A Wireless Sensor Network (WSN) is a network of small, low-cost, battery-powered sensor nodes, with limited computational and communication capabilities. The sensor nodes collect certain data from an environment and send it to a Base Station (BS). It is important that the data collected by the sensor nodes is delivered reliably to the BS and with as low computational and communication loads (thus with low energy consumption) imposed on the sensor nodes, as possible. However, with unreliable communication channels and battery-powered and resource-constrained sensor nodes, achieving both reliability and energy-efficiency in a data collection process in a WSN is a challenging issue. Transmission redundancy, a measure typically used to achieve reliability, also results in an increase in communication loads imposed on each sensor node. This thesis attempts to address this challenging issue by applying two novel measures, which are: (i) adjust the level of transmission redundancy in response to network conditions, and (ii) use computation to reduce communication load on sensor nodes. The first measure can reduce the number of messages in the network while still achieving reliability, and the second measure can reduce both the number and the length of messages transmitted in the network, thus reducing energy consumption. To this end, the thesis has made the following contributions. Firstly, it proposes a novel Multi-Aggregator based Adjusted Multicast (MAAM) method to optimise the trade-off in achieving reliability and reducing energy consumption in delivering collected data from the sensor nodes to the BS. The optimisation is done in two ways, using data aggregation and adjusting the level of transmission redundancy in response to network conditions. In this research, the level of transmission redundancy is adjusted manually and as such an algorithm for automatic adjustment of the transmission redundancy level is currently not a part of MAAM. The data delivery ratios and the energy costs of the method have been investigated and compared against three existing data aggregation approaches (namely Single-Aggregator based Uni-Cast (SAUC), Multi-Aggregator based Uni-Cast (MAUC) and Multi-Aggregator based Broad-Cast (MABC)) under various network conditions. The results show that: (i) MAUC gives the best reliability and lowest energy costs when there are no failed nodes in the network, (ii) MABC has the highest energy costs under all network conditions but only provides reliability when the network is not congested, (iii) MAAM gives higher reliability and lower energy costs than MAUC in a network with failed nodes, and (iv) MAAM gives higher reliability than MABC in a congested network and lower energy costs than MABC under all network conditions. Secondly, it proposes a novel idea called Structured-data-representation-and-next-hop-node-Interpretation-and-Aggregation (SIA) to maximise the amount of data which can be collected for a given message payload length. To do so, SIA implements a novel method, called Revised Partial Mapping (RPM). Different from existing topology coding methods which mainly rely on data aggregation to reduce message size, RPM uses structured data representation along with data aggregation such that data from more sensor nodes can be collected for a given message payload length. A theoretical evaluation of RPM has been carried out to determine the network size (i.e. the number of sensor nodes and number of hops in the network) supported with a given message payload length. The results show that a greater network size can be supported with RPM, as compared to the other methods. To further evaluate RPM, a novel Aggregation based Topology Learning (ATL) protocol has been designed. ATL uses RPM to collect topology data from a WSN and, based on the collected data, the BS can learn the topology of the WSN and identify high energy-consuming nodes in the network. The evaluation is done on a real WSN testbed. The results show that: (i) the learnt topology constitutes, on an average, 89% of the total nodes and identifies, on an average, 60% of these nodes in their correct hops, (ii) average energy consumption is almost the same as when data aggregation is carried out without the learning process, and (iii) on an average, 76.72% of the high energy-consuming nodes are correctly identified. Thirdly, it proposes a novel Localization and Mobility Modelling (LMM) system for wildlife tracking to demonstrate a potential application of the ATL protocol highlighted above. LMM makes use of three main building blocks, ATL, a novel Multi-Zone Multi-Hierarchy (MZMH) communication structure and a novel Location Estimation (LE) method for collecting data and estimating movement traces from the collected data. ATL is optionally integrated with MAAM for data collection to improve the accuracy of estimated movement traces. LMM has been evaluated on a WSN simulator using a dataset of zebras' movement traces collected from real-world deployments and the results have been compared against a wildlife tracking system, called ZebraNet, and a movement trace estimation method, called DV-hop. The results show that: (i) a node used in LMM weighs 93.31% less, uses 92.5% smaller storage capacity and has 98.5% lower energy costs than a node used in ZebraNet, and (ii) with the LE method, the error in the movement trace estimation is 61.56% lower, on an average, as compared to DV-hop.