A key limitation of the approach followed for the generation of future LUCC scenarios is that historic trend analysis only considers two discrete points in time. Moreover, Markov chain analysis is unable to assimilate ancillary data, which may constrain the rate and direction of change. Land-cover change is driven by socio-economic and biophysical factors (Lambin et al., 2001). Therefore, to make valid predictions about future land-cover change it is necessary to understand the causes of historic land-cover change. For instance, given the observed dependence of the agricultural sector on irrigation, one may hypothesize that an area of land located near an irrigation canal or well is more likely to be converted to cropland than land without a reliable water source. Furthermore, the expansion of cropland may be limited by the presence of the Himalayan mountain range in the north of the basin, where factors such as elevation and slope may impose physical limits on crop growth. Understanding the objectives of national and international policies on afforestation in the region may provide a means to constrain future scenarios of change. These examples demonstrate that land-cover change cannot be modelled unless the underlying driving factors are taken into account.
Future work could improve the prediction of land cover by applying modelling approaches that utilize biophysical and socio-economic datasets. This is a possible way to reduce uncertainty and provide more accurate projections for the future status of land cover in the UG basin. In addition, the development of historic land-use maps in seasonal or even monthly time-scales (if data become available), as opposed to the annual scales presented here, could provide useful insights on the study of past land-use change trends.
The unavailability of historical flow data is one of the main constraints in this work. This limitation severely impacts the historic and future hydrological model simulations, by restricting the streamflow validation in only one sub-catchment of the study area, and over a period of 10 years. Due to strict flow- data sharing policies posed by India, there is a limited number of water resources research studies and the lack of accessible data has been a great challenge for the present research. Ongoing debates about this issue have not managed to reform policies and resolve it yet, and this is something all future studies will have to deal with.
while the vast majority of parameters was held constant. However, the results showed that there is much room left to improve JULES so that it can function as a proper hydrological model and potential ways of addressing the model’s shortcomings were discussed.
At the daily time-scale, the simple runoff routing mechanism introduced uncertainty to the model outputs. The errors were reduced when looking at the outputs on a monthly basis, which smoothed the impact of the simple routing scheme. However, the lack of river channels, wetlands, or ground- water stores in JULES, along with the fact that the model assumes direct runoff into the river, increased the errors in the model outputs. This is related to the vertical structure of the model, which does not allow horizontal water movements, or any horizontal flux exchange in both surface and subsurface levels. The lack of groundwater component and its interactions with surface fluxes is crucial for the accurate representation of flux partitioning and is further affecting the model’s performance. The free gravity drainage assumption does not allow for any amount of moisture to be drawn from the soil when the water table in the field drops below the maximum soil depth. Coupling JULES to a groundwater model was not explored during the course of this thesis due to time constraints. It would have been a large project considering the lack of prior experience in the catchment, and the difficulties of parameterising the model with limited data.
The importance of in-situ observational datasets has been recognized across the land-surface mod- elling community. Specifically for India, where surface fluxes are poorly constrained, new observations could be brought together with historical data, to develop a more integrated understanding of land sur- face processes. Based on comparisons with new observational datasets, JULES could be evaluated and its parameters further constrained for the monsoon tropics.
The water management (hydropower, irrigation, industrial and domestic usage) and the dam struc- tures play a very significant role in the water balance of this region. The impacts of changing management practices in terms of water usage and storage could be much more important than the impacts posed by climate change or land-use change in terms of future water availability and demand. This has not been taken into account in the modelling work undertaken in the present thesis.
The role of feedbacks between the carbon and the water cycle under climate change conditions has
not been extensively discussed in this research. As mentioned earlier, increasedCO2 concentrations in
uptake and this would improve their water use efficiency by reduced transpiration rates. On the contrary, the increase in air temperature that is being projected for the future would lead to higher potential evap- oration as warmer air can hold more moisture, counteracting the above mentioned effects of increased
CO2 concentrations. The importance of these feedbacks between the carbon and the water cycle is sig-
nificant under risingCO2 and temperature conditions, given that their impacts are likely to affect both
food security and water availability. Further, the simulations of future climate scenarios (Chapter 7) took into account future projections of the meteorological variables used to force JULES, but did not account
for changes in theCO2 concentrations. It would be interesting for future work to look into the relative
effect of changes in precipitation, temperature andCO2concentration on the water cycle.
Neither JULES nor the new coupled scheme (JULES-Info) account for irrigation. There is lack of available spatial data that separate between rain-fed and irrigated areas of different crops whilst giving timely information regarding the amount of irrigation applied throughout the year, over the study area. Besides, the lack of sub-grid heterogeneity means that in this study the model could not simulate irrigation
practices, as irrigation would rarely cover an entire gridbox of 0.1◦. However, it is recognised that the
impact of irrigation is an extremely important aspect of hydrological modelling in a catchment that is 60% occupied by crops. Work is currently under development, both in respect to more accurate land-use classification of irrigated/rain-fed areas and irrigation representation in JULES. A very interesting future direction of research that would shine light onto water management practices and their feedbacks to the climate would be to compare the behaviour of various hydrological fluxes in irrigated versus rain-fed areas.
The most recent version of JULES (v4.1) includes the JULES-Crop model (currently undocumented) and an interesting suggestion for future work would be the comparison of JULES-Crop and JULES-Info. It is worth exploring how potential differences in the way that crop cycles are represented (global version of JULES-Crop vs locally parameterised JULES-Info), are affecting the model results.
In terms of climate projections, as previously mentioned, the strong inter-model uncertainties of GCM-derived data are attributed to inherent limitations of GCMs (Raty et al., 2014). However, these uncertainties were possibly amplified by the delta-change bias correction approach followed here and are posing a limitation on the confidence of the climate change impacts results. This method assumes a constant GCM bias through time, does not retain the change in variability of climatic variables, and does