D. Dutta, S. Narasimhan, J.R. Sharma
Regional Remote Sensing Service Centre, CAZRI Campus, Jodhpur – 342 003
E-mail: [email protected]
NNRMS, ISRO Headquarters, Bangalore
Dry land farming accounts for more than 60% of the cultivated lands which suffers from the vagaries of climate, resulting in periodic crop failure and acute shortage of water. Government of India has accorded highest priority to these areas for holistic development in the form of National Watershed Development Project for Rain fed Areas (NWDPRA). The project aims at in situ moisture conservation, primarily through vegetative measures to conserve as much rain water as possible, controlling soil erosion and regeneration of green biomass both on arable and non-arable lands. As the development process is an integrated approach and synchronized activities of various departments, there is need to monitor the development activities and assess their performance in terms of performance indicators. Conventional ground data collection in pre and post treatment period for monitoring is indeed a costly and time-consuming affair and subjected to human bias. However, synoptic viewing from space through multi-spectral capability is a cutting edge technology to provide information about various subtle biophysical parameters. Having suitably selected spectral bands sensitive towards vegetation pigment and cell structure, the image acquired over land surface gives spatial details of vegetation condition along with types and dynamics. Satellite data analysis before and after watershed treatment can provide valuable baseline information towards direction of change and its overall impact on biophysical environment.
Eight micro-watersheds, distributed over different agro climatic zones of Rajasthan, India, were studied for their performance in terms of biomass gain, landuse change and changes along drainage line, if any. The satellite data prior to treatment year (1988) was compared with post-treated year (1996), to reveal the noticeable changes over a span of 8 years. Visible surface manifestations due to watershed treatments were recorded in terms of increase in number and extent of water bodies, forest/horticultural plantation, and agro-forestry. For comparing the performance of the watersheds located at different agro climatic set up a comprehensive rating procedure was adopted for quantification and parametric evaluation of the changes, which considers all the visible impacts discernible in the data. The ratio of performance of watersheds in the year 1996 over base year i.e. 1988 gives the remote sensing based watershed health index (RSWHI). How the watersheds have performed at different agro-ecological set up has been described in details.
The study area is distributed across 8 agro-climatic zones (Figure 1) ranging from desertic western plain to humid southern and south-eastern plain. There is large variation in rainfall from 270 mm in the west to as high as 922 mm in the south. The annual and diurnal temperature variations are also very high. Mean monthly temperature goes below 7°C during winter and rises above 42°C during summer. The watersheds along with area, geographic locations and climate pattern are given in Table 1.
Fig. 1: Map of the study area
Table 1: Location, area, agro-climatic zones and associated climatic variables of the watersheds
|Agro climatic zone||Rainfall
|45I/8||6320||IIA: Transitional plain of inland drainage.||383||40.7||6.3|
|w2|| Bawliya pada
|46I/12||3820||IVB: Humid southern plain.||922||41.5||7.8|
|w3|| Birathai kalan
|45J/4||9019||IIB: Transitional plain of Luni basin.||472||40.2||9.5|
|45G/14||5862||IVA: Sub-humid southern plain and Aravalli hills.||650||38.6||7.8|
|54A/13,14||5581||VIIB: Flood prone eastern plain.||577||40.6||8.3|
|54C/5||12208||V: Humid south-eastern plain.||684||42.2||7.1|
|45C/1,2||8408||1A: Arid western plain.||270||39.5||7.7|
|w8||Syala(Tonk)||45O/9,13||6538||IIIA: Semi-arid eastern plain.||613||39.5||7.3|
Materials and methodology
Indian Remote Sensing Satellite, LISS II data of 1988 and 1996 (post monsoon period), with a spatial resolution of 36.25 m was used for the study. Data selection was done based on crop window, crop calendars and vegetation dynamics of the region. For baseline mapping Survey of India (SOI) topomaps in 1:50,000 scale and watershed maps with demarcated treatment areas were procured. Besides data on watershed activities, base line data for various development activities etc were also collected. For supervised landuse classification, ground truth from homogeneous areas were collected and used as training sites in digital landuse classification. Besides data collection, discussion was also made with the farmers and soil conservation officials.
Satellite data of both the years were normalized for minimizing the changes in spectral responses due to seasonal variation in atmospheric effects, sensor characteristics, sun illumination etc. that is necessary in change detection study, to capture subtle changes. Scene statistics is used to normalize band by band for bringing the mean and standard deviation of two dates at par with each other. Data processing was performed using both Digital Image Processing Software (EASI/PACE) and Geographic Information System (ARC/INFO).
Grid base generation, toposheet registration, mosaicing of scenes and image registration
Watershed wise geographic grid base of 5 minutes interval was generated and output projection was defined in Lambert Conformal Conic C, which is suitable for small area analysis. The purpose is to register the scanned maps into real world coordinate system. For registration of maps with georeferenced grid vectors, GCPworks function of PCI software was used, following nearest neighborhood resampling and affine transformation model. Once a registered database with topomaps was created the uncorrected raw satellite images were also registered in reference to the registered topomaps with minimum permissible rms errors (less than half of a pixel). Wherever single image does not cover the entire watershed, mosaicing was done with the adjoining scenes and then registration was performed. From the toposheets the watershed boundaries were digitized and saved as a vector segment, which was subsequently rasterized and used for watershed mask preparation.
Landuse classification was done following maximum likelihood classifier (which classifies each pixel based on their probability of being in a class). Training sites with known landuse classes were interactively applied for training the classifier, which subsequently classifies the whole image along with the training sets for a classified landuse output. Homogeneous training areas excluding bordering pixels, uniformly spread throughout the scene are collected for all the landuse types, which are spectrally separable. The spectral separability was also checked statistically by computing divergence matrix/confusion matrix. The statistics of training sets thus obtained are used for classifying the image. Using multivariate sample mean vector and inter band variance covariance matrix the probability of each pixel is calculated for each class and the pixel is assigned to that class which has the highest probability. The classified output was then verified in the field for 95% classification and overall accuracy. Broadly the lands are classified into agricultural lands, forestlands, wastelands and water body and their sub-classes.
The spectral response curve of healthy green vegetation is characterized by strong absorption in the red region together with a higher reflection in the NIR region of electromagnetic spectrum. Vegetation vigour is assessed through vegetation indices, which is a real number that is generated by linear combination of spectral bands of satellite predominantly the red and infrared band. Cell structure and plant architecture serve as effective scatters in the optical portion of electromagnetic spectrum due to large contrast in the refractive index between turgid cells of leaf tissues and the intercellular air space. One of the most common used indices is Normalized Difference Vegetation Index (NDVI), which is highly correlated with vegetation parameters such as green leaf biomass and leaf area. NDVI is given as the ratio of reflectance of (IR-R) and (IR+R). The index value varies between -1 and +1. In the present study the vegetation index values were grouped into 5 classes of same class interval based upon the maximum and minimum values. The groups were named as very poor, poor, moderate, good and very good respectively.
For detection of positive and/or negative changes in landuse and vegetation vigour, weights were assigned to various classes of landuse and NDVI independently. Then each pixel of 1996 scene is compared with the corresponding pixel of 1988 and the difference is noted. Positive difference in landuse gives an indication of better management practices. Both the overall changes and the degree of changes are recorded for estimating the degree of criticality. A programme was written in EASI/PACE digital image processing software, for change detection, which makes use of landuse/NDVI values of 1988 and 1996 using a logic table (look up table). The difference image was finally classified into no change, positive change and negative change categories.
Composite scoring for remote sensing based watershed health index
To evaluate relative performance of various watersheds in terms of landuse and biomass change the watershed quality indicators (derived through landuse and NDVI) were scaled and weights were assigned depending upon the impact of the health parameter on watershed. For various landuse types and vegetation vigour the weights were assigned as follows (Table 2 and 3).
Table 2: Weights assigned to various landuse classes
Where, W B = water body, AF = agricultural plantation, FP = forest plantation, GF = good forest, MF = moderate forest, CL = crop land, F = fallow, LWIS = land with scrub, SF = scrub forest, PF = poor forest, LWOS = land without scrub, SA = sandy area, B/R = barren/rocky, R/G = ravenous/gullied.
Table 3: Weights assigned to various vegetation vigour classes
|Very poor||Poor||Moderate||Good||Very good|
The area weighted spatial averaging was done for each watershed for both the years as follows.
Wav = [