Something like BIRCH could do the trick .. http://en.wikipedia.org/wiki/BIRCH

Dear Atheer,

You will find a proposition of K estimation in my paper here, i hope that it could be interesting for your work

https://www.researchgate.net/publication/224683111_Dimensionality_Reduction_Techniques_An_Operational_Comparison_On_Multispectral_Satellite_Images_Using_Unsupervised_Clustering

Best Regards

Ludovic

Conference Paper Dimensionality Reduction Techniques: An Operational Comparis...

What might work you are the 'online' clustering algorithms for streaming data. You can actually adapt them for your case. These papers will give you a plenty of options to implement what you may want:

1) Stream Clustering - Massive Online Analysis

http://moa.cms.waikato.ac.nz/details/stream-clustering/

2) Streaming k-means

http://papers.nips.cc/paper/3812-streaming-k-means-approximation.pdf

3) Density-based based clustering for real-time data

http://dl.acm.org/citation.cfm?id=1281210

You can use  any one of the dynamic clustering algorithms.

As a software engineer I'd do this in in two steps.  First of all, use a small data-sample to research the clustering algorithm and distance-measure that is most suitable for your problem.  You could do this using something like R (http://en.wikipedia.org/wiki/R_%28programming_language%29).  Once you're OK with the method, you can scale up to millions of points using, say, Apache Mahout... https://mahout.apache.org/.  It supports most of the main-stream clustering algorithms. Mahout also has other things like SVD that could be useful to you.

try to use fuzzy c means clustering algorithm using the sensed values only by giving desired no.of clusters.

1- there is so much algorithms for clustering or classification, just search the word, but the differences between them is time to solve or accuracy, for more information just see these algorithms:

a- Fuzzy Clustering (c-means) (this one it's great for you as a clustering junior researcher)

b- ANN

c- ACO

d- GA

e- SVM (this one it's great for you as a clustering junior researcher too)

... e.g.

2- see what is important for you? as you know we may have many dimensions of features but we are always trying to minimize that to small dimensions. I suggest to cluster your data set as them temperatures.

good luck

I am developing the method of descriptor fingerprints. They can be used for similarity clustering based on the Tanimoto index using the method of Butina. It is a fast method, especially if you use parallel programming. 

Incremental CFS Clustering algorithms are well suited for your problem.