Yet another post talking about the weather – but this time, from the perspective of what detail to throw away.

Table Of Contents

In part 1:

Chapter 5: Land Transport

5.1 Distance, Time, & Detriments

5.1.1 Time Vs Distance
5.1.2 Defining a terrain / region / locality

     5.1.2.1 Road Quality: An introductory mention

5.2 Terrain

5.2.0 Terrain Factor
5.2.1 % Distance
5.2.2 Good Roads
5.2.3 Bad Roads
5.2.4 Even Ground
5.2.5 Broken Ground
5.2.5 Marshlands
5.2.7 Swamplands
5.2.8 Woodlands
5.2.9 Forests
5.2.10 Rolling Hills
5.2.11 Mountain Slopes
5.2.12 Mountain Passes
5.2.13 Deserts
5.2.14 Exotic Terrain
5.2.15 Road Quality
     5.2.15.1 The four-tier system
     5.2.15.2 The five-tier system
     5.2.15.3 The eight-tier system
     5.2.15.4 The ten-tier system

5.2.16 Rivers & Other Waterways
     5.2.16.1 Fords
     5.2.16.2 Bridges
     5.2.16.3 Tolls
     5.2.16.4 Ferries
     5.2.16.5 Portage & Other Solutions

Today:

5.3 Weather

5.3.1 Seasonal Trend
5.3.2 Broad Variations
5.3.3 Narrow Variations
     5.3.3.1 Every 2nd month?
     5.3.3.2 Transition Months
     5.3.3.3 Adding a little randomness: 1/2 length variations
     5.3.3.4 Adding a little randomness: 1 1./2-, 2-, and 2 1/2-length variations

5.3.4 Maintaining The Average
     5.3.4.1 Correction Timing
          5.3.4.1.1 Off-cycle corrections
          5.3.4.1.2 Oppositional corrections
          5.3.4.1.3 Adjacent corrections
          5.3.4.1.4 Hangover corrections

     5.3.4.2 Correction Duration
          5.3.4.2.1 Distributed corrections: 12 months
               5.3.4.2.1.1 Even Distribution
               5.3.4.2.1.2 Random Distribution
               5.3.4.2.1.3 Weighted Random Distribution

          5.3.4.2.2 Distributed corrections: 6 months
          5.3.4.2.3 Distributed corrections: 3 months
          5.3.4.2.4 Slow Corrections (2 months)
          5.3.4.2.5 Normal corrections: 1 month
          5.3.4.2.6 Fast corrections: 1/2 month (2 weeks)
          5.3.4.2.7 Catastrophic corrections 1/4 month (1 week)

     5.4.4.3 Maintaining Synchronization
     5.4.4.4 Multiple Correction Layers

5.4 Losses & Hazards
5.5 Expenses – as Terrain Factors
5.6 Expenses – as aspects of Politics

In parts to come:

5.7 Inns, Castles, & Strongholds

5.7.1 Strongholds
5.7.2 Castles
5.7.3 Inns

5.8 Villages, Towns, & Cities

5.8.1 Villages
     5.8.1.1 Village Frequency
     5.8.1.2 Village Initial Size
     5.8.1.3 The Generic Village

5.8.2 Towns
     5.8.2.1 Towns Frequency
     5.8.2.2 Town Initial Size
     5.8.2.3 The Generic Town

5.8.3 Cities
     5.8.2.2 Small City Frequency
     5.8.2.3 Small City Size
     5.8.2.4 Size Of The Capital
     5.8.2.5 Large City Frequency
     5.8.2.6 Large City Size

5.8.4 Economic Factors, Simplified
     5.8.4.1 Trade Routes & Connections
     5.8.4.2 Local Industry
     5.8.4.3 Military Significance
     5.8.4.4 Scenery & History
     5.8.4.5 Other Economic Modifiers
     5.8.4.6 Upscaled Villages
     5.8.4.7 Upscaled Towns
     5.8.4.8 Upscaled Small Cities
     5.8.4.9 Upscaling The Capital & Large Cities

5.8.5 Overall Population
     5.8.5.1 Realm Size
     5.8.5.2 % Wilderness
     5.8.5.3 % Fertile
     5.8.5.4 % Good
     5.8.5.5 % Mediocre
     5.8.5.6 % Poor
     5.8.5.7 % Dire
     5.8.5.8 % Wasteland
     5.8.5.9 Net Agricultural Capacity

     5.8.5.10 Misadventures, Disasters, and Calamities
     5.8.5.11 Birth Rate per year
     5.8.5.12 Mortality
          5.8.5.12.1 Infant Mortality
          5.8.5.12.2 Child Mortality
          5.8.5.12.3 Teen Mortality
          5.8.5.12.4 Youth Mortality
          5.8.5.12.5 Adult Mortality
          5.8.5.12.6 Senior Mortality
          5.8.5.12.7 Elderly Mortality
          5.8.5.12.8 Venerable Mortality
          5.8.5.12.9 Net Mortality

     5.8.5.13 Net Population

5.8.6 Population Distribution
     5.8.6.1 The Roaming Population
     5.8.6.2 The Capital
     5.8.6.3 The Cities
     5.8.6.4 Number of Towns
     5.8.6.5 Number of Villages
     5.8.6.6 Hypothetical Population
     5.8.6.7 The Realm Factor
     5.8.6.8 True Village Size
     5.8.6.9 True Town Size
     5.8.6.10 Adjusted City Size
     5.8.6.11 Adjusted Capital Size

5.8.7 Population Centers On The Fly
     5.8.7.1 Total Population Centers
     5.8.7.2 The Distribution Table
     5.8.7.3 The Cities
     5.8.7.4 Village or Town?
     5.8.7.5 Size Bias
          5.8.7.5.1 Economic Bias
          5.8.7.5.2 Fertility Bias
          5.8.7.5.3 Military Personnel
          5.8.7.5.4 The Net Bias

     5.8.7.6 The Die Roll
     5.8.7.7 Applying Net Bias
     5.8.7.8 Applying The Realm Factor
     5.8.7.9 The True Size
          5.8.7.9.1 Justifying The Size
          5.8.7.9.2 The Implications

5.9 Compiled Trade Routes

5.9.1 National Legs
5.9.2 Sub-Legs
5.9.3 Compounding Terrain Factors
5.9.4 Compounding Weather Factors
5.9.5 Compounding Expenses
5.9.6 Compounding Losses
5.9.7 Compounding Profits
5.9.8 Other Expenses
5.9.9 Net Profit

5.10 Time
5.11 Exotic Transport

And, In future chapters:

6. Waterborne Transport
7. Spoilage
8. Key Personnel
9. The Journey
10. Arrival
11. Journey’s End
12. Adventures En Route

Background Landscape by Pete Linforth from Pixabay. Merchant Train image by G.C. (garten-gg) from Pixabay.

5.3 Weather

In a previous part of this series I spent an entire post detailing a very rich weather system because ships at sea are utterly dependent on the weather and on very precise details. On land, such specificity is a burden, not a luxury, and so this chapter includes a much simpler weather system. I very strongly want to let it be shaped by the previous one, though, so that if further detail is needed – and from time to time it will be – you can simply drop in whatever part of the previous system is relevant, use it for as long as you need it, and then put it away again.

5.3.1 Seasonal Trend

The place to start, therefore, is with a seasonal trend. How much hotter or colder does it get, each day, on average? And how much does the daily minimum change?

How much does the daily chance of precipitation in some form increase or decrease each day?

How much does the average amount of such precipitation rise or fall?

Is this region or area known for strong winds, and what times of year do they arrive?

These five facts define a seasonal weather pattern. You can specify it in months if you want, on the assumption that a season can be subdivided into three parts – early, middle, and late – or into 6-week spans (so a season of two halves) – or you can even do seasonal spans (4 weeks late of one season and 4 weeks early of the next as one long trend, then four weeks at the height of the new season, then start again – dividing the year into eight unequal portions).

Tropical climates don’t even need that much variation – they have two seasons, wet and dry. And there isn’t so much a transition as a catastrophic changeover.

As readers will have seen in my series, The Diversity Of Seasons, I like to model game weather in my campaigns on a real location as much as I can, for two reasons: first, there may be historical weather records that I can access, ensuring a realism that simply can’t be achieved any other way; and secondly, because an annual weather summary makes it a lot easier to boil a specific climate down into exactly the sort of trends described above.

But, for this system, I think even this is more richly detailed than necessary. So instead, let’s start here:

For a major location or a region, write down a one-line answer to each of the following questions:

1. What’s a good-weather day in summer like?
2. What’s a bad-weather day in summer like?
3. What’s a good-weather day in autumn like?
4. What’s a bad-weather day in autumn like?
5. What’s a good-weather day in winter like?
6. What’s a bad-weather day in winter like?
7. What’s a good-weather day in spring like?
8. What’s a bad-weather day in spring like?

Now, we’ve reduced those 5 parameters down to just one or two: How much more or less likely are bad weather days? How much more or less likely are good weather days?

The assumption, of course, is that most days are going to be somewhere in between.

If that’s all you’re ever going to need, that’s it, you can move on to the next section. Sadly, it probably isn’t, and you’re back to looking at those 5 numeric parameters.

So let’s pick somewhere and set them, just as an example. For no reason in particular, let’s pick a place I know nothing about aside from general perceptions: Brensbach, Germany. No, that doesn’t work, no weather data. All right, Bonn. There’s probably lots of climate info on Bonn.

Wikipedia tells me Bonn is in one of Germany’s warmest regions, so I’ll take that on board.

Looking at the climate chart, there are three values to pick between for daily maximum temperature – Record High, Mean Maximum, and Mean Daily Maximum. The first is the most extreme ever recorded, not useful in this context; the second is the average of the highest in any given month over a number of years, so that’s a contender; the third is the average of all maximum temperatures in the month over those many years. What’s the difference?

Here are three (invented) weather records in graph form for June for Bonn:



In the fictional-2022, there were some highs and some lows and a slight drop in average temperature over the month (the black line). In fictional-2023, there is a slightly more extreme drop in maximum temperature but it’s still fairly consistent. In fictional-2024, there are four major warm periods separated by colder ones and the overall temperature has declined noticeably by the end of the month.

Note the days circled in yellow – those are the individually-hottest days in the course of the month. In two cases out of three, they occur early in the month.

The figure supplied by Wikipedia for the Mean Maximum is the mean of all the individually hottest days for all the Junes on record in Bonn. In terms of forecasting the weather, not very useful. Averaging the maximum temperatures for every June day on record gives the Mean Dally Maximum. That’s the one we want.

But the difference to the other one is also useful – we can write the daily June average as (in this case) 22.5°C ± 9°.

The numbers for July are 24.1°C ± 8.8°.

The difference from June to July is +1.6°C ± 8.9°C

Notice that I’ve averaged the ± values for each month.

Over 30 days, that’s +0.0533333°C – too low to be useful – or 1°C every 18.75 days. Fahrenheit, because the size of a degree is smaller, might appear to be more useful for this, but here’s a though to consider: it’s my contention that 1°C is the smallest temperature change that we humans notice. If it’s 22°C right now, and an hour from now it’s going to be 23°C, we will actually notice the difference.

In some places, Summer will be May-June-July. In other places, it might be mid-May-June-July-mid-August.

I don’t care about any of that. It’s too detailed, too fussy. I want to compare the middle months of each season with the middle month of the next season, and those distinctions mostly get washed away in the process. So the process for setting the post-midsummer change is to compare June (midsummer) with September (mid-Autumn).

June numbers again: 22.5°C ± 9°.
September: 20.0°C ± 7.4°.

Ignoring mid-season, that means that late summer and early autumn in total (about 60 days) yield -2.5°, &PlusMinus 8.2°. And, 2.5 / 60 = 0.0416667, or -1° every 24 days.

I can do the same for the minimum daily temperatures:

June: 11.8°C ± 5.1°
September: 10.3°C ± 5.5°.

60-day change: -0.025°/day or 1° every 40 days, ± 5.3°C.

Average Precipitation is per month. In June it was 81.5mm (3.21 inches). But the row after next gives the average number of rainy days as 14.1 – so divide one by the other to get an average daily rainfall on wet days of 81.5/14.1 = 5.78mm / rainy day, or about 0.2276 inches. That’s more than a shower, it’s a light rain over several hours or a shorter, more intense, rain, like in a thunderstorm. 14.1 out of 30 = 47% chance of rain each day.

The September numbers: 62.5mm over 13.6 days = 4.59mm / rainy day; and 13.6/30 = 45.333% chance of rain each day.

Over 60 days, we go from 47% chance of rain to 45.333% chance of rain, so the overall chance is 47% -1% every 36 days.

The amount of rain on a rainy day goes from 5.78 to 4.59 mm, a change of -1.19mm, or 0.0198 mm/day or -1 mm every 50.42 days, or -0.1 mm every 5 days. That last isn’t bad, in terms of usefulness, but it might be even better to calculate it over rainy days and not all days.

Same change of -1.19mm; but instead of 60 days, we’re looking at the average of 47 and 45.333% times that sixty days, or 46.1665% x 60 = 27.7 rainy days.

-1.19mm / 27.7 rainy days = -0.04296 mm / rainy day or 23.277 rainy days per -1 mm or 2.327 rainy days per -0.1 mm.

Once a weather profile for an area has been calculated, it will never change absent some extraordinary geographic interference. The only remaining question before we can move on is how far does this climate representation extend?

This is a far more rubbery question. There are no good answers, because climate is just too complicated, and impacted by all sorts of factors. To offer some sense of the complexities involved: Africa and Australia are about as far apart as you can get and still be in the same hemisphere – except for South America, which is further from Australia..



This map was produced in 1961 by the Central Intelligence Agency of the USA and is available in raster form from the US Library Of Congress. As a work-product of the United States Government, it is considered to be in the Public Domain in the United States. Image courtesy of’ Wikimedia Commons. Cropped by me with notations moved inboard and contrast / color-depth increased.

When inland Australia has a relatively wet year, there is usually drought in Africa. When it rains there, we have drought here, and heightened bushfire dangers. You might expect that all that ocean in between had a decisive role to play, but no. Is it our weather that changes theirs, or their weather that changes ours, or is the relationship still more complicated than that? I would bet on the latter. It’s my understanding that it’s a question of whether or not the potential rain falls from the atmosphere there, or stays up there until it gets to here, but I have low levels of confidence in that understanding.

The fact that weather is so complicated is both a good and a bad thing, from an RPG systems perspective. It means that almost any output from a game mechanic can be rationalized and interpreted plausibly; but it also means that any decent system gets easily bogged down because there are so many aspects of the weather to pin down..

5.3.2 Broad Variations

In Australia, we are very familiar with El Nino and La Nina phenomena and the Southern Oscillation Index, and the impact that they have on our weather, and are slowly becoming more aware generally of the impact of the Indian Ocean Dipole on our weather. To quote from Wikipedia:

Across most of the continent, El Nino and La Nina have more impact on climate variability than any other factor. There is a strong correlation between the strength of La Nina and rainfall: the greater the sea surface temperature and Southern Oscillation difference from normal, the larger the rainfall change.

During El Nino events, the shift in rainfall away from the Western Pacific may mean that rainfall across Australia is reduced. Over the southern part of the continent, warmer than average temperatures can be recorded as weather systems are more mobile and fewer blocking areas of high pressure occur. The onset of the Indo-Australian Monsoon in tropical Australia is delayed by two to six weeks, which as a consequence means that rainfall is reduced over the northern tropics. The risk of a significant bushfire season in south-eastern Australia is higher following an El Nino event, especially when it is combined with a positive Indian Ocean Dipole event.

… Australia … experiences extensive droughts alongside considerable wet periods that cause major floods. There exist three phases – El Nino, La Nina, and Neutral… Since 1900, there have been 28 El Nino and 19 La Nina events in Australia including the current 2023 El Nino event. The[se] events usually last for 9 to 12 months, but some can persist for two years [or more], though the ENSO cycle generally operates over a time period from one to eight years.

What this means is that there are multiyear long-term cycles of varying duration and intensity, which compound with the typical climatic pattern. It’s the same everywhere, though the intensity of such effects can vary. As a general rule, you cycle from hotter-dryer to cooler-wetter, but there’s enough variability to cause complications.

I propose getting around these implications and complications by applying separate cycles for rainfall and temperature, even though that’s not actually accurate.

Any of the values recorded in the previous section can be affected. One combination of factors might mean warmer nights and cooler days; another might not affect night-time temperatures at all but could push daytime maximums up or down.

Ultimately, these boil down to two factors: how much, and for how long?

How much: roll d-something (based on the level of variability of the season and rounding down) and divide by 2. When the current phase of the cycle ends, make the same roll and apply it in the opposite direction – so if you’re getting +3.5°C on the daily maximums, and you roll 2.5°C, then the modifier drops to +1°C. And, since that’s still a positive number, the next roll will also subtract from it.

How long: 6+d12 months, but if you roll a 12, subtract 2 and add a d6 and another d12.

11/12ths of the time, you’ll get a flat 7-18 months, average 12 months. One time in 12, you’ll get a 16+d6+d12 pattern, which looks like this:



Each time you generate the weather, you simply add this modifier to the result and take it off the variability.

5.3.3 Narrow Variations

That leaves only the more frequent and common daily cycles. There are two phenomena that these have to replicate: one off-events where the system ‘clears its throat’ and protracted events that impact weather over multiple days.

This is rather trickier, because long-range events are rare, but get more likely the shorter they are; while short-term events last a day or two, three at the most, and the shorter the duration, the more likely they are. So we have two completely distinct probability curves, in terms of duration.

That requires, generally speaking, two die rolls. And, given the shapes involved, divided die rolls at that.

Roll 1: 2d8 / (d4+3), minimum result 1
Results > 2, use Roll #2 instead: [(3d8+4) / 2d4] +2



When you plot these, this is what you get: 93% of the time, you’re talking a 1-2 day event, with 1 happening 64% of the time. But on the remaining 7%, there’s a peak probability of 3-to-8 days, but an outside chance of an event lasting up to 12 days – and an extremely remote probability of another 4 days beyond that (amounting to just over 1% of the 7%).

Intensity of event follows the same basic curve as the second results roll without the +2, divided by 2 instead:

Roll3: [(3d8+4) / 2d4] /2, round down.

So instead of 4-5, the answers are 1-1.5. For practical purposes anything more than 4.5 is going to be 1% or less.

If you really want to, you can divide the result by 4.5 and multiply by double the variation determined after the long-term weather patterns are taken into account. I don’t think it’s worth the effort myself – but I would probably have stopped at the “% chance good days” in the first place.

5.3.3.1 Every 2nd month?

There are all sorts of things that can be done to make the system more robust. I don’t think they are necessary, either, with one possible exception.

The first is to divide the year up a little differently. Say that seasonal transitions take a little less time and are more abrupt, and that the mid-season period is longer and more consistent therefore. This is less realistic but makes the changes more dramatic, and hence, more impactful in game terms. It also means that you can ignore all the complicated stuff more of the time and just run off the core season.

Another approach is to ignore the seasons entirely, and let them emerge naturally from the weather. This approach means that you are always transitioning from the previous month to the next month and ignoring what’s recorded for the in-between.

So:
January = the average of December & February.
February = the average of January & March.
March = the average of February & April.
…. and so on.

This ‘second month’ approach smooths over the maths quite a lot. You’re no longer worried about trends; instead, you take the average that you calculate and that is the base weather. Apply the long-term cycles and short term variations as usual.

5.3.3.2 Transition Months

Another technique is to say that the seasons are 2 1/2 months long, and average the relevant months to get a base weather for the entire season. That leaves transitions taking 1 week of the end of the season and the first week of the next, which you get by averaging the two months.

Shortening the transition months like this makes weather transitions so sharp as to be unrealistic, but you are going to be counting on random variations to hide that fact. If that doesn’t seem enough, weight the outcomes of the random rolls to favor the results that should be trending – if you’re heading into summer and the dice indicate a cool day (down 4 degrees on the average), flip the result to make it 4 degrees warmer than the indicated average.

5.3.3.3 Adding a little randomness: 1/2 length variations

This is the one that might be worthwhile – it halves the length of random variations by making the results the number of “half-days” instead of “whole days”. You can then roll a random number for exactly when the change comes through – I would suggest adding or subtracting up to 6 hours, in other words, d12-6 hours. That means that you can get evening shifts, dawn shifts, and noon shifts as well as midnight shifts in the weather.

I would further simplify: if the indicated duration is 1/2 day or less, it counts for zero against the pattern; if it’s more than 1/2 a day, then it counts as a full day.

5.3.3.4 Adding a little randomness: 1 1./2-, 2-, and 2 1/2-length variations

I was talking about these general principles to someone once, and they suggested expanding the half-day variation principle to longer durations, effectively compensating for the /2 by multiplying a d5 by the duration, or maybe a d6. Presented here for the sake of completeness, this is not something I would recommend..

5.3.4 Maintaining The Average

Ever have a time when a series of die rolls for the same thing came up high all the time, even when rolled days apart? I’ve seen it happen. In the extremely long run, it all averages out of course, but the fact of the matter is that simple weather generation systems are necessarily poorly granular – there’s often quite a lot of variation and not a lot of subtlety to the results. So there’s one further refinement that’s worth considering, no matter how you roll for your weather.

If you only generate results when you need them, this lack of granularity gets amplified.

But there is a relatively simple solution: keep track of your results, I mean the end numbers not the bits in between. How hot did it get? How cool at night? How many rainy days were there, and how much rain fell?

From these, you can set yet another correction factor to bring the overall average back to whatever it’s supposed to be.

There are three factors to consider: when to calculate and apply a modifier, how long it needs to last, and how big of an adjustment?

If the weather system tells you there’s five days of heavy rain, you can expect floods. That much rain ruins the fit to the average expectation – the solution is to compensate for the greater rainfall by making other rainfall lighter or shorter or nothing more than threatening clouds, until the overall average is back where it’s supposed to be.

5.3.4.1 Correction Timing

Because you want as much of the correction to take place ‘naturally’, i.e. through daily variations and general weather patterns, these corrections shouldn’t happen all that frequently. For that reason, the timing of them should be linked to the long-term weather cycles and not to anything shorter.

There are four basic options. Which one should you choose? I would roll randomly, they are all as likely as each other.

5.3.4.1.1 Off-cycle corrections

Off-cycle means that as soon as the cycle ends you make the calculation and use the correction instead of rolling for a new long-tern cycle result. Once you’re back to average, you roll a new long-term cycle result to take it’s place and carry on your merry way.

5.3.4.1.2 Oppositional corrections

Oppositional Correction means that when the long term cycle indicates the opposite of what it did when the accumulated error took place, this additional modifier will compound with that long-term trend to make matters better – or worse.

You’re supposed to have a dry season but you roll a lot of rain? Compensate by making the following wet season trend drier than usual.

5.3.4.1.3 Adjacent corrections

This divides the compensation in two and applies them to both the next cycle phase and the one after that. In effect, it spreads the adjustment out over a longer period, making it less noticeable.

5.4.4.1.4 Hangover corrections

Finally, you can keep hold of any required adjustments until there’s a long-term trend in the indicated direction and then apply them. This means that long dry periods get (eventually) balanced by floods, and long wet periods get (eventually) balanced by droughts. The redress might not happen next year, or even the year after – it’s longer term than that.

This can mean that you end up with several such modifiers / corrections taking hold all at once. That happens in reality, too – a long run of medium-to-good years followed not just by a drought or cold snap, but by a massive drought – the kind that affects parts of Africa every now and then – or little ice age, like the ones reported in England in 1650, 1770, and 1850, each separated by intervals of slight warming. Modern climatology conflates all three into something referred to as the Little Ice Age.

Between 1649 and 1666, for four successive winters, the Thames froze over – something that only generally happened one year in ten even in the Little Ice Age. The most severe freezing of England ever recorded was during the Great Frost Of 1683-84 when the Thames froze for months at a time, and the ice reached as much as 11 inches thick (See River Thames Frost Fairs for more information).

That’s what an accumulation of corrections can look like.

5.3.4.2 Correction Duration

The next factor to consider is duration. There are five patterns, each roughly twice as likely as the one that follows it, and with the largest one subdivided along similar probability lines. When you map that onto a d% roll, you get:

01-29 Normal Correction (1 month)
30-44 Slow Correction (2 months)
45-52 Distributed Correction 3 months
52-77 Distributed Correction 6 months
78-90 Distributed Correction 12 months
91-97 Fast Correction (1/2 month)
98-00 Catastrophic Correction (1.4 month = 1 week)

The above takes into account two factors: that most errors will cancel out, leaving a relatively small net correction; and a lot of weather events are like a cascade of dominoes, or a chain reaction – something builds up until it can no longer be contained and then lets go – but it doesn’t happen all at once.

5.3.4.2.1 Distributed corrections: 12 months

I’m going to deal with these in sequence of decreasing length rather than a sequence derived from the probability shown above. A 12-month distribution is as long as a correction gets, and is one of the more improbable outcomes.

It does not mean that an adjustment is applied daily for 365 days. Rather, the size of the total adjustment required dictates the number of days.

The minimum adjustment is generally 0.5°C if your source measurements are to 0.1°, 1° otherwise if they are in °C, and 2° if they are in °F.’

For every 2 of this size, there will be one of twice as long. The pattern is A, BB, CCCC, DDDD DDDD, and so on.

You want the largest such pattern that is less than the total adjustment.

1 A = 1
2 B = 2 (+1 = 3)
3 C = 3 (+3 = 6)
4 D = 4 (+6 = 10)
5 E = 5 (+10 = 15)
6 F = 6 (+15 = 21)
7 G = 7 (+ 21 = 28)
8 H = 8 (+ 28 = 36)
9 I = 9 (+ 36 = 45)
10 J = 10 (+ 45 = 55)
11 K = 11 (+55 = 66)
12 L = 12 (+66 = 78)
13 M = 13 (+78 = 91)
14 N = 14 (+91 = 105)
15 O = 15 (+105 = 120)
16 P = 16 (+120 = 136)
17 Q = 17 (+136 = 153)
18 R = 18 (+153 = 171)
19 S = 19 (+171 = 190)
20 T = 20 (+190 = 210)

That should be far enough! In fact, it almost certainly goes too far, but that’s better than the alternative.

So let’s walk our way through the table and then I’ll explain the process of using it. the first number is the number multiplied by the minimum adjustment to get the actual adjustment for a specific scale. Those scales are labeled A, B, C, and so on. The third number is the weight of the adjustment, which I’ll explain in a moment, while the fourth number (in the brackets) is the cumulative weight.

The presence of a given level of adjustment – M, say – implies that there’s at least one adjustment of each smaller size also required. So the number to compare with the total is the cumulative.

Allocate the largest single adjustment and subtract it from the total correction. Divide the result by the cumulative value of the next lowest score – you’re only interested in whole numbers, so it’s not too difficult a calculation. That’s how many adjustments there are of that next smaller scale. So allocate them, deduct their total correction from the goal, and repeat until you know how many A events there are going to be.

An example: Let’s say that we need a total of 84mm of additional rainfall.

▪ 78 is less than 84, and gives an L adjustment. 91 is too high.
▪ So there is 1 L adjustment of 12 x the base. Let’s use 1 mm as the base. So on on day, there will be +12mm of rain.
▪ Subtract 12 from the total required: 84 – 12 = 72.

▪ The next lowest rating, K, has a cumulative weight of 66. There’s only room for one of those in 72.
▪ Subtract the value of K from the new target of 72: 72-11 = 61.

▪ The next lowest rating, J, has a cumulative weight of 55. There’s still only room for 1. So far, we have three events: L, K, J.
▪ Subtract the value of J from the target total: 61 – 10 = 51.

▪ The I rating scores 9 and has a total weight of 45. Still only 1 event.
▪ Subtract the value of I from the target total: 51-9=42.

▪ The H rating scores 8 and has a total weight of 36. Still only 1 event.
▪ Subtract the value of H from the target total: 42-8=34.

▪ The G rating scores 7 and has a total weight of 28. Still only one event. So far we have 1 each of L, K, J, H, and G.
▪ Subtract the value of G from the target total: 34-7=27.

▪ In the same way, we add a single F and a single E event, accounting for another 11 of the total adjustment and leaving 16 to get.

▪ And then we add a D event with a weight of 5, leaving 11.

▪ You can see that we’re close to a multiple higher than 1, but C doesn’t quite get us there; it has a total weight of 6. Add one C event.
▪ Subtract the C value from the target: 11-3=8.

▪ The weight of a B event is 3. So there are TWO b events in this adjustment, with a total value of 6.
▪ Subtract the 2xB value from the target: 8-6=2.

▪ There are two A events, worth 1 each.
▪ The total sequence is AA, BB, C, D, E, F, G, H, I, J, K, L.

if the scale of the proposed adjustment seems too high for you – it should be less than the total variability if you know it – the solution is to allocate more of the top level events, retreating steps down the table until you get a daily adjustment you can live with.

SECOND EXAMPLE:
We need a total adjustment after a long cold spell of +120°C. If we’re using 1/2 degrees, that’s a total requirement of 120 / 0.5 = 240 units if adjustment. Our daily capacity is no more than 4°C, or 8 of those adjustments. That’s an H event.

▪ H events have a total weight of 36. 240 / 36 = 6 and a remainder. So there are going to be 6 H events for a total adjustment of 6 x 8 = 48. This leaves 192 to go.

▪ G events have a total weight of 28. Divide the target of 192 by 28 and you get 6 and a remainder. So there are 6 G events, for a net correction of 6×7 = 42 units. This leaves 192-42=150 to go.

▪ F events have a value of 6 and a total weight of 21. Divide the target of 150 by the total weight and you get 7 and a remainder. There are 7 F events of total value 7×6 = 42. 108 to go.

▪ E events have a value of 5 and a total weight of 15 each. Divide the target of 108 by 15 and you get 7 and a remainder. 7 E events are a total correction of 35 units. 73 to go.

▪ D events have a value of 4 and a total weight of 10 each. 73 / 10 = 7 and a remainder. There are 7 D events, worth a total correction of 28. 45 to go.

▪ C events have a value of 3 and a total weight of 6 each. 45 / 6 = 7 and a remainder (what a surprise!) There are 7 C events which are a total correction of 21. That leaves just 24.

▪  events have a value of 2 and a weight each of 3. 24/3=8. There are 8 B events accounting for 16 units of adjustment and leaving 8.

▪ Which obviously means that there are 8 A events.

▪ 8 + 8 + 7×6 = 58 events in total.

Having broken the correction up into individual adjustments, the next step is to determine the distribution of those adjustments.

There are three basic models for doing so: Even distribution is the simplest, randomly even is the next most complicated, and weighted is the most realistic.

5.3.4.2.1.1 Even Distribution

The number of events of a given scale can either be lumped together to form a single longer event, or kept separate to form more events. Group them as you see fit.

EXAMPLE 1 CONTINUED:
▪ A, A, B, B, C, D, E, F, G, H, I, J, K, L = 14 events. That’s a workable number – no grouping.

EXAMPLE 2 CONTINUED:
▪ 8xA, 8xB, 7xC, 7xD, 7xE, 7xF, 7xG, 7xH – that’s a total of 58 events, as noted earlier. That’s a bit unwieldy. I’ll break each of the lower “7x” into a 3x, a 2x, and two 1x events (C through E)., and the higher ones into a 3x and a 4x grouping. So I end up with:

▪ 8xA, 8xB, C, C. CC, CCC, D, D, DD, DDD, E, E, EE, EEE, FFF, FFFF, GGG, GGGG, HHH, HHHH = 8 + 8 + 3+ 3×3 + 3×2 = 16+3+9+6 = 34 events.

For those events that have a compound duration, add 1 to the total duration of the adjustment and then subtract the total length of those events.

EXAMPLE 1 CONTINUED:
▪ There are no compound-length events.

EXAMPLE 2 CONTINUED:
▪ There are 12 compound length events – CC, CCC, DD, DDD, and so on. They total 2+3 + 2+3 + 2+3 + 3+4+ 3+4 + 3+4 = 15 + 21 = 36 days. 12 months = 365 days so 365+12-36 = 377-36 = 341 days.

Divide the result by the total number of correction event groups. Round down to (user’s choice) whole days, half days, or quarter days. That’s the interval between the end of one correction event and the start of the next.

EXAMPLE 1 CONTINUED:
▪ 365 days / 14 = 26 days. So every 26 days, there’s a correction event lasting a day.

EXAMPLE 2 CONTINUED:
▪ 341 / 34 = 10. So there are 10 whole days between correction events.

The final consideration is event sequence. For this, you need to actually list them without compression in 8xA and the like. Distribute them as evenly as possible. Then take a die of the next smaller size than the number of events and roll; count along the list to find the first event. List it in sequence, cross it out, and roll again, counting ‘1’ as the next listed entry. If you get to the end of the list, go back to the start and keep counting. When the number of events gets low enough, choose a smaller dice.

EXAMPLE 1 CONTINUED:
▪ A, A, B, B, C, D, E, F, G, H, I, J, K, L, 14 events in total, so use a d12.
▪ Distributed: A, B, C, D, K, G, H, A, B, E, F, I, J, L.
▪ Roll #1 is a 9. The 9th item in the list – A, B, C, D, K, G, H, A, B – is B. List it and cross it off the Distributed list (WordPress doesn’t make that very easy so I’ll put them in brackets instead).
▪ Roll #2 is an 11. E, F, I, J, L. gets me to 5. so A is 6, then B, C, D, K, G. The second event is G.
▪ Roll #3 is an 8. H, A, (B), E, F, I, J, L is 7, so A is 8. The third event is A.
I started with 14 events, now I’m down to 11, so a d12 is now too cumbersome and gets reduced to a d10.
▪ Roll #4 is another 8. B, C, D, K, (G), H, A, (B), E, F – the fourth event is F.
▪ Roll #5 is a 6. I, J, L gets me to three, so B is 4, C, 5, and D is 6. The fifth event is D.

▪ And so on – ultimately, my sequence list is B, G, A, F, D, K, B, L, J, H, C, E, A, I.

EXAMPLE 2 CONTINUED:
▪ 8xA, 8xB, C, C. CC, CCC, D, D, DD, DDD, E, E, EE, EEE, FFF, FFFF, GGG, GGGG, HHH, HHHH , 34 events in total.
▪ Distributed: A, FFF, B, E, C, A, B, DDD, A, GGG, B, C, A, HHHH, B, EEE, A, B, D, CC, A, FFFF, B, E, DD, A, GGGG, B, CCC, A, HHH, B, D, EEE.
▪ 34 events, so start with a d20: Roll#1 is a 3, so B.
▪ Roll #2 is 19, so FFFF.
▪ Roll #3 is 5, so GGGG.
▪ Roll #4 is 11, so C.
▪ Roll #5 is 19, so DD.
▪ Roll #6 is 8, so EEE.
▪ Roll #7 is 17, so D.
▪ Roll #8 is 15, so A.
▪ Roll #9 is 11, so A again.
▪ Roll #10 is 17, so DDD.
▪ Roll #11 is 7, so B.
▪ Roll #12 is 7 again, so A.
… and so on.

5.3.4.2.1.2 Random Distribution

Random distribution is not that much more difficult. Instead of a fixed number of days between events, double the number and choose the next smaller die size.

If there is a large size dice indicated, halve the number to get a more practical alternative and roll two of them.

EXAMPLE 1 CONTINUED:
▪ 26 days calculated between events. 26×2 = 52. I could rig up a d52 simulator using a deck of playing cards. Or I could use a function that I rarely need at AnyDice and get it to roll a string of d52s for me. But none of those are convenient enough.
▪ So halve it again. Now I’m back to a 26. I could use d20. But I actually have in my unusual dice collection a d24. So 2d24 will work just fine.
▪ In fact, my first 10 rolls are 18, 30, 10 ,36, 15, 15, 39, 20, 27, 36. Continue until you run out of events to schedule.

EXAMPLE 2 CONTINUED:
▪ 10 days calculated between events, double it to 20, and it’s tailor made for a d20.

The goal is to have a random roll that averages the gap between correction events that you calculated previously. It doesn’t hurt for it to be a dumbbell curve, either; in fact that’s probably preferable to a flat roll.

5.3.4.2.1.3 Weighted Random Distribution

In this model, half the events are to take place in the season most appropriate for them to do so, at half the base interval. One quarter of the events are to take place in the season most unusual for them to do so, at 3/4 the base interval. The rest occupy the rest of the total time-frame.

1. Calculate 1/2 the total events, round down. Multiply by 1/2 the base interval (rounded down).
2. Calculate 1/4 of the total events, round down. Multiply by 3/4 of the base interval (rounded up)
3 Subtract these two subtotals from the total correction time.
4. Subtract the two event count subtotals from the total number of events.
5. Calculate the rest-of-the-year average by dividing the remaining correct time by the remaining number of events. Round down.
6. Split these non-adjacent season events in two as evenly as possible.

EXAMPLE 1 CONTINUED:
14 events. Base interval 26 days. Total time allowed = 365 days.
1. 1/2×14 = 7. No rounding needed. So 7 events 13 days apart = 91 days.
2. 1/4×14 = 3.5, round down to 3. Calculate 3/4 x 26 days = 19.5 days, round up to to 20. So 3 events 20 days apart = 60 days.
3. 365 – 81 – 60 = 224 days.
4. 14 events – 7 – 3 = 4 events.
5. 224 days / 4 events = 56 days. So 4 events 56 days apart.
6. Over two non-adjacent seasons = 2 events per non-adjacent season.

EXAMPLE 2 CONTINUED:
34 events. Base Interval 10 days. Total time allowed = 341 days.
1. 1/2×34 = 17, no rounding needed. So 17 events 5 days apart = 85 days.
2. 1/4×34 = 8.5, round down to 8. 3/4 x 10 = 7.5, round up to 8. So 8 events 8 days apart = 64 days.
3. 341 – 85 – 64 = 192 days.
4. 34 – 17 – 8 = 9 events.
5. 192 days / 9 events = 21.33 days, rounds down to 21. So 9 events 21 days apart.
6. Over 2 non-adjacent seasons, one will have 4 events and one will have 5. Choose randomly.

5.3.4.2.2 Distributed corrections: 6 months

Having established the basic methods, I don’t need to repeat them. Instead of full seasons, though, we’re now talking about 1/2 seasons.

5.3.4.2.3 Distributed corrections: 3 months

It’s a similar story here, but three months is a season. So weighted adjustments are no longer necessary or possible, unless you want to break the 12-week season into sub-seasons of 3 weeks each.

5.3.4.2.4 Slow Corrections (2 months)

Two months is less than a season. Note that the shorter the time interval in which to make the corrections, the more frequent the corrections have to be and the more extreme they might need to be.

5.3.4.2.5 Normal corrections: 1 month

Sydney recently experienced a series of rain events lasting about 3 1/2 weeks. While there were one or two days without rain, and perhaps half-a-dozen in which part of a day had good weather, most of the time, it was dark and gray all day and raining for at least part of the day. This produced short-term flooding in some vulnerable areas and more serious flooding in one region.

Four weeks is unusually long for an intense heat-wave, but there have definitely been times when I have experienced a long “warm wave”. It’s also long for a “Cold Snap” but an appropriate duration for a period of “chilly weather”.

5.3.4.2.6 Fast corrections: 1/2 month (2 weeks)

Regionally catastrophic and generally uncomfortable one way or another. I have experienced deluges, heat-waves and chilly periods of this length before. By the time you get to this rapidity of correction, it’s not so much about intervals between days, it’s about how much on any given day.

If you have to, assign each day an adjustment and then use the interval / random systems (your choice) to either top up or mitigate the adjustment accordingly.

5.3.4.2.7 Catastrophic corrections 1/4 month (1 week)

This doesn’t happen very often. Maybe one year in ten or eleven, something like this will happen. Remember that “errors”” needing correction have built up over a 6-30 month time frame – so having all of that readjustment to average take place in just 1/4 of a month can’t be considered anything less than catastrophic. Weather events of 24-120 times the usual intensity. Monsoonal downpours (171mm is the record for my neck of the woods) – in One Day. That’s 6.7 inches. But, in actual fact, that happened in one HOUR.

I’ve experienced a week of 50°C+ (122°F) peak temperatures – measured, as always, in the shade. From memory, the unofficial “in the sun” number was more like 67°C (152°F). It didn’t cool off much at night, either. That year, it reached the point where we couldn’t concentrate enough to game – all you could do was sit and pant and melt (metaphorically). Roads were melting. Metal in street signs became fatigued. People died.

I’ve experienced a week or so of darn-near subzero temperatures, again unusual for this part of the world. Five layers and a heater and still cold (but part of that was too many layers, causing sweating, which made me even colder. I know better, now).

5.3.4.3 Maintaining Synchronization

Something else to watch for, with all these levels and layers of adjustment is when seasons actually turn. For many years now, it has seemed like the seasons have changed in Sydney a week or two sooner than expected.

Certain landmark dates spring to mind. The second weekend in October is one. That’s the weekend of the Bathurst 1000 motor race, probably the premium such event in the southern hemisphere and certainly the biggest in Australia. The weather has generally just started to turn warm, though the nights are still a bit chilly.

One such weekend, about twenty years ago, it was so hot that the local (to where I was living then) Mr Whippy van ran out of ice-cream and had to phone his wife to get more out of the storeroom and drive it to him. it was 40-something degrees (C) in the shade, and I was acutely uncomfortable because at the time I didn’t own a fan.

For the last decade or so, that race has taken place against a backdrop of what felt like the lingering depths of winter. In a phrase, it’s been “bloody cold”. So the onset of winter seems early, the onset of summer seems late, and the summer itself seems a lot milder overall, with fewer really hot days.

These are all within the scope of random variation. But they are also patterns.

People perceive patterns very readily – even patterns that are just coincidence or otherwise aren’t really there. The problem is that randomly-generated weather sometimes doesn’t exhibit those patterns, or doesn’t do so consistently.

You could ague that the long-cycle variations and extended corrections are the system’s attempts to create those patterns, ready for the players and PCs to recognize, and you would not be altogether incorrect. The problem is that those long-term cycles have been through multiple cycle periods and these patterns have persisted even in the face of them changing. So this is something even deeper and more long-term in its stability.

It’s not something that these mechanics, as presented so far, take into account. Instead, the system lays the onus of creating and accommodating these patterns on the GM. For some people, that’s not good enough. So for them, I recommend yet another layer of adjustments (and these do need to feed into the system for later average corrections). Pick two or three milestone weather events and schedule them through your campaign calendar year.

Suggested examples include:
★ The first long soaking rain of spring;
★ The first big thunderstorm of summer;
★ The first really hot day of summer;
★ The first night of frost on the ground;
★ The first snowfall of the year.
★ The commencement of the harvest.

Make whatever adjustment is necessary on that date each year to achieve that milestone. And half of it for the 2 days that follow, and 1/3 of it for the three days after that, and 1/4 of it for the four days after that.

And leave those adjustments there, for year after year. This effectively overrides the system’s randomness to synchronize the seasons with the calendar.

5.3.4.4 Multiple Correction Layers

I can’t recommend letting a spreadsheet do all the math for you, highly enough. You will have to input base values and the various layers of adjustment, but let it put everything together.

But, unless the PCs are mostly staying in one place for a really long period of time, they will never notice. Therefore, I would only worry about the multiple correction layers and detailed adjustments for the current Base Of Operations and maybe the nearest big city (if the two are not the same).

Maybe for the capital of the Kingdom.

Whenever the PCs venture away from this fixed point in the geography of the campaign, use the simplest possible system – the chance of good weather / chance of bad weather.

If it looks like it might be important, use the more detailed system to generate the weather experienced historically wherever the PCs are now. But the rest of the time, forget it. They will.

5.4 Losses & Hazards

In keeping with that principle, have a PC-owned or operated business experience losses and climatic hazards as seems appropriate. There will be some such almost every year. Use the detailed system as a guide to what could happen and invent the numbers out of whole cloth as you need them. Use the detailed system to justify whatever you come up with.

“As you know, when Skyrym is full in August, flooding of the plains is never far away.”

Is it reality? Is it superstition? is there a causal connection? You don’t care – the weather is what it is, never mind what caused it to be that way. So invent numbers that can be justified and that seem reasonable and that enhance the plot and make the background seem more real, and forget the simulationist reality as much as you can.

5.5 Expenses – as Terrain Factors

Okay, this is now edging it’s way back to where the subject of discussion left off last time. It’s also a part of the subject that has been touched on previously, but it’s time to integrate it into the overall conversation about land travel.

If you have a choice of two routes, both of the same length and difficulty, the better one to choose is the one that has the lower expenses, yes? Obvious.

It’s when things stop being equal that they get more complicated. Route X is shorter but steeper than Route Y, and that increases the overheads from using that route – but Route Y has a fee or charge that has to be paid. Now which one is better?

One way to decide is to arbitrarily set a terrain factor to represent the overheads and expenses that will be incurred.

Every route will have some cost attached to it. Some of those costs will happen every trip, some may be amortized over many trips, averaged out.

There’s also a relative factor to take into account – when you don’t have much money, every expense looms three times as large. When you have the luxury of being able to afford a faster but more expensive route, you can often get longer to sell your wares or a better marketplace in which to do so – higher demand and people willing to pay more. Handle such matters correctly, and make additional profits; handle them poorly and remain relegated to struggle-street.

There is a lot to be said for the notion of making the tale of the business’ success a narrative thread in it’s own right. Two steps forward, one step back, two steps forward, a challenge that has to be overcome… weave that into your background and you can get most of the benefits of a hands-on business operation run by the players through their PCs without actually bogging down with the minutia. The principle of ‘expenses as a terrain factor’ offers a method of creating the narrative that surrounds these events, justifying and explaining them. “Until now, we couldn’t afford X – but now we can.”

5.6 Expenses – as aspects of Politics

By far the bigger expenses, though, are not the relatively fixed and stable ones, they are the soft costs that derive from politics and the political winds that are blowing. Bribes and Taxes and Commissions and Levies and Surcharges – the whole nine yards.

Most rulers are rational; they won’t strangle the operations that are (or could) make their domains richer, because they will get to skim some of that cream off the top.

Rather than looking at this sort of expense as an aspect of a business operation, I recommend regarding it as a plot development that attaches significance and relevance to the campaign by means of the business operation. Use the profitability of the business as a delivery system to make the politics around them matter to the PCs.

Treat the overland business operation as a plot development delivery system.

Overland transport is uniquely positioned for this purpose, when you thin about it.

River travel leaves you no choice but to go where the river takes you. What are choices and options and variables for overland travel are relatively fixed and simplified on a river.

Maritime transport is mostly over the free-for-all seas and oceans. It’s only really when you put into port, or come across another vessel, that there’s any real chance for social or political engagement.

Overland travel means choices. And choices are both levers to be pulled and storytelling opportunities. Go out of your way to create them, manipulate them, and use them.

The more you bottleneck the choices open to the PCs, the less opportunity there is for engagement.

But, on the other hand, to a certain extent, you don’t want the players to be so actively engaged in the business operations. So there is a fine line to be trodden here. Always keep it in mind – but don’t waste the opportunities that land travel offers you; save them for when you need them.

Next week, while I gird my loins for the big posts to come in this chapter, I’ll take a time out with one of a pair of smallish posts that came to mind today. I think.