, and R. K. Cunningham, “Constellation parameters for minimizing propagation delay over LEO inter-satellite links,” IEEE Trans. Netw., doi: 10.1109/TON.2026.3661441. [18] X. Wang, S. Wang, X. Gao, Z. Qian, and Z. Han, “AMTOS: An ADMM-based multilayer computation offloading and resource allocation optimization scheme in IoV-MEC system,” IEEE Internet Things J., vol. 11, no. 19, pp. 30953-30964, 1 Oct.1, 2024.