This paper investigates the performance of adaptive
LoRa networks with dynamic SF allocation accounting for Duty
Cycle (DC) restrictions and quantifying the imperfect orthogonality of Spreading Factor (SF)s. The study presents a novel
spatiotemporal model that combines stochastic geometry and
queuing theory where LoRa devices are perceived as interacting
two-dimensional DTMCs. Each chain jointly tracks the number
of packets in the buffer and the node’s protocol state. Numerical
simulations are carried out to validate the accuracy of the
proposed model. The network performance is studied in terms of
Pareto frontiers under different orthogonality assumptions and
adaptation settings, showcasing the ranges of sensing applications
that LoRa can accommodate without compromising the network
stability. The evolution of SFs activity distribution, coverage
probability and average latency is examined against different
network parameters. The results show that activating SF adap tation with higher cardinality is not always advantageous and
evince the existence of an adaptation cardinality that minimises
the delay. The study also identifies regimes where SF adaptation
is advantageous for the network scalability and reveals ’SF-Up’
and ’SF-Down’ rates that maximise the coverage or minimise
the delay. Comparing dynamic to static SF allocations, the
results highlight a tradeoff between coverage and latency yielding
valuable insights into scenarios where either of the allocation
strategies would be more beneficial to the network.