From Classical Models to AI: Forecasting Humidity for Energy and Water Efficiency in Data Centers

1. of Humidity Forecasting for Reliable Data Centers

As the facility requirements of AI skyrocket, the infrastructure that makes all of it possible is pushing against limited resources. By 2028, recent research shows that AI could eat electricity that is the same as 22% of all US households [1].  Racks of high-performance AI chips eat a minimum of 10 times as much power as conventional servers in data centers. Accordingly, an infinite amount of warmth is produced, and cooling systems take up many of the constructing space [2]. Along with its carbon footprint, AI also has a considerable water footprint, much of it in regions of already high-water stress. For instance, GPT-3 requires 5.4 million liters of water to coach in Microsoft’s US data centers [3]. Seasonal forecasting is critical to the every day operation of apparatus inside data centers. Weather conditions, akin to temperature and humidity, affect how intensely cooling systems inside data centers must work [4].

In this text, the forecast of humidity is computed in several ways. A greater forecast of temperature and humidity can enable more efficient load planning, optimization of cooling schedules, and fewer demand placed on power and native water sources. Now, since we’re primarily discussing humidity in this text, allow us to see what the consequences of its extreme values are:

• High humidity: Condensation becomes an enormous issue — it might corrode hardware and trigger electrical failures. It also makes chillers work harder, costing more energy and water.
•  Low humidity: The danger flips: static and ESD (electrostatic discharge) can construct up and fry sensitive chips.

Accurate forecasting of humidity might help:

• Wonderful-tune cooling schedules
• Determine demand peaks
• Schedule maintenance
•  Redistribute workloads before environmental conditions cause costly downtime

By implementing the above protective measures, we reduce the strain on electricity and native water supplies, ensuring the resilience of AI centers and the general efficiency of the distributed computing infrastructure.

It isn’t only data centers that might be affected by humidity; edge devices, akin to sensors, might be affected as well. These are more vulnerable to weather conditions because they’re typically outdoors and in distant areas. Edge applications often need low-latency predictions. This favors lighter algorithms, akin to Because of this, within the forecasting section below, and other light algorithms are discussed.

Allow us to conclude this section by discussing the futuristic cover image of a knowledge center situated on the Moon. Lunar data centers could be impervious to lots of Earth’s constraints, akin to extreme weather and earthquakes. As well as, the Moon offers a superbly neutral place for data ownership. As a matter of fact, on 26^th February 2025, launched a rocket that carried lunar lander [5]. Amongst other things, contained a small data center, called , developed by . couldn’t manage a full upright landing, nevertheless, performed successful data operations prior to landing. As well as, despite the undeniable fact that the lander landed inside a crater, the data center survived and demonstrated the potential for a lunar data center [6].

2. A Real-World Case Study: Forecasting Humidity With a Precision Interval

Given the importance of weather forecasting for data centers, I turned to a real-world dataset from containing every day climate measurements from Delhi. India has a strong data center industry. Based on [7], Delhi currently has 30 data centers, and a Delhi developer will invest $2 billion to further expand the India data center growth [8].

The information contain temperature, humidity, wind speed, and atmospheric pressure measurements. A training set is provided on which we trained our models, and a test set, on which we tested the models. The link to the info and data about its license might be present in the footnote of this text.

Although temperature, wind, and pressure all influence cooling demand, I focused on humidity since it plays a vital role in evaporative cooling and water consumption. Humidity also changes more rapidly than temperature, and due to this fact, it’s a really meaningful goal for predictive modeling.

 I started with classical approaches akin to , then moved to more flexible models like and and concluded with deep learning models. Here is an entire list of forecasting methods in this text:

Along the way in which, I compared accuracy, interpretability, and deployment feasibility — not as an educational exercise, but to reply a practical query: which forecasting tools can deliver the type of reliable, actionable climate predictions that help data centers optimize cooling, lower energy costs, and conserve water?

As well as, every forecast plot will include a prediction interval, not only a single forecast line. A lone line might be misleading, because it implies, we “know” the precise humidity level on a future day. Because the weather is rarely certain, operators need greater than a single forecast. A prediction interval gives a variety of likely humidity values, reflecting each model limits and natural variability.

Confidence intervals tell us in regards to the mean forecast. Prediction intervals are broader — they cover where real humidity readings might fall. For operators, that difference is vital: underestimate the range and also you risk overheating; overestimate it and also you spend greater than you would like.

approach to judge prediction intervals is by . With a 95% confidence interval, we expect about 95 out of 100 points to fall inside it. If only 86 do, the model is simply too sure of itself. Conformal prediction adjusts the range so the lines up with what was promised.

Conformal prediction takes the model’s past errors (residuals = actual − predicted), finds a typical error size (quantile of those residuals), and adds it around each recent forecast to create an interval that covers the true value with the specified probability.

Here is the principal algorithm for the computation of the prediction interval:

1. Create a calibration set.
2. Compute the residuals:

where the primary term on the appropriate side of the equation is the actual observed value, and the second term is the model prediction for a similar point.

3. Find the quantile of residuals:

where alpha is the importance level, e.g. 0.05.

4. Form the conformal interval for a brand new forecast:

The interval at time t is the same as:

3. Data and Forecasting Methods (with Code)

The code for all forecasting methods discussed in this text is on . The directory link is at the top of the article. Before we discuss our forecasting methods, allow us to take a have a look at our data. Figure 1 shows the training data, and Figure 2 shows the test data. As seen in Figure 1, the training data behave in a stable, stationary manner. Yet Figure 2 tells a special story: the test period breaks that stability with a transparent downward drift. This stark contrast raises the stakes.

We expect that structure-based methods, akin to ARIMA, and traditional ML methods, akin to Random Forest, can have a tough time capturing the downward shift because they are usually not temporally aware. However, deep learning forecasting methods can understand that the test series mirrors similar seasonal segments throughout the training data, and due to this fact are more equipped to capture the downward shift.

3. A. Forecasting

(AutoRegressive Integrated Moving Average) models mix three elements:

• terms that capture the memory of past values
• terms that account for past forecasting errors
• Differencing (the “I”) to remove trends and make the series stationary.

3. A. 1. AutoARIMA Test Data Forecast

Traditionally, the analyst must test for stationarity and choose how much differencing to use before fitting the model. This can be a difficult process that may also be liable to error. removes that burden by running statistical tests under the hood. It mechanically decides the degree of differencing and searches across and mixtures to pick the very best fit based on information criteria. Briefly, you’ll be able to hand it raw, non-stationary data, and it should handle the detective be just right for you—making it each powerful and easy.

Figure 3 shows the forecast (orange dashed line) and the prediction interval (yellow shaded area).  can follow short-term fluctuations but is unable to capture the longer downward trend; due to this fact, the forecast becomes a gentle line. This can be a typical limitation: can capture local autocorrelation, but it surely cannot capture evolving dynamics. The widening prediction intervals make sense—they reflect growing uncertainty over time.

3. A. 2. Accuracy of and Coverage of Prediction Interval

Table 1. Errors of

In Table 1, we report three different errors: , and to supply an entire picture of model accuracy. and are the best to read, since they use the identical units because the goal. puts more weight on big misses, while tells you the typical size of an error. We also report , which is less intuitive but commonly used for comparison.

Regarding the prediction interval, we didn’t apply conformal prediction, since already returns model-based 95% prediction intervals. These intervals are derived from s statistical assumptions reasonably than from the model-agnostic conformal prediction framework. Nonetheless, not using conformal prediction yielded an imperfect of the prediction interval (85.96%).

3. A. 3. Interpretability of

Certainly one of the appealing points of is how easy it’s to “see” what the model is doing. Figure 4 depicts the partial autocorrelation function (), which computes the partial correlation of a stationary time series with lagged values of itself. This Figure shows that today’s humidity still “remembers” yesterday and the times before, with correlations fading over time. This lingering memory is precisely what  uses to construct its forecasts.

Moreover, we ran the test, which confirmed that the train data is indeed stationary.

3. A. 4. Mode of Deployment

is straightforward to deploy: once given a time series, it mechanically selects orders and matches without manual tuning. Its light computational footprint makes it practical for batch forecasting and even for deployment on edge devices with limited resources. Nonetheless, its simplicity means it’s best fitted to stable environments reasonably than settings with abrupt structural changes.  

3. B. Forecasting

On this section,we’ll discussan open forecasting library originally developed by (now ). treats a time series because the sum of three key pieces: a trend, seasonality, and holidays or special events:

• Trend: The trend is modeled flexibly with either a straight line that may bend at change-points or a saturating growth curve, which rises quickly after which flattens out. That is just like the cooling demand in a knowledge center that grows with workloads but eventually levels off once the system reaches capability.
• Seasonality is captured with smooth terms, so recurring patterns akin to weekly or yearly cycles are learned mechanically.
• Holidays or events might be added as regressors to elucidate one-off spikes.

Subsequently, we see that has a really convenient additive structure. This makes easy to know and robust to messy real-world data.

Code Snippet 1 below shows how you can train and fit the model and use it to forecast the test data. Note that the forecast returns and , that are the bounds of the prediction interval, and sets the prediction interval to 95% (line 1 of code). So, like  above, the prediction interval isn’t derived from conformal prediction.

#Train and Fit the Prophet Model
model = Prophet(interval_width=0.95)
model.fit(train_df)
#Forecast on Test Data
future = test_df[['ds']].copy()
forecast = model.predict(future)
cols = ['ds', 'yhat', 'yhat_lower', 'yhat_upper']
forecast_sub = forecast[cols]
y_true = test_df['y'].to_numpy()
yhat       = forecast['yhat'].to_numpy()
yhat_lower = forecast['yhat_lower'].to_numpy()
yhat_upper = forecast['yhat_upper'].to_numpy()

Code Snippet 1. Training and Forecasting with

3. B. 1. Test Data Forecast

Figure 5 shows s forecasting of the test data (the orange line) and the prediction interval (blue shaded area). In contrast to , we will see that s forecast captures well the downward trend of the info.  

3. B. 2. Accuracy and Prediction Interval Coverage

Table 2. Prophet errors.

The forecasting improvement of as compared to might be also seen in Table 2 above, which depicts the errors.

As we said above, the prediction interval was not derived using conformal prediction. Nonetheless, in contrast to , the prediction interval is a lot better: 93.86%.

3. B. 3. Interpretability

As we said above, is transparently additive: it decomposes the forecast into trend, smooth seasonalities, and optional holiday/regressor effects, so component plots show exactly how every bit contributes to and the way much each driver moves the forecast.

Figure 6 above shows the forecast components: a delicate downward trend over time (top), a weekly cycle where weekends are more humid and mid-week is drier (middle), and a yearly cycle with humid winters, a dip in spring, and rising values again in summer and fall (bottom).

3. B. 4. Mode of Deployment

is easy to deploy, runs efficiently on standard CPUs, and might be used at scale or on edge devices, making it well-suited for business applications that need quick, interpretable forecasts.

3. C. Forecasting With

is a neural-network-based extension of . It keeps the identical core structure (trend + seasonality + events) but adds:

• A feed-forward neural network to capture more complex, nonlinear patterns.
• Support for lagged regressors and autoregression (can use past values directly, like AR models).
• The flexibility to learn multiple seasonalities and higher-order interactions more flexibly.

has the good characteristics of being statistical and additive, which enable transparency and quick forecasts. builds on that framework but brings in deep learning. can pick up nonlinear and autoregressive effects, but that extra flexibility makes it harder to interpret.

As Code Snippet 2 below shows, we used seasonality in our model to take advantage of the seasonal mode of humidity.

model = NeuralProphet(
    seasonality_mode='additive',
    yearly_seasonality=False,
    weekly_seasonality=False,
    daily_seasonality=False,
    n_changepoints=10,
    quantiles=[0.025, 0.975]  # For 95% prediction interval
)
# Add custom seasonality (~6 months)
model.add_seasonality(name='six_month', period=180, fourier_order=5)
model.fit(train, freq='D', progress='bar')
future=model.make_future_dataframe(train,periods=len(test), n_historic_predictions=len(train))
forecast = model.predict(future)

Code Snippet 2. Training and forecasting with

3. C. 1. Test Data Forecast

Figure 7 shows forecasting (the dashed green line) and the prediction interval (light green shaded area). Just like forecast captures well the downward trend of the info. 

3. C. 2. Accuracy and Prediction Interval Coverage

Table 3. errors.

It’s interesting to notice that, despite neural augmentation and the addition of seasonality, errors are barely higher than . adds more moving parts, but that doesn’t at all times translate into higher forecasts. On limited or messy data, its extra flexibility can actually work against it, while simpler setup often keeps the predictions steadier and a bit more accurate.

Regarding the precision interval, it’s drawn using the limit variables, and returned by . The of the 95% prediction interval is 83.33%. That is low, but it surely is predicted since it isn’t computed using conformal prediction.

3. C. 3. Interpretability

The three panels in Figure 8 below show, respectively:

• Panel 1. : Shows the learned baseline level and where the slope changes (changepoints) within the piecewise-linear trend.
• Panel 2. : Bars/spikes indicating how much the trend’s slope jumps at each changepoint (positive = faster growth, negative = slowdown/downturn).
• Panel 3. The one-period shape/strength of the seasonal component.

3. C. 4. Mode of Deployment

runs well on CPUs and might be utilized in scheduled jobs or small APIs. While heavier than it’s still practical for many containerized or batch deployments, and may also run on edge devices like a with some setup.

3. D. Forecasting

is a machine learning technique that may also be used for forecasting. That is achieved by turning past values and external aspects into features. That is how it really works: First, it builds multiple decision trees on randomly chosen parts of the info. Then, it averages their results. This helps avoid overfitting and capture nonlinear patterns.

3. D. 1. Forecast

Figure 9 below shows the forecast (orange line) and the prediction interval (the blue shaded area). We will see that doesn’t perform as well. This happens because doesn’t really “understand” time. As a substitute of following the natural sequence of the info, it just looks at lagged values as in the event that they were peculiar features. This makes the model good at capturing some nonlinear patterns but weak at recognizing longer trends or shifts over time. The result’s forecasts that look overly smooth and fewer accurate, which explains the upper

3. D. 2. Accuracy and Precision Interval

Table 4. Random Forest Errors

The poor performance of  can be evident within the high error values shown in Table 4 above.

Regarding the prediction interval, that is the primary forecasting technique where we used conformal prediction to compute the prediction interval.

The coverage of the prediction interval was estimated to be a powerful 100%.

3. D. 3. Interpretability

provides some interpretability by rating the importance of the features utilized in its predictions. In time-series forecasting, this often means examining which lags of the goal variable contribute most to the model’s predictions. The feature importance plot in Figure 10 above shows that the very recent lag (at some point back) dominates, carrying nearly 80% of the predictive weight, while all longer lags contribute almost nothing. This means that the relies heavily on the immediate past value to make forecasts, smoothing over longer-term dependencies. While such interpretability helps us understand what the model is “,” it also highlights why may underperform in capturing broader temporal dynamics in comparison with methods higher suited to sequential structure.

3. D.4. Random Forest Mode of Deployment

models are relatively lightweight to deploy, since they consist of a set of decision trees and require no special hardware or complex runtime. They might be exported and run efficiently on standard servers, embedded systems, and even edge devices with limited “compute”, making them practical for real-time applications where resources are constrained. Nonetheless, their memory footprint can grow when many trees are used, so compact versions or tree pruning might be applied in edge environments.

3. E. Forecasting

is a boosting algorithm that builds trees one after one other, with each recent tree correcting the mistakes of previous trees. In forecasting, we offer it with features akin to lagged values, rolling averages, and external variables, allowing it to learn time patterns and relationships between variables. It really works well since it incorporates strong regularization, which enables it to handle large and complicated datasets more effectively than simpler methods. But, like , it doesn’t naturally handle time order, so its success depends heavily on how well the time-based features are designed.

3. E. 1. XGBoost Test Data Forecast

Figure 11 shows the forecast (orange line) and the prediction interval (blue shaded area). We will see that the forecast closely follows the humidity signal and is due to this fact very successful at predicting humidity. This may also be confirmed in Table 5 below, which depicts relatively small errors, particularly as compared to

builds trees sequentially, and that is the source of its strength. As we previously said, each recent tree corrects the mistakes of the previous ones. This boosting process is combined with strong regularization. This method can pick up quick changes, cope with tricky patterns, and still stay reliable. That sometimes makes its forecasts closer to reality than .

3. E. 2. Forecasting Accuracy and Prediction Interval Coverage

Table 5. forecasting errors.

Here, we also used conformal prediction for the computation of the prediction interval. Because of this, the precision interval is high: 94.74%

3. E. 3. Forecasting Interpretability

despite its complexity, stays fairly interpretable in comparison with deep learning models. It provides feature importance scores that show which lagged values or external variables drive the forecasts. We will have a look at feature importance plots, very like with . For a deeper view, values show how each factor influenced a single prediction. This offers each an overall picture and case-by-case insight.

Figure 12 below shows the burden of a feature, e.g. how often it’s utilized in splits.

The series below shows the gain for every lag, i.e., the typical improvement when a lag is used.

{‘humidity_lag_1’: 3431.917724609375, ‘humidity_lag_2’: 100.19515228271484, ‘humidity_lag_3’: 130.51077270507812, ‘humidity_lag_4’: 118.07515716552734, ‘humidity_lag_5’: 155.8759307861328, ‘humidity_lag_6’: 152.50379943847656, ‘humidity_lag_7’: 139.58169555664062}

The summary plot in Figure 13 shows that  is by far probably the most influential feature, with high recent humidity values pushing forecasts upward and low recent humidity values pulling them downward. Later lags (2–7) play only a minor role, indicating the model relies mainly on probably the most recent statement to make predictions.

3. E. 4. Mode of Deployment

can be straightforward to deploy across platforms, from cloud services to embedded systems. Its principal advantage over is efficiency: models are typically smaller and faster at inference. This makes the model practical for real-time use. Its support across many languages and platforms makes it easy to implement in various settings.

3. F. () Forecasting

The approach combines several specialized models (“experts”), each tuned to capture different points of the info, with a that determines the burden each expert must have in the ultimate forecast. 

In Code Snippet 3, we see the keywords and . Allow us to explain what they’re: We implemented the using models integrated through , with serving as considered one of the experts. is a family of time-series forecasting models built using transformers. is a helpful framework that may handle tabular, text, image, and time series data. is just considered one of its many strategies to spice up performance using model ensembling.

from autogluon.timeseries import TimeSeriesDataFrame, TimeSeriesPredictor
MODEL_REPO = "autogluon/chronos-bolt-small"  
LOCAL_MODEL_DIR = "models/chronos-bolt-small
predictor_roll = TimeSeriesPredictor(
    prediction_length=1,
    goal="humidity",
    freq=FREQ,
    eval_metric="MSE",
    verbosity=1
)
predictor_roll.fit(train_data=train_tsd, hyperparameters=hyperparams, time_limit=None)

Code Snippet 3: Fitting the model

In Code Snippet 3 above, the predictor is named because forecasting generates predictions in a rolling fashion: each forecasted value is fed back into the model to predict the subsequent step. This approach reflects the sequential nature of time series data.  It also allows the gating network to dynamically adjust which experts it relies on at each point within the horizon. Rolling forecasts also expose how errors accumulate over time. This manner, we achieve a more realistic view of multi-step performance.

3. F. 1. Test Data Forecast

As shown in Figure 14 above,  performs extremely well and closely follows the actual test data. As Table 6 below shows,

3. F. 2. Forecasting Accuracy and Prediction Interval Coverage

Table 6. Mixture of Experts Forecasting Errors.

The of the 95% prediction interval is incredibly good (97.37%) because we used conformal prediction.

3. F. 3. Forecasting Interpretability

There are several ways to realize insight into how works:

• Gating network weights: By inspecting the gating network’s outputs, you’ll be able to see which expert(s) got probably the most weight for every prediction. This reveals and certain experts are trusted more.
• Expert specialization: Each expert might be analyzed individually—e.g., one may capture short-term fluctuations while one other handles longer seasonal trends. Taking a look at their forecasts side by side helps explain the ensemble’s behavior.
• Feature attribution (SHAP/feature importance): If the experts are themselves interpretable models (like trees), their feature importances might be computed. Even for neural experts, we will use or integrated gradients to know how features influence decisions.

So while isn’t as “out-of-the-box interpretable” as or you open the black box by analyzing which expert was chosen when, and why.

3. F. 4. Mode of Deployment

Deploying is more demanding than tree ensembles. The rationale is that it involves each the expert models and the . In data centers, on servers, or within the cloud, implementation is simple because modern frameworks like  and  can easily handle orchestration. For edge devices, nevertheless, deployment is far more difficult. The precise challenges are the complexity and size of . Subsequently, pruning, quantization, or limiting the variety of lively experts is commonly obligatory to maintain inference lightweight. frameworks akin to simplify deployment by wrapping your entire pipeline. The site also hosts large-scale models that might help us scale as much as production-grade AI systems.

3. G. Forecasting

[9] is a deep learning model for time series forecasting built from stacks of fully connected layers grouped into blocks. Each block outputs a forecast and a backcast, with the backcast faraway from the input so the subsequent block can concentrate on what stays. By chaining blocks, the model step by step refines its predictions and captures complex patterns. In our implementation, we used a sliding-window setup: the model examines a hard and fast window of past observations (and external drivers, akin to mean temperature) and learns to predict several future points concurrently. The window then shifts forward step-by-step across the info, giving the model many overlapping training examples and helping it generalize to unseen horizons.

In this text, was implemented using , which is an extension of the unique  architecture that features exogenous drivers. and are a part of the library [10], which offers several neural forecasting models. As might be seen in Code Snippet 4, was arrange using a factory function (), which lets us define the forecast variable and add mean temperature () as an additional input. The concept behind including is simple: the model doesn’t just learn from past values of the goal series, but additionally from this key outside factor.

def make_model(horizon):
    return NBEATSx(
        input_size=INPUT_SIZE,
        h=horizon,
        max_steps=MAX_STEPS,
        learning_rate=LR,
        stack_types=['seasonality','trend'],
        n_blocks=[3,3],
        futr_exog_list=['meantemp'],
        random_seed=SEED,
        # early_stop_patience=10,  # optional
    )
# Fit model on train_main
model_cal = make_model(horizon=CAL_SIZE)
nf_cal = NeuralForecast(models=[model_cal], freq='D')

Code Snippet 4: model creation and fitting.

3. G. 1. Test Data Forecast

Figure 15 shows the forecasting model (orange line) and the prediction interval (blue area). We will see that the forecast is capable of follow the downward trend of the info, but stays above the info line for a good portion of the info.

3. G. 2. N-BEATS Accuracy and Prediction Interval Coverage

Table 7. forecasting errors.

For , we used conformal prediction, and, because of this, the prediction interval is superb: 98.25%

3. G. 3. Interpretability

In our experiments, we used the generic type of , which treats the model as a black-box forecaster. Nonetheless, also offers one other architecture with “interpretable blocks” that explicitly model trend and seasonality components. This implies the network not only produces accurate forecasts but may also decompose the time series into human-readable parts, making it easier to know what drives the predictions.

3. G. 4. Mode of Deployment

Because is built entirely from feed-forward layers, it is comparatively lightweight in comparison with other deep learning models. This makes it straightforward to deploy not only on servers but additionally on edge devices, where it might deliver multi-step forecasts in real time without heavy hardware requirements.

Conclusion

In this text, we compared several forecasting approaches—from classical baselines akin to and to machine-learning methods akin to and deep learning architectures akin to and Simpler models offered transparency and straightforward deployment but struggled to capture the complexity of the humidity series. In contrast, modern deep learning and ensemble-based approaches significantly improved accuracy, with the achieving the bottom error (MSE = 45). T

Below we see a summary of the mean square errors:

• MSE = 398.19
• MSE = 105.26
• MSE = 145.31
• MSE = 448.77
• MSE = 57.46
• MSE = 45.52
• MSE = 166.76

Besides accuracy, we also computed a prediction interval for every forecasting method and demonstrated using conformal prediction to compute an accurate prediction interval. The conformal prediction code for every forecasting method might be present in my notebooks on . Prediction intervals are vital because they provide a sensible sense of forecast uncertainty.

For every forecasting method, we also examined its interpretability and mode of deployment. With models like and , interpretation comes straight from their structure. shows how past values and errors influence the current, while splits the series into components like trend and seasonality that might be plotted and examined. Deep learning models akin to or act more like black boxes. Nonetheless, of their case, we will use tools akin to or error evaluation to get insights.

Deployment can be vital: lighter models, akin to  can run efficiently on edge devices. Larger deep learning models can utilize frameworks akin to AutoGluon to streamline their training. An awesome profit is that these models might be deployed locally to avoid API limits.

In conclusion, our results show that reliable humidity forecasts are each possible and useful for day-to-day data center operations. By adopting these methods, data center operators can expect energy demand peaks and optimize cooling schedules. This manner, they’ll reduce each energy consumption and water use. Provided that AI power demands continuously rise, the power to forecast environmental drivers, akin to humidity, is crucial because it might make digital infrastructure more resilient and sustainable.

Thanks for reading!

Your entire code of the article might be found at:

https://github.com/theomitsa/Humidity_forecasting

References

[1] J. O’ Donnell, and C. Crownhart, We Did the Math on AI’s Energy Footprint. Here’s The Story You Haven’t Heard (2025), MIT Technology Review.

[2] Staff writers, Contained in the Relentless Race for AI Capability (2025), Financial Times, https://ig.ft.com/ai-data-centres/

[3] P.  Li, et al, Making AI Less Thirsty: Uncovering and Addressing the Water Footprint of AI Models (2025), Communications of the ACM, https://cacm.acm.org/sustainability-and-computing/making-ai-less-thirsty/

[4] Jackson Mechanical Service Blog, Managing Humidity Levels: A Key Factor For Data Center Efficiency and Uptime (2025), https://www.jmsokc.com/blog/managing-humidity-levels-a-key-factor-for-data-center-efficiency-and-uptime/#:~:text=Inadequate%20management%20of%20humidity%20within,together%20might%20precipitate%20revenue%20declines.

[5] D. Genkina, Is It Lunacy to Put a Data Center on the Moon?  (2025), IEEE Spectrum.

[6] R. Burkett, Lunar Data Center Intact Despite Lunar Lander’s Botched Landing, St. Pete Company Says (2025), https://www.fox13news.com/news/lunar-data-center-intact-despite-lunar-landers-botched-landing-st-pete-company-says

[7] Data Centers in Delhi, https://www.datacenters.com/locations/india/delhi/delhi

[8] Staff writers, Delhi Developer to Invest $2 Billion on India Darta Centre Boom (2025), Economic Times of India Times,  https://economictimes.indiatimes.com/tech/technology/delhi-developer-to-invest-2-billion-on-india-data-centre-boom/articleshow/122156065.cms?from=mdr 

[9] B. N. Oreshkin et al., N-BEATS, Neural Basis Expansion for Interpretable Time Series Forecasting (2019), https://arxiv.org/abs/1905.10437

[10] Library, https://github.com/Nixtla/neuralforecast?tab=readme-ov-file

Footnote:

1. All images/figures are by the creator, unless otherwise noted.
2. Link to data used for forecasting in this text: https://www.kaggle.com/datasets/sumanthvrao/daily-climate-time-series-data/data
3. Data License: The information has a Creative Commons License: CC0 1.0. Link to data license: https://creativecommons.org/publicdomain/zero/1.0/

Excerpt from license deed mentioning business use: