A rooftop array rarely increases the chance of a direct strike, but it sharply increases what a nearby one costs you. What to bond, which SPD type belongs where, and the corners that cheap installations cut.
A jute mill in Narsingdi lost two inverters and a monitoring gateway inside one week in April. Nothing was struck. Nothing on the roof was even marked. The damage was entirely induced, from strikes some distance away, and it was preventable for a fraction of what the replacements cost.
That is the shape of the problem, and it is not the shape most people expect.
A flat array on a large industrial roof barely increases the building's collection area for a direct strike. The modules are low, the building was already the tallest thing in its own footprint, and if there is a lightning protection system the air terminations are still above everything. So the probability of being hit does not move much.
What changes is the consequence of a strike that lands nearby. You have just installed several kilometres of conductor across a roof, arranged in loops that enclose real area, terminated in power electronics that are sensitive to microsecond transients. A strike to a tree, a chimney or a pole a few hundred metres away induces voltage in those loops by simple magnetic coupling. It does not have to touch you. Most photovoltaic lightning damage in Bangladesh is of this kind, and the giveaway is that everything looks perfect on the roof while the inverter input stage is destroyed.
Anybody who has spent a March or April in Sylhet or on the northern plains knows how many storm days the pre-monsoon brings. This is not a theoretical risk to be handled with a paragraph in a specification. It is the single most common cause of unplanned inverter replacement on Bangladeshi rooftops.
Conversations about this get confused because three separate systems are all described as lightning protection.
The framework that ties them together is the IEC 62305 series. Part two is a risk assessment: you put in the structure's dimensions, its location and ground flash density, what it is built of, what is inside it and what the consequences of a failure are, and the calculation tells you whether protection is required and at which protection level. This matters commercially, because a risk assessment is a document a client can ask for at tender stage. Without one, the decision about how much protection to fit is being made by whoever is quoting, and the incentive there is obvious.
On a building with an LPS there is a genuine design fork, and it is the fork most often fudged.
The isolated approach keeps the array physically clear of the air terminations and down conductors by at least the calculated separation distance, so that lightning current flowing in the LPS cannot flash across to the array. The separation distance is a calculation, not a guess, and it depends on the protection level, the number of parallel down conductor paths and the length of conductor involved. If you can achieve it, the DC side then only needs to deal with induced surges.
The bonded approach accepts that the array structure is connected to the LPS, deliberately or unavoidably. That is a legitimate design, but it has a direct consequence: part of the lightning current can now flow in the array's earthing and, by coupling, in its conductors. Once that is true, the DC surge protection has to be rated for lightning impulse current with the long 10/350 microsecond waveform, which means Type 1 devices, not Type 2 devices rated for the shorter 8/20 microsecond induced surge.
Fitting Type 2 SPDs on a bonded array is the most common invisible corner cut in the whole discipline. The plant works. It passes a visual inspection. It fails the first time it is asked to do the job, and by then the SPD is a burnt component in a photograph.
Module frames and mounting rails are anodised aluminium. Anodising is an oxide layer, and an oxide layer is an insulator. A module clamped to a rail is mechanically secure and electrically unreliable, and the resistance across that joint changes with weather, torque and age.
So the earthing of the array has to be deliberate:
The voltage induced in a circuit by a nearby strike is proportional to the area the circuit encloses. This gives you a genuinely free way to cut the damage, and it is a wiring practice rather than a purchase.
Route the positive and negative of each string together, tied along the same path, from the first module to the inverter input. A string wired with the positive down one edge of the array and the negative back along the other edge encloses the whole array and behaves like a pickup coil. Keeping the pair together, close to the earthed mounting structure, can reduce the induced voltage by a large factor for no material cost at all.
Then bring the DC cables down inside the building, close to the earthed structure, rather than draping them down an outside wall. Keep DC and AC in separate containment but running parallel and close to bonded metalwork. And keep the run from the array to the inverter as short as the layout allows, because both the surge exposure and the voltage drop reward it.
Four placements matter, and the fourth is the one that gets left out of every budget quotation.
DC side. An SPD at the inverter DC input, and a second set at the array end if the cable run between them is long. The rule of thumb used in practice is that beyond roughly ten metres of DC cable you need protection at both ends. The device must be specifically rated for photovoltaic DC use, not simply for DC. A PV array is a current source that will feed a fault indefinitely with no zero crossing, so a PV SPD carries a thermal disconnector and an internal arrangement designed for that duty. Its maximum continuous operating voltage has to exceed the array's open circuit voltage at the coldest expected cell temperature, which brings you straight back to the string design calculation.
AC side at the inverter. A Type 2 device on the inverter's AC output terminals.
AC side at the origin of the installation. A Type 1 plus Type 2 device at the main incoming LT panel. If two SPDs on the same circuit are close together they can interfere with each other's operation unless the manufacturer states that they are coordinated, so either keep the required separation in cable length or use a decoupled set from one manufacturer.
Data and signal. The RS485 run between inverters, the Ethernet cable to the office, the CT wiring for the export limiter, the weather station cable. These are long, often unscreened, cross zone boundaries, and are usually the first things to die. A signal SPD on each of them costs very little and prevents the classic outcome where the inverters survive and the gateway, the switch and the office PC do not.
One detail decides whether any of this works. The protection level an SPD achieves is its own clamping voltage plus the inductive voltage developed along its connecting leads during the surge, and that inductive term is large. The total lead length, line side plus earth side, should be kept as short as physically possible, which in practice means well under a metre. A properly specified SPD on a long earth tail neatly routed around the panel is not protecting anything. If the panel layout will not allow short leads, use a V-connection or move the device.
Bangladeshi sites accumulate earth pits. There is the original building earth, a separate pit somebody drove for a generator, one for the transformer neutral, one for the lightning protection, and now a new one for the solar. If those electrodes are not bonded together, the site is more dangerous than it would be with a single one, because a surge raises the potential of the electrode nearest the event and the full difference appears across whatever equipment bridges the two systems. That is often the inverter, sitting with its earth on one system and its DC cables running to a structure earthed on the other.
Everything on the site bonds to one earthing system. That is the rule, and it is not negotiable for the sake of tidiness in the drawing.
On resistance targets: the number depends on what the electrode is for, and quoting a single figure without saying which duty it serves is meaningless. In the soft delta soils around much of Bangladesh a low value is straightforward to reach with an electrode driven down into the water table. On higher ground, in laterite, and in the dry season before the monsoon recharges the ground, it is harder, and an electrode measured in August is not the same electrode in April. Measure by a proper fall of potential test at the driest time of year, record it, and install a test link and an inspection pit at every electrode so it can be measured again without excavation. Then put that measurement in the annual maintenance schedule, because earth connections are the part of an installation that is made once and never looked at again.
If you are inspecting a quotation or an existing plant, this is the list, and every item on it has been found on rooftops here: